Operation Wood

EEL 6266
Power System Operation and Control
Chapter 3
Economic Dispatch of Thermal Units
The Economic Dispatch Problem
Consider a system that consists of N thermal-generating
units serving an aggregated electrical load, Pload
input to each unit: cost rate of fuel consumed, Fi
output of each unit: electrical power generated, Pi
total cost rate, FT, is the
sum of the individual
unit costs F1 1 T G P1
essential constraint:
the sum of the output F2 2 T G P2
powers must equal
Pload
the load demand
the problem is to FN N T G PN
minimize FT
© 2002, 2004 Florida State University EEL 6266 Power System Operation and Control 2
The mathematical statement of the problem is a constrained
optimization with the following functions:
objective function:
N
FT = Fi (Pi )
i =1
equality constraint:
N
φ = 0 = Pload − Pi
i =1
note that any transmission losses are neglected and any operating
limits are not explicitly stated when formulating this problem
Problem may be solved using the Lagrange function
N N
L = FT + λ φ = Fi (Pi ) + λ Pload − Pi
i =1 i =1
Principles
the Lagrange function establishes the necessary conditions for
finding an extrema of an objective function with constraints
taking the first derivatives of the Lagrange function with
respect to the independent variables allows us to find the
extreme value when the derivatives are set to zero
there are NF + Nλ derivatives, one for each independent variable
and one for each equality constraint
the derivatives of the Lagrange function with respect to the
Lagrange multiplier λ merely gives back the constraint equation
the NF partial derivatives result in
∂L dFi (Pi )
= −λ =0
∂Pi dPi
Example
determine the economic operating point for the three
generating units when delivering a total of 850 MW
input-output curves
unit 1: coal-fired steam unit: H1 = 510 + 7.2 P1 + 0.00142 P12
unit 2: oil-fired steam unit: H 2 = 310 + 7.85P2 + 0.00194 P22
unit 3: oil-fired steam unit: H 3 = 78 + 7.97 P3 + 0.00482 P32
fuel costs
coal: $ 3.30 / MBtu
oil: $ 3.00 / MBtu
the individual unit cost rate functions
F1 (P1 ) = H 1 (P1 ) × 3.3 = 1683 + 23.76 P1 + 0.004686 P12
F2 (P2 ) = H 2 (P2 ) × 3.0 = 930 + 23.55P2 + 0.00582 P22
F3 (P3 ) = H 3 (P3 ) × 3.0 = 234 + 23.70 P3 + 0.01446 P32
Example
the conditions for an optimal dispatch
dF1 dP1 = 23.76 + 0.009372 P1 = λ
dF2 dP2 = 23.55 + 0.01164 P2 = λ
dF3 dP3 = 23.70 + 0.02892 P3 = λ
P1 + P2 + P3 = 850
solving for λ yields
λ − 23.76 λ − 23.55 λ − 23.70
+ + = 850
0.009372 0.01164 0.02892
λ = 27.41
then solving for the generator power values
P1 = ( 27.41 − 23.76) 0.009372 = 389.8
P2 = ( 27.41 − 23.55) 0.01164 = 331.8
P3 = ( 27.41 − 23.70) 0.02892 = 128.4
In addition to the cost function and the equality constraint
each generation unit must satisfy two inequalities
the power output must be greater than or equal to the minimum
power permitted: Pi ≥ Pi ,min
minimum heat generation for stable fuel burning and temperature
the power output must be less than or equal to the maximum
power permitted: Pi ≤ Pi ,max
maximum shaft torque without permanent deformation
maximum stator currents without overheating the conductor
then the necessary conditions are expanded slightly
dFi dPi = λ ∀Pi : Pi ,min ≤ Pi ≤ Pi ,max
dFi dPi ≥ λ ∀Pi = Pi ,min
dFi dPi ≤ λ ∀Pi = Pi ,max
Example
reconsider the previous example with the following generator
limits and the price of coal decreased to $2.70 / MBtu
generator limits
unit 1: 150 ≤ P1 ≤ 600 MW
unit 2: 100 ≤ P2 ≤ 400 MW
unit 3: 50 ≤ P3 ≤ 200 MW
new fuel cost rate function for unit 1:
F1 (P1 ) = H1 (P1 ) × 2.7 = 1377 + 19.44 P1 + 0.003834 P12
solving for λ yields
λ − 19.44 λ − 23.55 λ − 23.70
+ + = 850
0.007668 0.01164 0.02892
λ = 24.82
P1 = 701.9 P2 = 109.3 P3 = 38.8
Example
this solution meets the constraint of generation meeting the
850 MW load demand, but units 1 and 3 are not within limit
let unit 1 be set to its maximum output and unit 3 to its minimum
output. The dispatch becomes:
P1 = 600 MW P2 = 200 MW P3 = 50 MW
hence, λ must equal the incremental cost of unit 2 since it is the
only unit not at either limit
= 23.55 + 0.01164(200) = 25.88
dF
λ= 2
dP2 P =200
2
next compute the incremental costs for units 1 and 3
= 19.44 + 0.007668(600) = 24.04
dF1
dP1 P =600
1
= 23.7 + 0.02892(50) = 25.15

dF3
dP3 P3 =50
Example
note that the incremental cost for unit 1 is less than λ indicating
that it is at its maximum
however, the incremental cost for unit 3 is not greater than λ so it
should not be forced to its minimum
rework with units 2 and 3 incremental cost equal to λ
λ − 23.55 λ − 23.70
+ = 850 − P1 = 250
0.01164 0.02892
λ = 25.67
P1 = 600 P2 = 182.0 P3 = 68.0
note that this dispatch meets the necessary conditions
incremental cost of electricity = 2.567 cents / kilowatt-hour
Network Losses
Consider a similar system, which now has a transmission
network that connects the generating units to the load
the economic dispatch problem is slightly more complicated
the constraint equation must include the network losses, Ploss
the objective
function, FT F1 1 T G P
1
is the same Pload

as before F2 2 T G P
2
Transmission
the constraint network with
equation must losses, Ploss
be expanded FN N T G P
N
as: N
φ = 0 = Pload + Ploss − Pi
i =1
Network Losses
The same math procedure is followed to establish the
necessary conditions for a minimum-cost operating solution
Lagrange function and its derivatives w.r.t. the input power:
N N
L = FT + λ φ = Fi (Pi ) + λ Pload + Ploss − Pi
i =1 i =1
∂L dFi ∂P dFi ∂P
= − λ 1 − loss = 0 → + λ loss = λ
∂Pi dPi ∂Pi dPi ∂Pi
the transmission network loss is a function of the impedances

and the currents flowing in the network
for convenience, the currents may be considered functions of the
input and load powers
it is more difficult to solve this set of equations
Network Losses
Example
repeat the first example, but include a simplified loss
expression for the transmission network
Ploss = 0.00003P12 + 0.00009 P22 + 0.00012 P32
the incremental cost functions and the constraint function are
formed as:
dFi ∂Ploss
= λ 1−
dPi ∂Pi
dF1 dP1 = 23.76 + 0.009372 P1 = λ [1 − 2(0.00003)P1 ]
dF2 dP2 = 23.55 + 0.01164 P2 = λ [1 − 2(0.00009 )P2 ]
dF3 dP3 = 23.70 + 0.02892 P3 = λ [1 − 2(0.00012 )P3 ]
P1 + P2 + P3 − 850 = Ploss = 0.00003P12 + 0.00009 P22 + 0.00012 P32
this is no longer a set of linear equations as before
Network Losses
Example
a new iterative solution procedure
step 1 pick starting values for P1, P2, and P3 that sum to the load
step 2 calculate ∂Ploss/∂Pi and the total losses Ploss
step 3 calculate λ that causes P1, P2, & P3 to sum to Pload & Ploss
step 4 compare P1, P2, & P3 of step 3 to the values used in step 2; if
there is significant change to any value, go back to step 2, otherwise,
the procedure is done
pick generation values
P1 = 400 MW P2 = 300 MW P3 = 150 MW
find the incremental losses
∂Ploss ∂P1 = 2(0.00003)(400 ) = 0.0240
∂Ploss ∂P2 = 2(0.00009 )(300) = 0.0540
∂Ploss ∂P3 = 2(0.00012 )(150) = 0.0360
Network Losses
Example
total losses
Ploss = 3 × 10−5 (400 ) + 9 × 10−5 (300 ) + 12 × 10−5 (150 ) = 15.6 MW
2 2 2
solve for λ
23.76 + 0.009372 P1 = λ (1 − 0.024 )
23.55 + 0.01164 P2 = λ (1 − 0.054 )
23.70 + 0.02892 P3 = λ (1 − 0.036 )
P1 + P2 + P3 − (850 + 15.6 ) = 0
in matrix form
− 0.009372 0 0 0.976 P1 23.76
0 − 0.01164 0 0.946 ⋅ P2 = 23.55
0 0 − 0.02892 0.964 P3 23.70
1 1 1 0 λ 865.6
P1 = 437.20 P2 = 296.49 P3 = 131.91 λ = 28.54
Network Losses
Example
since the values of P1, P2, and P3 are quite different from the
starting values, we return to step 2
find the incremental losses and total losses
∂Ploss ∂P1 = 2(0.00003)(437.2 ) = 0.0262
∂Ploss ∂P2 = 2(0.00009 )(296.5) = 0.0534
∂Ploss ∂P3 = 2(0.00012 )(131.9 ) = 0.0317
Ploss = 3×10−5 (437.2 ) + 9×10 −5 (296.5) + 12×10−5 (131.9 ) = 15.73 MW
2 2 2
solve for λ in matrix form

− 0.009372 0 0 0.9738 P1 23.76
0 − 0.01164 0 0.9466 ⋅ P2 = 23.55
0 0 − 0.02892 0.9683 P3 23.70
1 1 1 0 λ 865.73
P1 = 431.03 P2 = 298.38 P3 = 136.33 λ = 28.55
Network Losses
Example
summarization of the iteration process
iteration P1 P2 P3 losses λ
count (MW) (MW) (MW) (MW) ($/MWh)
1 400.00 300.00 150.00 15.60 28.54

2 437.20 296.49 131.91 15.73 28.55
3 431.03 298.38 136.33 15.82 28.55
4 432.45 297.92 135.45 15.80 28.55
5 432.11 298.06 135.63 15.81 28.55
6 432.19 298.02 135.59 15.80 28.55
7 432.17 298.03 135.60 15.80 28.55
EEL 6266
Chapter 3
Numerical Methods for Economic Dispatch
The Lambda-Iteration Method
The solution to the optimal dispatch can be approached by
graphical methods
plot the incremental cost characteristics for each generator
the operating points must have minimum cost and satisfy load
that is, find an incremental cost rate, λ that meets the demand PR
graphically:
dF1 dF2 dF3
dP1 dP2 dP3
($/MWh) ($/MWh) ($/MWh)
P1 (MW) P2 (MW) P3 (MW)

Σ
PR = P1 + P2 + P3
An iterative process error (e)
assume an incremental cost [1]
solution: (|e| < tolerance)
rate λ and find the sum of the
power outputs for this rate [3]
the first estimate will be
incorrect
0 λ
if the total power output is too λ[2] λ[3] λ[1]
low, increase the λ value, or if
too high, decrease the λ value
[2]
with two solutions, a closer
N
value of total power can be
e= Pi − PR
extrapolated or interpolated i =1
the steps are repeated until the Lambda projection
desired output is reached
This procedure can be adopted for start set λ
a computer implementation
the implementation of the power calculate Pi
for i = 1 to N
output calculation is rather
independent of the solution method calculate
N
each generator output could be ε = Pload − Pi
solved by a different method i =1
as an iterative procedure, a stopping

criterion must be established first iteration?
two general stopping rules are
project λ
appropriate for this application
total output power is within a |ε| ≤ tolerance?
specified tolerance of the load demand
iteration loop count exceeds
end print schedule
a maximum value
Example
consider the use of cubic functions to represent the input-
output characteristics of generating plants
H (MBtu/h ) = A + BP + CP 2 + DP 3 (P in MW )
for three generating units, find the optimum schedule for a
2500 MW load demand using the lambda-iteration method
generator characteristics:
A B C D Pmax Pmin
Unit 1 749.55 6.95 9.68×10-4 1.27×10-7 320 800
Unit 2 1285.0 7.051 7.375×10-4 6.453×10-8 300 1200
Unit 3 1531.0 6.531 1.04×10-3 9.98 ×10-8 275 1100
assume that the fuel cost to be $1/MBtu

set the value of λ on the second iteration at 10% above or below
the starting value depending on the sign of the error
Example
initial iteration: λstart = 8.0
incremental cost functions
λ = dF1 dP1 = 6.95 + 2(9.68 × 10−4 )P1 + 3(1.27 × 10−7 )P12
λ = dF2 dP2 = 7.051 + 2(7.375 × 10−4 )P2 + 3(6.453 × 10−8 )P22
λ = dF3 dP3 = 6.531 + 2(1.04 × 10−3 )P3 + 3(9.98 × 10−8 )P32
find the roots of the three incremental cost functions at λ = 8.0
P1 = (–5575.6, 494.3), P2 = (–8215.9, 596.7), P3 = (–7593.4, 646.2)
use only the positive values within the range of the generator upper
and lower output limits
calculate the error
e = 2500 − (494.3) − (596.7 ) − (646.2 ) = 762.9 MW/h
with a positive error, set second λ at 10% above λstart: λ[2] = 8.8
Example
second iteration: λ[2] = 8.8
find the roots of the three incremental cost functions at λ = 8.8
P1 = (–5904, 822.5), P2 = (–8662, 1043.0), P3 = (–7906, 958.6)
calculate the error
e = 2500 − (822.5) − (1043) − (958.6 ) = −324.0 MW/h
error out of tolerance
project λ
[2 ] [1]
λ − λ 8 .8 − 8 .0
λ[3] = [1] [2 ] (e[2 ] ) + λ[2 ] = (− 324.0) + 8.8 = 8.5615
e −e 762.9 + 324.0
continue with third iteration
Example
results of all iterations
Iteration λ Total Generation P1 P2 P3
1 8.0 1737.2 494.3 596.7 646.2
2 8.8 2824.1 822.5 1043.0 958.6
3 8.5615 2510.2 728.1 914.3 867.8
4 8.5537 2499.9 725.0 910.1 864.8
Issues
under some initial starting points, the lambda-iteration
approach exhibits an oscillatory behavior, resulting in a non-
converging solution
try the example again with a starting point of λstart = 10.0
The Gradient Method
Suppose that the cost function is more complex
P − a4
example: F ( P ) = a0 + a1 P + a2 P 2.5 + a3 e a5
the lambda search technique requires the solution of the

generator output power for a given incremental cost
possible with a quadratic function or piecewise linear function
hard for complicated functions; we need a more basic method
The gradient search method uses the principle that the
minimum is found by taking steps in a downward direction
from any starting point, x[0], one
finds the direction of steepest
descent by computing the ∂F dx1
negative gradient of F at x[0]: − ∇F (x [0] ) = −
∂F dxn
The Gradient Method
to move in the direction of maximum descent from x[0] to x[1]:
x [1] = x [ 0 ] − α ∇f (x [ 0] )
α is a scalar that when properly selected guarantees that the
process converges
the best value of α must be determined by experiment
for the economic dispatch N N
problem, the gradient L= Fi (Pi ) + λ Pload − Pi
technique is applied directly i =1 i =1
to the Lagrange function

∂L ∂P1
(d dP1 )F1 (P1 ) − λ
the gradient function is: ∇L = = (d dPN )FN (PN ) − λ

∂L ∂PN N
∂L ∂λ Pload − Pi
this formulation does not i =1
enforce the constraint function
The Gradient Method
Example
solve the economic dispatch for a total load of 800 MW using
these generator cost functions
F1 (P1 ) = 1683 + 23.76 P1 + 0.004686 P12
F2 (P2 ) = 930 + 23.55P2 + 0.00582 P22
F3 (P3 ) = 234 + 23.70 P3 + 0.01446 P32
use α = 100% and starting from
P1[ 0 ] = 300 MW, P2[ 0 ] = 200 MW, and P3[ 0 ] = 300 MW
λ is initially set to the average of the incremental costs of the
generators at their starting generation values:
23.76 + 0.009372(300) +
Fi (Pi [ 0 ] ) =
3
d 1
λ[0] = 13 23.55 + 0.01164( 200) + = 28.27
i =1 dP1 3
23.70 + 0.02892(300)
The Gradient Method
% Example 3E
gendata = [ 1683 23.76 0.004686
Example
930 23.55 0.00582
234 23.70 0.01446 ]; Matlab program to perform the
power = [ 300, 200, 300 ];
alpha = 1.00, Pload = 800; gradient search method
% find lambda0
n = length( gendata );
lambda0 = 0;
for i = 1 : n
lambda0 = lambda0 + gendata(i,2) + 2 * gendata(i,3) * power(i);
end
lambda0 = lambda0 / 3
clear x0
x0 = power, x0(n+1) = lambda0;
% calculate the gradient
for kk = 1 : 10
disp(kk)
clear gradient
gradient = [];
Pgen = 0, cost = 0;
for i = 1 : n
gradient(i) = gendata(i,2) + 2 * gendata(i,3) * x0(i) - x0(n+1);
Pgen = Pgen + x0(i);
cost = cost + gendata(i,1) + gendata(i,2) * x0(i) + gendata(i,3) * x0(i) * x0(i);
end
gradient(n+1) = Pload - Pgen;
disp( [x0, Pgen, cost/1000] )
x1 = x0 - gradient * alpha;
x0 = x1;
end
The Gradient Method
Example
the progress of the gradient search is shown in the table below
Iteration λ Total Generation P1 P2 P3 Cost

1 28.28 800.0 300.0 200.0 300.0 23,751
2 28.28 800.0 301.7 202.4 295.9 23,726
3 28.28 800.1 303.4 204.8 291.9 23,704
4 28.35 800.2 305.1 207.1 288.1 23,685
5 28.57 800.7 306.8 209.5 284.4 23,676
6 29.23 801.8 308.7 212.1 281.0 23,687
7 31.06 805.0 311.3 215.3 278.4 23,757
8 36.08 813.7 315.7 220.3 277.7 23,983
9 49.79 837.4 325.1 230.3 282.1 24,632
10 87.19 901.9 348.0 253.8 300.0 26,449
note that there is no convergence to a solution
The Gradient Method
A simple variation
realize that one of the generators is always a dependent
variable and remove it from the problem
for example, picking P3, then P3 = 800 − P1 − P2
then the total cost function becomes
C = F1 (P1 ) + F2 (P2 ) + F3 (800 − P1 − P2 )
this function stands by itself as a function of two variables with
no load-generation balance constraint d dF1 dF3
the cost can be minimized C −
by a gradient method such as: ∇C =
dP1 dP1 dP1
=
d dF2 dF3
C −
dP2 dP2 dP2
note that the gradient goes to zero
when the incremental cost at generator 3
is equal to that at generators 1 and 2
The Gradient Method
A simple variation
the gradient steps are performed in like manner as before
x[1] = x [ 0 ] − ∇C ⋅ α
and
P1
x=
P2
Example
rework the previous example with the reduced gradient
dF1 dF3
−
dP1 dP1 23.76 + 2(0.004686)P1 − 23.70 − 2(0.01446)(800 − P1 − P2 )
∇C = =
dF2 dF3
− 23.55 + 2(0.00582)P2 − 23.70 − 2(0.01446)(800 − P1 − P2 )
dP2 dP2
α is set to 20.00
The Gradient Method
% Example 3F Example
gendata = [ 1683 23.76 0.004686
930 23.55 0.00582
234 23.70 0.01446 ];
Matlab program to perform the
power = [ 300, 200, 300 ];
alpha = 20.00;
simplified gradient search method
Pload = 800;
% form lambda0
clear x0
x0 = power(1:n-1);
% calculate the gradient
for kk = 1 : 10
disp(kk)
clear gradient
gradient = [];
Pn = Pload;
for i = 1 : n - 1
Pn = Pn - x0(i);
end
cost = gendata(n,1) + gendata(n,2) * Pn + gendata(n,3) * Pn * Pn;
for i = 1 : n - 1
gradient(i) = gendata(i,2) + 2 * gendata(i,3) * x0(i) - gendata(n,2) - 2 * gendata(n,3) * Pn;
cost = cost + gendata(i,1) + gendata(i,2) * x0(i) + gendata(i,3) * x0(i) * x0(i);
end
disp( [x0, Pn, 800, cost/1000] )
x1 = x0 - gradient * alpha;
x0 = x1;
end
The Gradient Method
Example
the progress of the simplified gradient search is shown in the
table below
Iteration Total Generation P1 P2 P3 Cost
1 800.0 300.0 200.0 300.0 23,751

2 800.0 416.1 330.0 54.0 23,269
3 800.0 368.1 287.4 144.5 23,204
4 800.0 381.5 307.1 111.4 23,194
5 800.0 373.3 303.0 123.7 23,193
6 800.0 373.6 307.0 119.3 23,192
7 800.0 371.4 307.6 121.0 23,192
8 800.0 370.6 309.0 120.4 23,192
9 800.0 369.6 309.7 120.7 23,192
10 800.0 368.9 310.4 120.6 23,192
note that there is a solution convergence by the 6th iteration

Newton’s Method
The solution process can be taken one step further
observe that the aim is to always drive the gradient to zero
∇Lx = 0
since this is just a vector function, Newton’s method finds the
correction that exactly drives the gradient to zero
Review of Newton’s method
suppose it is desired to drive the function g(x) to zero
the first two terms of the Taylor’s series suggest the following
g ( x + ∆x ) = g (x ) + [g ′(x )]∆x = 0 g1 ( x1 , , xn )
the objective function g(x) is defined as: g ( x ) =
g n ( x1 , , xn )
then the Jacobian is:
∂g1 ∂x1 ∂g1 ∂xn
g ′( x ) =
∂g n ∂x1 ∂g n ∂xn
Newton’s Method
the adjustment at each iteration step is ∆x = −[g ′(x )] g ( x )
−1
if the function g is the gradient vector ∇Lx, then

−1
∂
∆x = − ∇Lx ∆L
∂x N N
For economic dispatch problems: L = Fi (Pi ) + λ Pload − Pi
i =1 i =1
d2 L d2 L
dx12 dx1dx2
2
and ∂ d L d2 L
∇Lx = dx dx dx22
∂x 2 1
d2 L d2 L
dλdx1 dλdx2
note that in general, one Newton step solves for a correction that
is closer to the minimum than would the gradient method
Newton’s Method
Example
solve the previous economic dispatch problem example using
the Newton’s method
the gradient function is the same
as in the first example
let the initial value of d 2 F1
0 0 −1
λ be equal to zero dP12
the Hessian matrix d 2 F2
−1
takes the following form: [H ] = 0
dP22
0
d 2 F3
0 0 −1
dP32
the initial generation values −1 −1 −1 0
are also the same as in the
first example
Newton’s Method
% Example 3G Example
gendata = [ 1683 23.76 0.004686
930 23.55 0.00582
234 23.70 0.01446 ];
Matlab program to perform the
power = [ 300, 200, 300 ];
Pload = 800;
Newton’s method
% form H
H = zeros(n+1,n+1);
for i = 1 : n
H(i,i) = gendata(i,3) * 2;
H(i,n+1) = -1, H(n+1,i) = -1; end
x0 = zeros(n+1,1);
x0(1:n,1) = transpose( power );
% calculate the gradient and Hessian matrices
for kk = 1 : 10
disp(kk)
gradient = zeros(n+1,1);
gradient(n+1,1) = Pload;
for i = 1 : n
gradient(i,1) = gendata(i,2) + 2 * gendata(i,3) * x0(i,1) - x0(n+1,1);
gradient(n+1,1) = gradient(n+1,1) - x0(i,1); end
dx = H \ gradient;
cost = 0;
for i = 1 : n
cost = cost + gendata(i,1) + gendata(i,2) * x0(i) + gendata(i,3) * x0(i) * x0(i); end
disp( [x0', cost/1000] )
x0 = x0 - dx;
end
Newton’s Method
Example
the progress of the gradient search is shown in the table below
Iteration λ Total Generation P1 P2 P3 Cost

1 0.00 800.0 300.0 200.0 300.0 23,751
2 27.19 800.0 366.3 313.0 120.7 23,192
3 27.19 800.0 366.3 313.0 120.7 23,192
4 27.19 800.0 366.3 313.0 120.7 23,192
note the quick convergence to a solution

compare with the solution of the previous example
EEL 6266
Chapter 3
Economic Dispatch
Using Dynamic Programming
Piecewise Linear Cost Functions
Common practice
many utilities prefer to represent their generator cost functions
as single- or multiple-segment, linear cost functions
Typical examples:
F(P) F(P)
Pmin Pmax Pmin Pmax

dF(P)/dP dF(P)/dP
Pmin Pmax Pmin Pmax

Piecewise Linear Cost Functions
Piecewise linear cost functions can not be used with gradient
based optimization methods
like the lambda-iteration
such methods will always land on Pmin or Pmax
A table-based method resolves this problem
technique
for all units running, begin to raise the output of the unit with the
lowest incremental cost segment
if this unit hits the right-hand end of a segment or hits Pmax, find
the unit with the next lowest incremental cost segment and begin
to raise its output
eventually, the total of all units outputs equals the total load
the last unit is adjusted to have a generation, which is partially
loaded for one segment
Dynamic Programming
A wide variety of control and dynamic optimization
problems use dynamic programming (DP) to find solutions
can greatly reduce the computation effort in finding optimal
trajectories or control policies
DP applications have been developed for
economic dispatch
hydro-thermal economic-scheduling
unit commitment
methods are based on the calculus of variations
but, applications are not difficult to implement or program
principles are introduced by presenting examples of one-
dimensional problems
Dynamic Programming
Example
consider the cost of transporting a unit shipment from location
A to location N
there are many short paths that connect many stops along the
way, which offers numerous parallel routes from getting from A
to N
each path has an associated cost
e.g., distance and level of difficulty results in fuel costs
the total cost is the sum of the path costs of the selected route
from the originating location to the terminating location
the problem is to find the minimum cost route
Dynamic Programming
H
11 3
B
8 9
E
5 2
8 3
I L 4
2 4 6
A C 11
N
7
9 F
5
J M 3
3 8
6
9
D 5
6 G K
4
1-D Dynamic Programming Example

Dynamic Programming
There are various stages traversed
starting at A, the minimum cost path to N is ACEILN
starting at C, the least cost path to N is CEILN
starting at E, the least cost path to N is EILN
starting at I, the least cost path to N is ILN
starting at L, the least cost path to N is LN
Obtaining the optimal route
the choice of the route is made in sequence
Theory of optimality
the optimal sequence is called the optimal policy
any sub-sequence is called a sub-policy
the optimal policy contains only optimal sub-policies
Dynamic Programming
Example (continued)
divide up the field of paths into stages (I, II, III, IV, V)
at the terminus of each stage, there is a set of nodes (stops), {Xi}
at stage III, the stops are [{X3} = {H, I, J, K}]
a set of costs can be found for crossing a stage, {VIII(X2, X3)}
a cost is dependent on the starting and terminating nodes of a
stage, VIII(E, H) = 3, VIII(F, I) = 11
the minimum cost for traversing from stage I to stage i and
arrive at some particular node (stop), Xi, is defined as fI(Xi)
the minimum costs from stage I to stage II for nodes {B, C, D}
are:
fI(B) = VI(A, B) = 5, fI(C) = VI(A, C) = 2, fI(D) = VI(A, D) = 3
the minimum cost from stage I to stage III for node {E} is:
fII(E) = min [ fI(X1) + VII(X1, E) ] = min[ 5 + 11, 2 + 8, 3 + inf. ]
{X1} X1 = B =C =D
fII(E) = 10 via ACE
Dynamic Programming
I II III IV V
H
B 11 3 13
5 E 9
8
5 10 2
8 I 3
L
12 15
4
A 2 C 4 6 N
11
0 2 F 7 19
9 6 5 J M
3
3
6 11 8 18
9
D G 5
3 6 K
9 4
13
1-D dynamic programming example: cost at each node
Dynamic Programming
at each stage, the minimum cost should be recorded for all the
terminus nodes (stops)
use the minimum cost of the terminus of the previous stage
identify the minimum cost path for each of the terminating nodes
of the current stage
(X1) fI(X1) path (X2) fII(X2) path (X3) fIII(X3)path (X4) fIV(X4)path (X5) fV(X5) path
B 5 A E 10 AC H 13 ACE L 15 ACEI N 19 ACEIL
C 2 A F 6 AC I 12 ACE M 18 ADGK
D 3 A G 9 AD J 11 ACF
K 13 ADG
Dynamic Programming
Economic dispatch
when the heat-rate curves exhibit nonconvex characteristics
it is not possible to use an equal
H(P)
incremental cost method
multiple values of MW output
exist for a given value of
incremental cost Pmin Pmax
dynamic programming finds dH(P)/dP
optimal dispatch under such
circumstances
the DP solution is accomplished
as an allocation problem Pmin Pmax
A nonconvex heat rate curve
the approach generates a set of outputs and its corresponding
for an entire set of load values incremental heat rate curve
Dynamic Programming
Example
consider a three-generator system serving a 310 MW demand
the generator I/O characteristics are not smooth nor convex
Power Levels (MW) Costs ($/hour)
P1 = P2 = P3 F1 F2 F3
0 ∞ ∞ ∞
50 810 750 806
75 1355 1155 1108.5
100 1460 1360 1411
125 1772.5 1655 1704.5
150 2085 1950 1998
175 2427.5 ∞ 2358
200 2760 ∞ ∞
225 ∞ ∞ ∞
the demand does not fit the data exactly, interpolate is needed
between the available closest values, 300 and 325 MW
Dynamic Programming
Example
the minimum cost function for scheduling units 1 and 2:
f 2 (D ) = F1 (D − P2 ) + F2 (P2 )
let P2 cover its allowable range for demands of 100 to 350 MW
D F1(D) P2 = 50 75 100 125 150 (MW) f2 P2*
(MW) ($/h) F2(P2)= 750 1155 1360 1655 1950 ($/h) ($/h) (MW)
50 810 ∞ ∞ ∞ ∞ ∞ ∞ -
75 1355 ∞ ∞ ∞ ∞ ∞ ∞ -
100 1460 1560 ∞ ∞ ∞ ∞ 1560 50
125 1772.5 2105 1965 ∞ ∞ ∞ 1965 75
150 2085 2210 2510 2170 ∞ ∞ 2170 100
175 2427.5 2522.5 2615 2715 2465 ∞ 2465 125
200 2760 2835 2927.5 2820 3010 2760 2760 150
225 ∞ 3177.5 3240 3132.5 3115 3305 3115 125
250 ∞ 3510 2582.5 3445 3427.5 3410 3410 150
275 ∞ ∞ 3915 3787.5 3740 3722.5 3722.5 150
300 ∞ ∞ ∞ 4120 4082.5 4035 4035 150
325 ∞ ∞ ∞ ∞ 4415 4377.5 4377.5 150
350 ∞ ∞ ∞ ∞ ∞ 4710 4710 150
Dynamic Programming
Example
the minimum cost function for scheduling units 1, 2 and 3:
f 3 (D ) = f 2 (D − P3 ) + F3 (P3 )
let P3 cover its allowable range for the demand
D f2(D) P3 = 50 75 100 125 150 175 (MW) f3 P3*
(MW) ($/h) F3(P3) = 806 1108.5 1411 1704.5 1998 2358 ($/h) ($/h) (MW)
100 1560 ∞ ∞ ∞ ∞ ∞ ∞ ∞ -
125 1965 ∞ ∞ ∞ ∞ ∞ ∞ ∞ -
150 2070 2366 ∞ ∞ ∞ ∞ ∞ 2366 50
175 2465 2771 2668.5 ∞ ∞ ∞ ∞ 2668.5 75
200 2760 2976 3073.5 2971 ∞ ∞ ∞ 2971 100
225 3115 3271 3278.5 3376 3264.5 ∞ ∞ 3264.5 125
250 3410 3566 3573.5 3581 3669.5 3558 ∞ 3558 150
275 3722.5 3921 3868.5 3876 3874.5 3963 3918 3868.5 75
300 4035 4216 4223.5 4171 4169.5 4168 4323 4168 150
325 4377.5 4528.5 4518.5 4526 4464.5 4463 4528 4463 150
350 4710 4841 4831 4821 4819.5 4758 4823 4758 150
375 ∞ 5183.5 5143.5 5133.5 5114.5 5113 5118 5113 150
400 ∞ 5516 5486 5446 5427 5408 5473 5408 150
425 ∞ ∞ 5818.5 5788.5 5739.5 5720.5 5768 5720.5 150
Dynamic Programming
Example
the results show:
D Cost P1 P2 P3
300 4168 150 100 50
325 4463 150 125 50
generator #2 is the marginal unit
it picks up all of the additional demand increase between 300 MW
and 325 MW
P1 = 50 MW, P2 = 110 MW, P3 = 150 MW, and Ptotal = 310 MW
the cost is easily determined using interpolation
1655 − 1360
F2 (110) = ⋅ (110 − 100) + 1360 = 1478
125 − 100
F1 = $ 810, F2 = $1478, F3 = $ 1998, and Ftotal = $4286
Ramp Rate Constraints
Generators are usually under automatic generation control
(AGC)
a small change in load and a new dispatch causes the AGC to
change the outputs of appropriate units
generators must be able to move to the new generation value
within a short period of time
large steam units have a prescribed “maximum rate limit”,
∆P/ ∆t (MW per minute)
the AGC must allocate the change in generation to other units,
so that the load change can be accommodated quickly enough
the new dispatch may be at the most economic values, but the
control action may not be acceptable if the ramp rate for any
of the units are violated
To produce an acceptable dispatch to the control system, the
ramp rate limits are added to the economic dispatch
formulation
requires a short-range load forecast to determine the most
likely load and load-ramping requirements of the units
system load is given to be supplied at time increments
t = 1 … tmax with loading levels of Ptload
the N generators on-line supply the load at each time increment
N
Pi t = Pload
t
i =1
each unit must obey a rate limit such that

Pi t +1 = Pi t + ∆Pi
− ∆Pi max ≤ ∆Pi ≤ ∆Pi max
The units are scheduled to minimize the cost to deliver the
power over the time period
Fi (Pi t )
tmax N
F total
=
t =1 i =1
constraints
N
Pi t = Pload
t
∀t = 1 tmax
i =1
and
Pi t +1 = Pi t + ∆Pi
− ∆Pi max ≤ ∆Pi ≤ ∆Pi max
the optimization problem can be solved with dynamic

programming
EEL 6266
Chapter 4
Transmission System Effects
Overview
The network’s incremental power losses causes a bias in the
optimal economic scheduling of the generators
the total real power loss increases the total generation demand
the power flow forms the basis for the development of loss
factors for predicting the real power loss
the generation schedule may need to be adjusted by shifting
generation to reduce flows on transmission circuits because
they would otherwise become overloaded
difficult to include the last effect into optimum dispatching
the power flow must be solved to check for violations
The Power Flow Problem
Power flow is the name given to a network solution that
shows currents, voltages, and active & reactive power flows
at every bus within the system
assumptions
balanced system - positive sequence solution only
simple generation and load models
non-linear problem
relates active and reactive power consumption and generation
with voltage magnitudes
uses
design procedures (system planning)
study unique operating problems
provide accurate calculations of loss penalty factors
The Power Flow Problem
The power flow problem consist of a given transmission
network
lines are represented by a pi-equivalent circuit
transformers are represented by a series impedance circuit
generators and loads represent the boundary conditions
loads are given as active and reactive power consumptions
generators are usually described by the active power production
and the terminal voltage
The general power flow equation
one set of equations for each bus in the network
N
Pinj−i = ∑ Ei Ek (Gik cos(θ i −θ k )+ Bik sin (θ i −θ k ))
k =1
N
Qinj−i = ∑ Ei Ek (Gik sin (θ i −θ k )− Bik cos(θ i −θ k ))
k =1
The AC Power Flow
Formulation of the power flow
building the bus-admittance matrix
modeling the transmission network of complex impedances as
related to the system buses
includes line, transformer, and shunt element impedances
general construction rule
if a branch exists between nodes i and j,
Yij = − yij
and
Yii = yi 0 + ∑ yij
j
where j is defined for all branches connected to i
The AC Power Flow
define bus characteristics based on typical information
available
Power flow bus specifications
Active Reactive Voltage Voltage
Bus Type Comments
Power, P Power, Q Magn., |E| Angle, θ
Constant Power Load, Standard load
Constant Power Bus Scheduled Scheduled Calculated Calculated representation
Load / Shunt Element, Only impedance

Constant Impedance Calculated Calculated Calculated Calculated value is given
On-Load Secondary side of

Tap Changer, Voltage Calculated Calculated Scheduled Calculated OLTC
Controlled Bus transformers
Generator/Synchronous Standard
Condenser, Voltage Scheduled Calculated Scheduled Calculated generator
Controlled Bus representation
Reference / Swing Must adjust net
Generator, Slack Bus Calculated Calculated Scheduled Scheduled power to hold
voltage constant
The AC Power Flow
Non-linear system solution method start Gauss-Seidel
PF Method
the ac power flow problem select initial
is cast into a root finding problem voltages
loop
 0
 M  = F ([x L x ]T ) do for all solve for
  1 N
i=1…N Einew=f(Pj,Ej)
0 i≠ref j=1…N
common solution techniques

calculate
Gauss-Seidel save maximum
line flows
voltage change
first AC power flow method
developed for digital computers no yes print
∆Emax≤ε results
method has linear convergence
governing equation: end
1  Pk[sch] − jQk[sch] [ξ −1] 
E [ξ ]
k =  [ξ −1]∗
− ∑ Ykj E j − ∑ Ykj E j 
[ξ ]
Ykk  Ek j<k j >k 

The AC Power Flow
Non-linear system solution techniques
common solution methods
Newton-Raphson method
instead of treating each bus individually in each iteration, the
correction is found for the whole system
Newton’s method is based on the idea of driving the error of a
function to zero by making a correction on all the independent
variables (bus voltage magnitude and angle)
• setting up the equation: f(x) = K
• pick a starting point x0: f(x0) + ε = K
• use Taylor expansion about x0: f(x0) + (df(x0)/dx) ∆x + ε = K
• setting the error to zero: ∆x = [df(x0)/dx]–1 [K – f(x0)]
Newton’s method is an iterative process for a non-linear system of
equations, but it possesses quadratic convergence to a solution
• the set of first order partial differential eq. is called the Jacobian
The AC Power Flow
the set of power equations for each bus in the network
N
Pinj−i = ∑ Ei Ek (Gik cos(θ i −θ k )+ Bik sin (θ i −θ k ))
k =1
N
Qinj−i = ∑ Ei Ek (Gik sin (θ i −θ k )− Bik cos(θ i −θ k ))
k =1
the injected powers on the left-hand side are the knowns for a
load bus
the power mismatch or error is the difference between the left-
hand and right-hand sides with a particular guess of voltages
= Pinj −i − ∑ Ei[ζ ] Ek[ζ ] (Gik cos(θ i[ζ ] − θ k[ζ ] ) + Bik sin (θ i[ζ ] − θ k[ζ ] ))
N
[ζ ]
∆Pi
k =1
∆Qi[ζ ] = Qinj −i − ∑ Ei[ζ ] Ek[ζ ] (Gik sin (θ i[ζ ] − θ k[ζ ] ) − Bik cos(θ i[ζ ] − θ k[ζ ] ))
N
k =1
The AC Power Flow
the incremental correction is defined as
N
∂Pi N
∂P
∆Pi [ζ ]
=∑ ∆θ k + ∑ i ∆ Ek[ζ ]
[ζ ]
k =1 ∂θ k k =1 ∂ Ek
N
∂Qi N
∂Qi
∆Qi[ζ ] =∑ ∆θ k[ζ ] + ∑ ∆ Ek[ζ ]
k =1 ∂θ k k =1 ∂ Ek
the matrix representation of the incremental correction

 ∆P1   ∂P1 ∂θ1 ∂P1 ∂θ 2 L ∂P1 ∂ En −1 ∂P1 ∂ En   ∆θ1 
 ∆P   ∂P ∂θ ∂P2 ∂θ 2 L ∂P2 ∂ En −1

∂P2 ∂ En   ∆θ 2 
 2   2 1
 M = M M O M M  M 
    
 n −1  ∂Qn −1 ∂θ1 ∂Qn −1 ∂θ1
∆ Q L ∂Qn −1 ∂ En −1 ∂Qn −1 ∂ En   ∆ En −1 
 ∆Qn   ∂Qn ∂θ1 ∂Qn ∂θ1 L ∂Qn ∂ En −1 ∂Qn ∂ En   ∆ En 
The AC Power Flow
deriving the Jacobian terms
off-diagonal terms
∂Pi
= Ei Ek (Gik sin (θ i − θ k ) − Bik cos(θ i − θ k ))
∂θ k
∂Pi
= Ei Ek (Gik cos(θ i − θ k ) + Bik sin (θ i − θ k ))
∂ Ek E k
∂Qi
= − Ei Ek (Gik cos(θ i − θ k ) + Bik sin (θ i − θ k ))
∂θ k
∂Qi
= Ei Ek (Gik sin(θ i − θ k ) − Bik cos(θ i − θ k ))
∂ Ek E k
The AC Power Flow
deriving the Jacobian terms
diagonal terms
∂Pi
= −Qi − Bii Ei
2
∂θ i
∂Pi
= Pi + Gii Ei
2
∂ Ei Ei
∂Qi
= Pi − Gii Ei
2
∂θ i
∂Qi
= Qi − Bii Ei
2
∂ Ei Ei
The AC Power Flow
solving the incremental equation
 ∆θ 1   ∆P1 
 ∆θ 2   ∆P 
   2 
 M  = [J ]−1  M 
   
∆ E E
 n −1 n −1  ∆ Q
 n −1 
 ∆ En En   ∆Qn 
gaussian elimination is often used to solve directly for the

changes in voltage magnitudes and angles instead of finding the
matrix inverse of the Jacobian explicitly
the changes in voltage magnitudes and angles are added to the
values that were used at the beginning of the iteration
The AC Power Flow
Newton-Raphson method start Newton-Raphson
PF Method
the solution process runs select initial
voltage values
according to the flowchart
note that the Jacobian calculate all ∆P & ∆Q
matrix is sparse form the Jacobian matrix
find max power mismatch
the matrix algebra is calculate
carried out using Gaussian ∆Pmax≤ε line flows
elimination or one of the ∆Qmax≤ε
print
other various numerical results
methods solve for ∆|E| and ∆θ
using the Jacobian
and power mismatch end
do for all update voltages

i=1…N θiξ+1= θiξ +∆θi
i≠ref |Ei|ξ+1= |Ei|ξ+∆|Ei|
The AC Power Flow
Example 0.01 + j0.2
100 MW
three bus system 0.03 + j0.25
solve the power flow Unit 1

65 MW Bus 1 Bus 2
1.02 pu
0.02 + j0.4
Bus 3
Three Bus Network Unit 2
100 MVA base 1.05 pu
The AC Power Flow
Non-linear system solution techniques
common solution methods
fast-decoupled power flow
the Newton-Raphson is the most robust algorithm used in practice
the Jacobian matrix must be recalculated for each iteration
the set of linear equations must be resolved for each iteration
a faster method was sought after
simplifications in the Jacobian
in high voltage systems, the branch impedance is primarily
reactive, X >> R, X/R > 20
• Bik >> Gik
bus angles values are relatively close in value, |θi – θj| < 5°
• cos(θi – θj) ≈ 1
• sin (θi – θj) ≈ 0
The AC Power Flow
Fast Decoupled Power Flow
simplifications
several of the off-diagonal terms tend towards zero
∂Pi
= Ei Ek (Gik cos(θ i − θ k ) + Bik sin (θ i − θ k )) →= 0
∂ Ek E k
∂Qi
= − Ei Ek (Gik cos(θ i − θ k ) + Bik sin (θ i − θ k )) →= 0
∂θ k
the remaining off-diagonal terms can be reduced
∂Pi
= Ei Ek (Gik sin(θ i − θ k ) − Bik cos(θ i − θ k )) ≈ − Ei Ek Bik
∂θ k
∂Qi
= Ei Ek (Gik sin(θ i − θ k ) − Bik cos(θ i − θ k )) ≈ − Ei Ek Bik
∂ Ek E k
The AC Power Flow
simplifications
several of the diagonal terms also tend towards zero
∂Pi
= Pi + Gii Ei →= 0
2
∂ Ei Ei
∂Qi
= Pi − Gii Ei →= 0
2
∂θ i
the remaining diagonal terms can also be reduced
Qi << Bii Ei
2
∂Pi
= −Qi − Bii Ei ≈ − Bii Ei
2 2
∂θ i
∂Qi
= Qi − Bii Ei ≈ − Bii Ei
2 2
∂ Ei Ei
The AC Power Flow
simplifications
taking into account the zero terms, the incremental corrections
can be rewritten as
N
∂Pi
∆Pi = ∑
[ζ ]
∆θ k[ζ ]
k =1 ∂θ k
N
∂Qi ∆ Ek[ζ ]
∆Qi[ζ ] =∑
k =1 ∂ Ek Ek E k
substituting in the reduced terms
N
∆Pi [ζ ]
= ∑ − Ei Ek Bik ∆θ k[ζ ]
k =1
N ∆ Ek[ζ ]
∆Qi[ζ ] = ∑ − Ei Ek Bik
k =1 Ek
The AC Power Flow
further simplifications
divide the terms by |Ei|
assume |Ek| is approximately equal to one
∆Pi[ζ ] N
= ∑ − Bik ∆θ k[ζ ]
Ei k =1
∆Qi[ζ ] N
= ∑ − Bik ∆ Ek[ζ ]
Ei k =1
the equations in matrix form

[ζ ] [ζ ]
 ∆P1 E   − B11 − B12 L ∆θ1[ζ ]   ∆Q1 E   − B11 − B12 L ∆ E1[ζ ] 
 ∆P[ζ ] 1     [ζ ]  ∆Q2  
 ζ
1
  [ζ ]

E2  = − B21 − B22 L ∆θ 2   E2  = − B21 − B22 L ∆ E2 
[ ]
 2
   
 M   M M O   M   M   M M O  M 
         
The AC Power Flow
resulting equations in general form
 ∆P1 E1   ∆θ1 
 
∆ P
 2 2E = [B ′]  ∆θ 
 2
 M   M 
 ∆Q1 E1   ∆ E1 
   
∆ Q
 2 2 E = [B ′
′ ] ∆
 2E
 M   M 
where the terms in B´ and B" come from the susceptances of the
bus admittance matrix terms
The AC Power Flow
advantages
B´ and B" are constant
calculated once
only B" may need to change, resulting
from a generation VAR limit violation
about 1/4 the number of terms found in the Jacobian
disadvantages
solution convergence failure
when underlying assumptions do not hold
i.e., X/R > 20 and |θi – θj| < 5°
“DC” Power Flow
A significant simplification of the power flow analysis
drop the Q-V equations altogether in the fast-decoupled
approach
results in a completely linear, non-iterative power flow
algorithm
simply assume that all voltage magnitudes, |Ei|, equal 1.0 pu
the system equation becomes
 ∆P1   ∆θ1 
 ∆P   ∆θ 
 2  = [B′]  2 
 M   M 
   
∆
 NP ∆θ
 N
the terms of B´ are as described for the fast decoupled method
“DC” Power Flow
The DC power flow is only good for calculating the MW
flows on transmission lines and transformers
does not solve the problem of MVAR and MVA flows
the power flowing on each branch (line or transformer) is:
Pi k =
1
(θ i − θ k )
xi k
the power injection at each node (bus) is:

N N
Pi = ∑ Pi k = ∑ Bi′k (θ i − θ k )
k =1 k =1
“DC” Power Flow
Example j0.2
100 MW
solve for the MW flows
the system equation Unit 1
65 MW Bus 1 j0.25 Bus 2
j0.4 Unit 2
 7.5 − 5.0 θ1   P1  35 MW
 − 5.0 9.0  θ  =  P 
   2  2
θ 3 = 0.0
Bus 3
Three Bus Network
60 MW
all units are expressed
100 MW
in per units
Unit 1
the solution 65 MW Bus 1
Bus 2
40 MW Unit 2
θ1  0.2118 0.1177 0.65 5 MW 35 MW
θ  = 0.1177 0.1765  − 1 
 2   
Bus 3
 0.02 
= 
Flows Calculated by the DC Power Flow
 − 0. 1
EEL 6266
Chapter 4
Transmission System Effects
Transmission Losses
An illustration using a simple system Ploss = 0.0002 P12
consider a two generator system P1

Min = 70 MW
the generating units are identical Max = 400 MW 500 MW
production costs are modeled P2
using a quadratic equation Min = 70 MW
Max = 400 MW
F1 (P1 ) = F2 (P2 ) = Fi (Pi ) = 400 + 7 Pi + 0.002 Pi 2

the losses on the transmission line are proportional to the square
of the power flow
let both units be loaded to 250 MW Ploss = 12.5 MW
the load would be under served

P1
by 12.5 MW 250 MW
487.5
P2 MW
250 MW
Transmission Losses
solution using the Lagrange equation
L = F1 (P1 ) + F2 (P2 ) + λ (500 + Ploss − P1 − P2 )
Ploss = 0.0002 P12
then
∂L
= 7.0 + 0.004 P1 − λ (1 − 0.0004 P1 ) = 0
∂P1
∂L
= 7.0 + 0.004 P2 − λ = 0
∂P2
∂L
= 500 + 0.0002 P12 − P1 − P2 = 0
∂λ
solution: P1 = 178.88, P2 = 327.50, Ploss = 6.38 MW
cost = F1(P1) + F2(P2) = 4623.15
Transmission Losses
if the optimal dispatch is ignored Ploss = 13.93 MW
and generator 1 is set to supply all

P1
the losses, then 263.93 MW
P1 = 263.93 and Ploss = 13.93 MW 500 MW
P2
total cost = F1(263.93) + F2(250) 250 MW
= 4661.84
optimum dispatch tends toward supplying the losses from the unit
close to the load, resulting in a lower value of losses
the best economics are not necessarily attained at minimum
losses P = 2.08 MW loss
the minimum loss solution:

P1 = 102.08 and P2 = 400 MW P1
102.08 MW
Ploss = 2.08 MW 500 MW
total cost = 4655.43 P2
400 MW (at limit)
Transmission Losses
Derivation of the penalty factor from incremental losses
start with the Lagrange equation for the economic dispatch
N
 N

L = ∑ Fi (Pi ) + λ  Pload + Ploss (P1 , P2 ,K, PN ) − ∑ Pi 
i =1  i =1 
∂L
min L → = 0 ∀Pi min ≤ Pi ≤ Pi max
P ∀i =1KN
i ∂Pi
then
∂L dFi (Pi )  ∂P 
= − λ 1 − loss  = 0
∂Pi dPi  ∂Pi 
rearranging the equation
−1
 ∂Ploss  dFi (Pi )
1 − ∂P  =λ
 i  dPi
Transmission Losses
the incremental loss for bus i is defined as
∂Ploss
∂Pi
the penalty factor for bus i is given as
−1
 ∂Ploss 
Pf i = 1 − 
 ∂Pi 
if the losses increase for an increase in power from bus i, the

incremental cost is positive and the penalty factor is greater than
unity
the rearranged minimizing equations become
dFi (Pi )
Pf i = λ ∀Pi min ≤ Pi ≤ Pi max
dPi
which are called the coordination equations
Transmission Losses
Behaviors of the penalty factor on the coordination equations
when transmission losses are ignored, the penalty factor takes
on the value of unity, making the coordination equation the
same as the incremental cost equation
a penalty factor greater than one (Pfi > 1), representing
increased losses for increased generation at bus i, acts on the
coordination equation as if the original incremental cost
function has been slightly increased
graphically being moved upward
a Pfi < 1 acts on the coordination equation as if the incremental
cost has been slightly decreased
graphically being moved downward
Transmission Losses
Graphical comparison of the coordination equations with
and without accounting for the losses
dF1 dF dF2 dF dF3 dF3

dP1 dF1 Pf1 1 dP2 dF2 Pf 2 2 dP3 dF3 Pf 3
dP1 dP2 dP3
with
dP1 dP2 dP3
penalty
λ ′′ factors
λ′
no penalty
factors
P1′′ P1′ P1 P2′ P2′′ P2 P3′ P3′′ P3
Pf1 = 1.05 Pf2 = 1.00 Pf3 = 0.95
Pi′ = Dispatch ignoring losses

Pi′′= Dispatch with penalty factors
Transmission Losses
The B matrix loss formula
simplified, practical method for loss and incremental cost
calculations
basic formula
Ploss = PT [B ]P + B0T P + B00
where P is the vector of all generator bus net power injections

[B] is a square loss factor matrix of the same dimension as P
B0 is a loss factor vector of the same length as P
B00 is a loss factor constant
alternative form of the equation
Ploss = ∑∑ Pi Bij Pj + ∑ Bi 0 Pi + B00
i j i
Transmission Losses
using the B coefficients in the start
select starting value given: total
of Pi, i = 1…N load, Pload
economic dispatch equations
calculate Ploss using B matrix Economic
equality constraint find demand PD = Pload + Ploss Dispatch
N N N N with
φ = −∑ Pi +Pload +∑∑ Pi Bij Pj +∑ Bi 0 Pi +B00 calculate penalty factor eqs. updated
i =1 i =1 j =1 i =1 for Pfi for i = 1…N penalty
factors
incremental cost equations pick starting λ
∂L dFi  N

= − λ 1 − 2∑ Bij Pj − Bi 0 
 solve coordination equations
∂Pi dPi  j =1  for Pi for i=1…N
the presence of the incremental

check demand
losses couples together the adjust λ
|ΣPi–PD|<ε?
coordination equations
compare new Pi to Pi of last
makes the solution process iteration, save max. change
more difficult check solution
print max|Pi–Pi′|<δ?
iterative solution algorithm end results
Transmission Losses
Example
loss coefficients
 0.0676 0.00953 − 0.00507  − 0.0766 
[ B ] =  0.00953 0.0521 0.00901  B0 =  − 0.00342
   
 − 0.00507 0.00901 0.0294   0.0189 
B00 = 0.040357
cost functions
F1 (P1 ) = 213.1 + 11.669 P1 + 0.00533P12 50.0 ≤ P1 ≤ 200
F2 (P2 ) = 200.0 + 10.333P2 + 0.00889 P22 37.5 ≤ P2 ≤ 150
F3 (P3 ) = 240.0 + 10.833P3 + 0.00741P32 45.0 ≤ P3 ≤ 180
Pload = 210 MW
Transmission Losses
Example
iteration results
iteration Ploss PD λ P1 P2 P3
1 17.8 227.8 12.8019 50.00 85.34 92.49
2 11.4 221.4 12.7929 74.59 71.15 75.69
3 9.0 219.0 12.8098 73.47 70.14 75.39
4 8.8 218.8 12.8156 73.67 69.98 75.18
5 8.8 218.8 12.8189 73.65 69.98 75.18
6 8.8 218.8 12.8206 73.65 69.98 75.18
Transmission Losses
Reference bus versus load center based penalty factors
B matrix assumption
all load currents conform to an equivalent total load current
equivalent load current is the negative sum of all generation
important concepts
an incremental loss is the change in losses ∂Ploss
for an incremental change in generation output ∆P = ∆Pi
∂Pi
loss
the incremental loss for generator bus i assumes

∆Pj = 0 ∀ j ≠ i

that all other generator outputs remain fixed
implied principles when using the B matrix
an incremental increase in generator output is
matched by an equivalent increment in load
alternative approach is to use a reference generator bus
Transmission Losses
Reference generator bus based penalty factors
reference bus always moves for a change in generation
load demand stays constant
Pi new = Pi old + ∆Pi
Prefnew = Prefold + ∆Pref
a change in reference generation would be the negative of the
generation change
flows in the system changes as a result of any generator
adjustments
the change in flows is apt to cause a change in losses
∆Pref = − ∆Pi + ∆Ploss
Transmission Losses
Beta factors
ratio of the negative power change in the reference bus to a
change in generator i:
∆Pref = − β i ∆Pi βi = −
∆Pref
=
(∆Pi − ∆Ploss ) = 1 − ∂Ploss
∆Pi ∆Pi ∂Pi
economic dispatch can now be defined as follows
economic dispatch is reached when an incremental power shift
from any generator to the reference results in no change in net
production cost for any arbitrarily small power change
∂Fi (Pi )
C = ∑ Fi (Pi ) ∆C = ∑ ∆Pi = 0 ∆Pi ≤ ε
∂Pi
this implies
dFi (Pi ) dFref (Pref ) dFi (Pi ) dFref (Pref )

∆Ci = ∆Pi + ∆Pref = ∆Pi − β i ∆Pi = 0
dPi dPref dPi dPref
Transmission Losses
Beta factors
economic dispatch using beta factors
rewriting the equations
dFi (Pi ) dFref (Pref ) 1 dFi (Pi ) dFref (Pref )
∆Pi = β i ∆Pi → ∆Pi = ∆Pi
dPi dPref β i dPi dPref
this is very similar to the coordination equation where the
reciprocal of beta replaces the penalty factor
first order gradient solution method
pick a generation value for the reference bus
set all other generation according to the equation above
dFref (Pref )
check for total demand
 dFi (Pi )

readjust the reference as ∆C = ∑  − βi  ∆Pi

i ≠ ref 
 dPi dPref 
needed until a solution is
reached
Transmission Losses
Finding the reference-bus-based penalty factors
computed directly from the N-R power flow equations
seek to find the ratio of power change at the reference bus when
there is a change, ∆Pi at the i-th generator
∂Pref ∂Pref ∂Pref ∂θ i ∂Pref ∂ Ei
∆Pref = ∑ ∆θ i + ∑ ∆ Ei = ∑ ∆Pi + ∑ ∆Pi
i ∂θ i i ∂ Ei i ∂θ i ∂Pi i ∂ Ei ∂Pi
likewise, find the ratio when there is a reactive power change,

∆Qi at the i-th generator
∂Pref ∂Pref ∂Pref ∂θ i ∂Pref ∂ Ei
∆Pref = ∑ ∆θ i + ∑ ∆ Ei = ∑ ∆Qi + ∑ ∆Qi
i ∂θ i i ∂ Ei i ∂θ i ∂Qi i ∂ Ei ∂Qi
the terms ∂Pref∂θi and ∂Pref∂Ei are derived by

differentiating the power equation for the reference bus
Transmission Losses
the remaining terms are taken from the inverse Jacobian matrix
the resulting equation is
 ∂Pref ∂Pref ∂Pref ∂Pref ∂Pref ∂Pref 
 ∂P L
 1 ∂Q1 ∂P2 ∂Q2 ∂PN ∂QN 
 ∂Pref ∂Pref ∂Pref ∂Pref ∂Pref ∂Pref  −1
= L  [J ]
 ∂θ1 ∂ E1 ∂θ 2 ∂ E2 ∂θ N ∂ EN 
or
 ∂Pref ∂P1   ∂Pref ∂θ1 
 ∂P ∂Q   ∂P ∂ E 
 ref 1  ref 1  Note: in practice, Gaussian
 ∂Pref ∂P2   ∂Pref ∂θ 2 
[ ]
  −1   elimination is employed to
 ∂P ref ∂Q 2  = J T
 ∂Pref ∂ E2  find the reference-bus penalty
 M   M  factors.
   
∂ P
 ref ∂ PN   ∂Pref ∂θ N 
∂Pref ∂QN  ∂Pref ∂ E N 
   
EEL 6266
Chapter 5
Unit Commitment
Load Demand Cycles
Human activity follows cycles
systems supplying services will also experience usage cycles
transportation, communication, and electric power systems
electric power consumption follows a daily, weekly, and
seasonal cycles
high power usage during the day and evening hours
industrial and commercial operations and lighting loads
lower usage on the weekends
higher usage during the summer and winter
greater temperature extremes
Load cycles create economic problems for power generation
it is quite expensive to continuously run all generation, which
is needed to meet the peak power demands
Load Demand Cycles
Definition
commitment means to turn-on a given generation unit
have the prime mover operating the unit at synchronous speed
synchronize and connect the unit to the network grid
Economics
savings are gained by decommitting some of the generation
units when they are not need to meet the current load demand
the engineering problem is committing enough units to meet
current and future load demands while minimizing starting and
operating costs
Load Demand Cycles
Example
consider the cost for operating three generation units
Unit 1: F1(P1) = 561 + 7.92 P1 + 0.001562 P12 150 ≤ P1 ≤ 600
Unit 2: F2(P2) = 310 + 7.85 P2 + 0.00194 P22 100 ≤ P2 ≤ 400
Unit 3: F3(P3) = 93.6 + 9.56 P3 + 0.005784 P32 50 ≤ P3 ≤ 200
what combination of units is best to supply a 550 MW load?

Unit 1
Unit 2
Unit 3
Ftotal
Max
Gen
Gen
Min
P1
P2
P3
F1
F2
F3
Off Off Off 0 0 Infeasible
Off Off On 200 50 Infeasible
Off On Off 400 100 Infeasible
Off On On 600 150 0 400 150 0 3760 1658 5418
On Off Off 600 150 550 0 0 5389 0 0 5389
On Off On 800 200 500 0 50 4911 0 586 5497
On On Off 1000 250 295 255 0 3030 2440 0 5471
On On On 1200 300 267 233 50 2787 2244 586 5617
Load Demand Cycles
Example
notes:
the least expensive way to supply the generation is not with all
three units running or with any combination involving two units
the optimal commitment is to only run unit #1, the most
economic unit
by only running the most economic unit, the load can be supplied
by that unit operating closer to its best efficiency
if another unit is committed, both unit #1 and the other unit will be
loaded further from their best efficiency points, resulting in a
higher net cost
Load Demand Cycles
Daily load patterns
consider the load demand with a simple peak-valley pattern
in order to optimize the operation of the system
units must be shut down as load goes down
then the units must be recommitted as load goes back up
simple approach to the solution is a simple priority list scheme
1500 MW
1150 MW Peak Load
Total Loading
1000 MW
500 MW
450 MW Min. Load
3 PM 9 PM 3 AM 9 AM 3 PM
Time of day
Load Demand Cycles
Example
use a brute force technique to obtain a “shut-down rule” for the range of
loads from 1200 to 500 MW in steps of 50 MW
when load is above 1000 MW, run all three units
loading between 600 MW and 1000 MW, run units #1 & #2
loading below 600 MW, only run unit #1

Load Unit 1 Unit 2 Unit 3
1200 on on on
1500 MW 1150 on on on
1100 on on on
Total Loading
1050 on on on
200 MW Unit #3
1000 MW 1000 on on off
400 MW 950 on on off
Unit #2 900 on on off
850 on on off
500 MW 600 MW 800 on on off
Unit #1 750 on on off
700 on on off
650 on on off
3 PM 9 PM 3 AM 9 AM 3 PM 600 on off off
Time of day 550 on off off
500 on off off
Constraints in Unit Commitment
Primary constraints
enough units are committed to supply the load economically
Spinning reserve constraints
spinning reserve definition
the total amount of on-line, synchronized generation power
committed less the current loading and power losses supplied
protects the network from an unexpected loss of one or more
generation units
typical spinning reserve rules
the reserve is a given percentage of the forecasted demand
must be capable of making up the loss of the most heavily
loaded generation unit
reserves must be spread around the system to avoid transmission
limitations (bottling) and permit parts of the system to run as
“islands”
Example
consider a power system consisting of two isolated regions
transmission tie-lines join the regions and may transfer power up
to a maximum of 550 MW in either direction
five units have been committed to supply 3090 MW of loading
Interchange
Unit Output
Generation
Load (MW)
Regional
Regional
Capacity
Spinning
Reserve
Region
(MW)
(MW)
(MW)
(MW)
(MW)
Unit
Unit
Western region
Units 1, 2 & 3
550 MW 1 1000 900 100

maximum Western 2 800 420 1740 380 1900 160
3 800 420 380 (in)
Eastern region 4 1200 1040 160 160
Eastern 1350 1190
Units 4 & 5 5 600 310 290 (out)
Total 4400 3090 3090 1310 3090
Example
verify the allocation of spinning reserves in the system
western region
generation of largest unit: 900 MW
available spinning reserve
• local: 760 MW; tie-line capacity: 390 MW; eastern region: 450 MW
• total: 1150 MW - load can be supplied
eastern region
generation of largest unit: 1040 MW
available spinning reserve
• local: 450 MW; tie-line capacity: 550 MW; western region: 700 MW
• total: 1000 MW - load can not be completely supplied
lack 40 MW of spinning reserve in the eastern region
commit 40 MW of new generation within the eastern region
Thermal unit constraints
a thermal unit can undergo only gradual temperature changes
results in a time period of several hours to bring a unit on-line
minimum up time: it should not be turned off immediately
minimum down time: once decommitted, the minimum time
before a unit can be recommitted
crew constraint: at a multiple unit plant, there is usually only
enough personnel to start one unit at at time
a certain amount of energy is expended to bring a unit on-line
to slowly bring up the temperature and pressure
this energy does not result in any power delivered from the unit
the energy cost is brought into the unit commitment problem as
a start-up cost
Start-up costs
the start-up cost can vary from a maximum
cold-start value to a much smaller warm-start value
warm unit: a recently turned-off unit with latent heat
that is near the normal operating temperature
two approaches available to treating a thermal unit during its
down time
allow the boiler to cool down and then heat it back up to
operating temperature in time for a scheduled turn-on
provide enough fuel to supply sufficient energy to the boiler to
just maintain the operating temperature
Start-up cost comparison
cooling
allowing the unit to cool down
Cstart-up
start-up cost function:
cooling
− t shut − down

= H cold 1 − e α 
Ccold  F fuel + C fixed break-even
  banking point
banking Cfixed
input sufficient energy into
the boiler to just maintain
the operating temperature
banking cost function:
0
Cbank = H bank F fuel tshut −down + C fixed 0 1 2 3 4 5 6 7 hr
Time-dependent start-up costs
Unit Commitment Solution Methods
Typical utility situation involving the commitment problem
must establish a loading pattern for M periods
have N generation units available to commit and dispatch
the M load levels and operating limits on the N units are such
that any one unit can supply the load demand and any
combination of units can also supply the loads
Commitment by enumeration
a brute force method
total combinations to investigate: 2N – 1
for the total period of M intervals, the maximum number of
possible combinations is: (2N – 1)M
example: for a 24-hour period made up of 1-hr intervals,
a 5 unit network become 6.2 × 1035 combinations
Priority-List Methods
consist of a simple shut-down rule
obtained by an exhaustive enumeration of all unit combinations
at each load level
or obtained by noting the full-load average production cost of
each unit
the full-load average production cost is the net heat rate at full load
multiplied by the fuel cost
various enhancements can be made to the priority-list scheme
by the grouping of units to ensure that various constraints are
met
Typical shut-down rules
at each hour when load is dropping, determine whether
dropping the next unit on the list leaves sufficient generation
to supply the load plus the spinning-reserve requirements
if the supply is not sufficient, keep the unit committed
determine the number of hours before the unit is needed again
if the time is less than the minimum shut-down time for the unit,
keep it committed
perform a cost comparison
the sum of the hourly production costs for the next number of
hours with the next unit to be dropped being committed
and the sum of the restart costs for the next unit based on the
minimum cost of cooling the unit or banking the unit
Example
construct a priority list for the units in the first example using
the same cost equations
Unit 1: F1(P1) = 561 + 7.92 P1 + 0.001562 P12 150 ≤ P1 ≤ 600
Unit 2: F2(P2) = 310 + 7.85 P2 + 0.00194 P22 100 ≤ P2 ≤ 400
Unit 3: F3(P3) = 93.6 + 9.56 P3 + 0.005784 P32 50 ≤ P3 ≤ 200
the full-load average production costs

Unit 1: F1(600) 600 = 9.7922
Unit 2: F2(400) 400 = 9.4010
Unit 3: F3(200) 200 = 11.1848
a strict priority order for these units: [ 2, 1, 3 ]
Example
the commitment scheme
ignoring minimum up/down times and start-up costs
Combination Min MW Max MW
1+2+3 300 1200
1+2 250 1000
2 100 400
notes
this scheme does not completely parallel the shut-down
sequence described in the first example
there unit 2 was shut down at 600 MW leaving unit 1
here unit 1 is shut down at 400 MW leaving unit 2
why the differences? where is the problem?
EEL 6266
Chapter 5
Unit Commitment
Dynamic programming
chief advantage over enumeration schemes is the reduction in
the dimensionality of the problem
in a strict priority order scheme, there are only N combinations
to try for an N unit system
a strict priority list would result in a theoretically correct
dispatch and commitment only if
the no-load costs are zero
unit input-output characteristics are linear
there are no other limits, constraints, or restrictions
start-up costs are a fixed amount
Dynamic programming
the following assumptions are made in this implementation of
the DP approach
a state consists of an array of units
with specified units operating and the rest decommitted (off-line)
a feasible state is one in which the committed units can supply the
required load and meets the minimum capacity for each period
start-up costs are independent of the off-line or down-time
i.e., it is a fixed amount w.r.t. time
no unit shutting-down costs
a strict priority order will be used within each interval
a specified minimum amount of capacity must be operating
within each interval
The forward DP approach
runs forward in time from the initial hour to the final hour
the problem could run from the final hour back to the initial hour
the forward approach can handle a unit’s start-up costs that are a
function of the time it has been off-line (temperature dependent)
the forward approach can readily account for the system’s history
initial conditions are easier to specified when going forward
the minimum cost function for hour K with combination I:
Fcost (K , I ) = min[Pcost (K , I ) + Scost (K − 1, L : K , I ) + Fcost (K − 1, L )]
{ L}
Fcost(K, I) = least total cost to arrive at state (K, I)

Pcost(K, I) = production cost for state (K, I)
Scost(K–1, L: K, I) = transition cost from state (K–1, L) to (K, I)
state (K, I) is the Ith commitment combination in hour K
a strategy is the transition or path from one state at a given
hour to a state at the next hour
X is defined as the number of states to search each period
N is defined as the number of strategies to be saved at each step
these variable allow control of the computational effort
for complete enumeration, the maximum value of X or N is 2N – 1
for a simple priority-list ordering, the upper bound on X is n, the
number of units
reducing N means that information is discarded about the highest
cost schedules at each interval and saving only the lowest N
paths or strategies
there is no assurance that the theoretical optimum will be found
restricted
search
paths
N=3
X=5
X X N X
Interval Interval Interval

K–1 K K+1
Example
consider a system with 4 units to serve an 8 hour load pattern
Incremental No-load Full-load Min. Time
Unit Pmax Pmin Heat Rate Cost Ave. Cost (h)
(MW) (MW) (Btu / kWh) ($ / h) ($ / mWh) Up Down
1 80 25 10440 213.00 23.54 4 2
2 250 60 9000 585.62 20.34 5 3
3 300 75 8730 684.74 19.74 5 4
4 60 20 11900 252.00 28.00 1 1
Initial Condition Start-up Costs Hour Load (MW)

Unit off(-) / on(+) Hot Cold Cold start 1 450
(h) ($) ($) (h) 2 530
1 -5 150 350 4 3 600
2 8 170 400 5 4 540
3 8 500 1100 5 5 400
4 -6 0 0.02 0 6 280
7 290
8 500
Example
to simplify the generator cost function, a straight line
incremental curve is used
the units in this example have linear F(P) functions:
 dF 
F ( P ) = Fno−load +  P
 dP 
the units must operate within their limits
F(P) No-load Incremental
Unit Pmax Pmin Cost Cost
(MW) (MW) ($ / h) ($ / MWh)
1 80 25 213.00 20.88
Fno-load 2 250 60 585.62 18.00
3 300 75 684.74 17.46
4 60 20 252.00 23.80
P
Pmin Pmax
Case 1: Strict priority-list ordering
State Unit status Capacity
the only states examined each hour
5 0010 300 MW
consist of the listed four: 12 0110 550 MW
state 5: unit 3, state 12: 3 + 2 14 1110 630 MW
state 14: 3 + 2 + 1, state 15: all four 15 1111 690 MW
all possible commitments start from state 12 (initial condition)

minimum unit up and down times are ignored
in hour 1:
possible states that meet load demand (450 MW): 12, 14, & 15
Pcost (1, 15) = F1 (25) + F2 (105) + F3 (300) + F1 (20) Economic Dispatch Eq.
= 1735.36 + 20.88(25) + 18.00(105) + 17.46(300) + 23.80(20) = 9861.36
Fcost (1, 15) = Pcost (1, 15) + Scost (0, 12 : 1, 15) DP State Transition Eq.
= (9861.36) + [(350) + (0.02 )] = 10211.38
Case 1
K Pcost Scost Fcost
in hour 1:
15 9861 350 10211
14 9493 350 9843
12 9208 0 9208
minimum at state 12 (9208)

in hour 2:
possible states that meet load demand (530 MW): 12, 14, & 15
Pcost (2, 15) = F1 (25) + F2 (185) + F3 (300) + F1 (20)
= 1735 + 20.88(25) + 18.00(185) + 17.46(300) + 23.80(20) = 11301
Fcost (2, 15) = Pcost (2, 15) + Scost (1, L : 2, 15) DP State Transition Eq.
350 + 9208
= (11301) + min  0 + 9843  = 20859
 
 0 + 10211 
Case 1: DP diagram
state
unit status hour 0 hour 1 hour 2 hour 3 hour 4 hour 5 hour 6 hour 7 hour 8
capacity 450 MW 530 MW 600 MW 540 MW 400 MW 280 MW 290 MW 500 MW
15 9861 11301 13410 11481 8964 6869 7043 10761

1111 10211 20860 33618 43953 52615 58828 64976 74442
690 MW 12 12 12 14 12 12 5 5
0
350 350 350 350 350 750 750
14 9493 10933 12265 0 11113 8593 6490 6665 10393
1110 9843 20492 32472 43585 52244 58449 64597 74074
630 MW 350 12 350 12 350 12 14 350 12 350 12 5 5
0
750 750
12 0 9208 0 10648 10828 0 8308 0 6192 6366 10108
0110 9208 19857 43300 51609 57800 63949 73439
550 MW 12 12 14 12 12 400 5 400 5
0
5 5574 0 5748
0010 57182 62930
300 MW 12 5
total cost: 73,439

priority order list, up-times and down-times neglected
Case 2:
complete enumeration (2.56 × 109 possibilities)
fortunately, most are not feasible because they do not supply
sufficient capacity
in this case, the true optimal commitment is found
the only difference in the two trajectories occurs in hour 3
it is less expensive to turn on the less efficient peaking unit #4 for
three hours than to start up the more efficient unit #1 for that same
time period
only minor improvement to the total cost
case 1: 73,439
case 2: 73,274
Case 2: DP diagram
state
unit status hour 0 hour 1 hour 2 hour 3 hour 4
capacity 450 MW 530 MW 600 MW 540 MW
15 9861 11301 13410 11481

1111 10211 20860 33618 43953
690 MW 12 12 12 14
350 350 350

14 9493 10933 12265 11113
1110 9843 20492 32472 43585
630 MW 12 12 12 14
350 350 350

13 9576 11016 12450 11113
0111 9576 20221 32507 43503
610 MW 12 12 12 13
12 9208 10648 10828

0110 9208 19857 43135
550 MW 12 12 13
5
0010
300 MW
Lagrange Relaxation
dual variables and dual optimization
consider the classical constrained optimization problem
primal problem: minimize f(x1,…,xn), subject to ω(x1,…,xn) = 0
the Lagrangian function: L(x1,…,xn) = f(x1,…,xn) + λ ω(x1,…,xn)
define a dual function
q(λ ) = min L( x1 , x2 , λ )
x1 , x2
then the “dual problem” is to find
q∗ (λ ) = max q(λ )
λ ≥0
the solution involves two separate optimization problems
in the case of convex functions this procedure is guaranteed to
solve the problem
Example
minimize f(x1,x2) = 0.25x12 + x22
subject to ω(x1,x2) = 5 – x1 – x2
the Lagrangian function: L(x1 , x2 , λ ) = 0.25 x12 + x22 + λ (5 − x1 − x2 )
λ
the dual function: q(λ ) = min L( x1 , x2 , λ ) → x1 = 2λ & x2 =
x1 , x2 2
q(λ ) = − 54 λ2 + 5λ
∂
the dual problem: q∗ (λ ) = max q(λ ) → q(λ ) = 52 λ − 5 = 0
λ ≥0 ∂λ
λ=2
Iterative form of Lagrange relaxation method
the optimization may contain non-linear or non-convex
functions
iterative process based on incremental improvements of λ is
required to solve the problem
select a arbitrary starting λ
solve the dual problem such that q(λ) becomes larger

 d 
update λ using a gradient adjustment: λt = λt +  q(λ )α
d λ 
find closeness to the solution by comparing
the gap between the “primal” function and the dual function
primal function: J* = min L
relative duality gap: J *
− q(*
)
, in practice the gap never reaches
*
zero q
Lagrange relaxation for unit commitment
loading constraint
N
t
Pload − ∑ Pi t U it = 0 ∀t = 1KT
i =1
unit limits
U it Pi min ≤ Pi t ≤ U it Pi min ∀i = 1K N and t = 1KT
unit minimum up-time and down time constraints

the objective function
∑∑ [F (P ) + S ]U = F (Pi t ,U it )
T N
t t
i i start −up i ,t i
t =1 i =1
Formation of the Lagrange function
in a similar way to the economic dispatch problem
L(P,U , λ ) = F (Pi t ,U it ) + ∑ λt  Pload
 T N

t
− ∑ Pi t U it 
t =1  i =1 
unit commitment requires that the minimization of the
Lagrange function subject to all the constraints
the cost function and the unit constraints are each separated over
the set of units
what is done with one unit does not affect the cost of running
another unit as far as the cost function, unit limits, and the up-time
and down-time constraints are concerned
the loading constraint is a coupling constraint across all the units
the Lagrange relaxation procedure solves the unit commitment
by temporarily ignoring the coupling constraint
The dual procedure attempts to reach the constrained
optimum by maximizing the Lagrangian with respect to the
Lagrange multiplier
q∗ (λ ) = max q(λ ) where q(λ ) = min L(P,U , λ )
λt Pit ,U it
done in two basic steps
Step 1: find a value for each λt which moves q(λ) towards a
larger value
Step 2: assuming that λt found in Step 1 is fixed, find the
minimum of L by adjusting the values of Pt and Ut
minimizing L
[
L = ∑∑ Fi (Pi ) + Si ,t U + ∑ λ P ]
T N T T N
t
i
t t t
load − ∑∑ λt Pi t U it
t =1 i =1 t =1 t =1 i =1
[ ]
= ∑∑ { Fi (Pi ) + Si ,t U − λ Pi U }+ ∑ λt Pload
N T T
t t t t t t
i i
i =1 t =1 t =1
separation of the units from one another; the inside term can
now be solved independently for each generating unit
∑ [F (P ) + S ]U
T
i i
t
i ,t i
t
− λt Pi t U it
t =1
the minimum of the Lagrangian is found by solving for the

minimum for each generating unit over all time periods
[ ]
min q(λ ) = ∑ min ∑ { Fi (Pi t ) + Si ,t U it − λt Pi t U it }
N T
i =1 t =1
subject to the up-time and down-time constraints and
U it Pi min ≤ Pi t ≤ U it Pi min ∀t = 1KT
Ui = 1
this is easily solved as a two-
Si Si Si
state dynamic programming
Ui = 0
problem of one variable t=1 t=2 t=3 t=4
Minimizing the function with respect to Pit
at the Uit = 0 state, the minimization is trivial and equals zero
at the Uit = 1 state, the minimization w.r.t. Pit is:
[
min Fi (Pi ) + λt Pi t]
d
dPi t
[F (P ) + λ P ] = ddP F (P ) + λ = 0
i i
t
i
t
t i i
t t
→
d
dPi t
F (P
i i
t
) = λt
i
there are three cases to be considered for Piopt and the limits
if Piopt ≤ Pimin then min [Fi(Pi) – λt Pit] = Fi(Pimin) – λt Pimin
if Pimin ≤ Piopt ≤ Pimax then min [Fi(Pi) – λt Pit] = Fi(Piopt) – λt Piopt
if Pimax ≤ Piopt then min [Fi(Pi) – λt Pit] = Fi(Pimax) – λt Pimax
the two-state DP is solved to minimize the cost of each unit

for Uit = 0 the minimum is zero; therefore the only way to have a
lower cost is to have [Fi(Pi) – λt Pit] < 0
Adjusting λ
λ must be carefully adjusted to maximize q(λ)
various techniques use a mixture of heuristic strategies and
gradient search methods to achieve a rapid solution
for the unit commitment problem λ is a vector of λt’s to be
adjusted each hour
simple technique
 d 
gradient component: λ = λ +  q(λ )α where
t t
d λ 
q(λ ) = Pload − ∑ Pi t U it
N
heuristic component: d t t
α = 0.01 when d q(λ ) is positive d λ i =1
dλ
α = 0.002 when d q(λ ) is negative
dλ
The relative duality gap
used as a measure of the closeness to the solution
for large real-size power-systems unit-commitment calculations,
the duality gap becomes quite small as the dual optimization
proceeds
the larger the commitment problem, the smaller the gap
the convergence is unstable at the end
some units are being switched in and out
the process never comes to a definite end
there is no guarantee that when the dual solution process stops, it
will be at a feasible solution
the gap equation: (J *
− q* )
q*
Lagrange relaxation algorithm start
using dual optimization pick starting λt

for t = 1 … T
k=0
two-state dynamic
update λt loop for program with T
for all t each unit i stages and solve
for Pit and Uit
solve for the
dual value q*(λt)
solve the economic

dispatch for each hour
using committed units
make final
calculate the adjustments
primal value (λt) to schedule
to achieve
feasibility
J*–q*
gap > threshold
q* gap < threshold end
Example
find the unit commitment for a three-generator system over a
four-hour segment of time
assume no start-up cost and no minimum up-time or down-time
t t
Pload (MW )
F1 (P1 ) = 500 + 10 P1 + 0.002 P12 100 < P1 < 600 1 170
F2 (P2 ) = 300 + 8 P2 + 0.0025P22 100 < P2 < 400 2 520
F3 (P3 ) = 100 + 6 P3 + 0.005P32 50 < P3 < 200 3 1100
4 330
start with all λt values set to zero
execute an economic dispatch for each hour when there is
sufficient generation committed for that hour
if there is not enough generation, arbitrarily set the cost to 10,000
the primal value J* is the total generation cost summed over all t
Iteration 1:
Hour λ u1 u2 u3 P1 P2 P3 dq(λ)/dλ P1edc P2edc P3edc
1 0.0 0 0 0 0 0 0 170 0 0 0
2 0.0 0 0 0 0 0 0 520 0 0 0
3 0.0 0 0 0 0 0 0 1100 0 0 0
4 0.0 0 0 0 0 0 0 330 0 0 0
q(λ) = 0.0 J* = 40,000 (J* – q*) / q* = undefined

Iteration 2:
dynamic programming for unit #3
λ = 1.7 5.2 11.0 3.3
F(P) – λP = 327.5 152.5 -700.0 247.5
P31 = P3min P32 = P3min P33 = P3max P34 = P3min
U3 = 1
U3 = 0
t=1 t=2 t=3 t=4
Iteration 2:
1 1.7 0 0 0 0 0 0 170 0 0 0
2 5.2 0 0 0 0 0 0 520 0 0 0
3 11.0 0 1 1 0 400 200 500 0 0 0
4 3.3 0 0 0 0 0 0 330 0 0 0
q(λ) = 14,982 J* = 40,000 (J* – q*) / q* = 1.67
Iteration 3:
1 3.4 0 0 0 0 0 0 170 0 0 0
2 10.4 0 1 1 0 400 200 -80 0 320 200
3 16.0 1 1 1 600 400 200 -100 500 400 200
4 6.6 0 0 0 0 0 0 330 0 0 0
q(λ) = 18,344 J* = 36,024 (J* – q*) / q* = 0.965
Iteration 4:
1 5.1 0 0 0 0 0 0 170 0 0 0
2 10.24 0 1 1 0 400 200 -80 0 320 200
3 15.8 1 1 1 600 400 200 -100 500 400 200
4 9.9 0 1 1 0 380 200 -250 0 130 200
q(λ) = 19,214 J* = 28,906 (J* – q*) / q* = 0.502
Iteration 5:
1 6.8 0 0 0 0 0 0 170 0 0 0
2 10.08 0 1 1 0 400 200 -80 0 320 200
3 15.6 1 1 1 600 400 200 -100 500 400 200
4 9.4 0 0 1 0 0 200 130 0 0 200
q(λ) = 19,532 J* = 36,024 (J* – q*) / q* = 0.844
Iteration 6:
1 8.5 0 0 1 0 0 200 -30 0 0 170
2 9.92 0 1 1 0 384 200 -64 0 320 200
3 15.4 1 1 1 600 400 200 -100 500 400 200
4 10.7 0 1 1 0 400 200 -270 0 130 200
q(λ) = 19,442 J* = 20,170 (J* – q*) / q* = 0.037

Remarks
the commitment schedule does not change significantly with
further iterations
however, the solution is not stable (oscillation of unit 2)
the duality gap does reduce
after 10 iterations, the gap reduces to 0.027
good stopping criteria would be when the gap reaches 0.05
EEL 6266
Chapter 6
Generation with Limited Energy Supply
Economic operation requires that expenditures for fuel be
minimized over a period of time
condition #1: no limitation on fuel supply
economic dispatch using only the present conditions as the data
condition #2: energy resources available at a particular plant is
a limiting factor in operations
economic dispatch calculations must account for what has
happened before and what will happen in the future
examples include limited fuel supplies, fix cost fuels, take-or-
pay contracts, surplus fuels, etc.
Take-or-pay fuel supply contracts
consider a system with N classical thermal plants and one
turbine generator fueled under a take-or-pay agreement
the utility agrees to use a minimum amount of fuel over a
contracted time period in order to purchase at a bulk price
if the utility fails to use the minimum amount, it agrees to pay
the minimum charge for the minimum amount
while this unit’s cumulative fuel consumption is below the
minimum, the system is schedule to minimize the total cost,
subject to the constraint that the total fuel consumption for the
period for this unit is equal to a specified amount
once the min. fuel amount is used, the unit is scheduled normally
as a simplification, we will let the maximum fuel consumption is
equal to the minimum amount
consider an N + 1 unit system with operation over jmax time
intervals, and let:
P1
Pij = unit i power output at time j F1 1
PT
Fij = cost for unit i for interval j P2 T FT
F2 2
qTj = fuel input for unit T at time j
FTj = cost for unit T for interval j
PN
Pload j = total load at time j FN 3 Pload
nj = number of hours in interval j

then the problem is defined as
 N
jmax
 jmax jmax N
min ∑  n j ∑ Fij  +∑ n j FTj φ =∑ n j qTj − qtotal = 0 ψ j = Pload , j −∑ Pij − PTj = 0
j =1  i =1  j =1 j =1 i =1
ignoring for the moment the generator limits, the term
jmax
∑n F
j =1
j T j
is a constant because the total fuel to be used at plant T is fixed

the total fuel cost of that fuel is constant and is dropped from the
Lagrange function
  jmax 
L = ∑ n j ∑ Fi j + ∑ λ j  Pload , j − ∑ Pi j − PT j  + γ  ∑ (n j qT j ) − qtotal 
jmax N jmax
 N
j =1 i =1 j =1  i =1   j =1 
the independent variables are the powers Pij and PTj
for any given time period, j = k
∂L dFi k ∂L dqT k
= 0 = nk − λk i = 1K N and = 0 = γ nk − λk
∂Pi k dPi k ∂PT k dPT k
Take-or-pay fuel supply start
γ is referred to as a pseudo- select value for γ

price or a shadow price
for load Pload j
revalues the fuel price of calculate the
a limited fuel supply for economic dispatch
economic dispatching loop
over all j dFij
nj = λ j , i = 1K N
discrete load patterns intervals dPij
∂qTj
solving the fuel limited γn j = λj
∂PTj
dispatch requires dividing
the pattern into intervals
ε = ∑ [n j qT j ] − qtotal
jmax
Load (MW)
j =1
ε ≤ ktolerance end
time False True
Example
find the optimal dispatch for a gas-fired steam plant
HT(PT) = 300 + 6.0PT + 0.0025PT2 [MBtu/h]
50 ≤ PT ≤ 400
Frate = 2.0 $/ccf (1 ccf = 103 ft3)
Hrate = 1100 Btu/ft3
the plant must burn 40×106 ft3 of gas
composite of remaining generation
FS(PS) = 120 + 5.1PS + 0.0012PS2 [$/h]
50 ≤ PS ≤ 500
load pattern
0h - 4h: 400 MW, 4h - 8h: 650 MW, 8h - 12h: 800 MW
12h - 16h: 500 MW, 16h - 20h: 200 MW, 20h - 24h: 300 MW
Example
ignoring the gas constraint, the optimum economic schedule is
period PS PT
1 350 50
2 500 150
3 500 300
4 450 50
5 150 50
6 250 50
operating cost of the composite unit for the 24h period: $ 52,128
total gas consumed: 21.8×106 ft3 (at a cost of $ 80k / 40×106 ft3)
total cost: $ 132,128
Example
consider now the gas constraint
using the gamma search method, the γ value ranges from 0.5 to
0.875 with a final value of 0.8742 $/ccf
the optimum economic schedule is
period PS PT
1 197.3 202.6
2 353.2 296.8
3 446.7 353.3
4 259.7 240.3
5 72.6 127.4
6 135.0 165.0
operating cost of the composite unit for the 24h period: $ 34,937
total gas consumed: 40×106 ft3 (at a cost of $ 80k)
Composite generation production cost functions
composite production cost curves are a useful technique to
mix fuel-constrained and non-fuel-constrained generation
combines N non-constrained units into an equivalent generator
FS (PS ) = F1 (P1 ) + F2 (P2 ) + K + FN (PN )
PS = P1 + P2 + K + PN
dF1 dF2 dF
= =K= N = λ
dP1 dP2 dPN
if one of the units hits a limit, its output is held constant
a simple procedure for generating FS(PS) consists of adjusting
λ from λmin to λmax, where
 dFi   dFi 
λ min
= min , i = 1K N  & λ = max 
max
, i = 1K N 
 dPi   dPi 
Curve finding start
at each increment of λ, set λα = λmin
calculate the total fuel

calculate Piα such if unit i hits
consumption costs and that dFi /dPi = λiα a limit, set
the total power output for i = 1 … N Piα at limit
for all the units N
these data points represent

PS = ∑P
i =1
i
points on the FS(PS) curve N
the composite curve can FS (PS ) = ∑ Fi (Pi )

i =1
be piecewise linear or
smooth (using curve fitting) λα +1 = λα + ∆λ
fit curve to
λα +1 ≥ λmax points of (PS, FS)
False True
end
Example
consider the three generation
units with the following cost
functions
F1 (P1 ) = 561 + 7.92 P1 + 0.001562 P12
150 ≤ P1 ≤ 600
F2 (P2 ) = 434 + 10.99 P2 + 0.002716 P22
100 ≤ P2 ≤ 400
F3 (P3 ) = 117 + 11.955P3 + 0.00723P32
50 ≤ P3 ≤ 200
combine the units into an
equivalent composite
generating unit and find
the equivalent cost function
Lambda Steps for Composite Cost Curve Equivalent unit input / output curve
λ
FS equivalent unit operating cost ($/h)

step Ps Fs Fs (fitted)
1 8.3886 300 4077 4138
2 8.7115 403 4961 4924
3 9.0344 507 5878 5799
4 9.3574 610 6829 6762
5 9.6803 714 7813 7812
6 10.0032 750 8168 8205
7 11.6178 766 8349 8375
8 11.9407 825 9049 9045
9 12.2636 885 9768 9744
10 12.5866 944 10507 10471
11 12.9095 1019 11470 11437
12 13.2324 1088 12369 12361
13 13.5553 1111 12669 12668
14 13.8782 1133 12975 12980
15 14.2021 1155 13288 13295 PS equivalent unit output (MW)
16 14.5241 1178 13609 13615
17 14.8470 1200 13937 13939 Equivalent unit input / output function
FS −approx . (PS ) = 2352.65 + 4.7151PS + 0.0041168PS2
300 ≤ PS ≤ 1200
Finding a solution by the gradient search technique
consider the following equations produced by the
minimization of the Lagrange function
dFi k dqT k
nk = λk i = 1K N and γ nk = λk
dPi k dPT k dFi k
dPi k
then combining the equations yields: γ = dq
Tk
for an optimal dispatch, γ is constant dPT k
for all intervals j, j = 1 … jmax
this fact is used to form a search process by refining γ
qTj is treated as a vector containing jmax terms
γ indicates the gradient of the objective function with respect to qTj
Two gradient methods start
from arbitrary PS schedule
compute FS(PS) and dFS/dPS
a simple search approach
assume feasible schedule for
does not require an PS and PT for all j = 1,…,jmax
initial feasible schedule calculate γj for j = 1,…,jmax
does not require an
ε = ∑ [n j qT j ] − qtotal
jmax
initial feasible fuel

j =1
usage schedule dFS j*
but the approach does dPS j*
γ j* = ε ≤ k tolerance end
dqT j*
not insure optimality dPT j*
False True
ε >0
False True
find j* with maximum γj find j* with minimum γj

and increase fuel use and decrease fuel use
qTj = qTj + ∆qTj for j = j* qTj = qTj – ∆qTj for j = j*
Two gradient methods from a feasible schedule
start
compute FS(PS) and dFS/dPS
a relaxation technique
calculate
requires an initial jmax
assume feasible schedule such that
jmax
Ftotal = ∑ n j FS j
dFS
feasible schedule j =1
∑ n j qT j = qtotal
j =1
& nj
dPS
= λj
requires an initial
feasible fuel usage calculate γj for select j+ and j– such that γj +
j = 1,…,jmax is maximum for j = j+ and
schedule γj – is minimum for j = j–
approach insures calculate new values
of γj for j+ and j– adjust q in j+ and j–,
optimality qT j + = qT j + ∆qj/nj, j = j+
qT j– = qT j – ∆qj/nj, j = j–
adjust PT j+, PT j–
calculate
∆ Ftotal ≤ ε ∆Ftotal = ∆FS j + ∆FS j
j= j+ j= j−
False True
end
Example
find the optimal dispatch for a gas-fired steam plant
HT(PT) = 300 + 6.0PT + 0.0025PT2 [MBtu/h]
50 ≤ PT ≤ 400
Frate = 2.0 $/ccf (1 ccf = 103 ft3)
Hrate = 1100 Btu/ft3
the plant must burn 40×106 ft3 of gas
composite of remaining generation
FS(PS) = 120 + 5.1PS + 0.0012PS2 [$/h]
50 ≤ PS ≤ 500
load pattern
0h - 4h: 400 MW, 4h - 8h: 650 MW, 8h - 12h: 800 MW
12h - 16h: 500 MW, 16h - 20h: 200 MW, 20h - 24h: 300 MW
Example
initial dispatch
period PS PT qTj γj
1 350 50 2.205 1.0454
2 500 150 4.568 1.0267
3 500 300 8.455 0.9240
4 450 50 2.205 1.0877
5 150 50 2.205 0.9610
6 250 50 2.205 1.0032
total fuel consumption: 21.84 M ft3

maximum γ : 1.0877, minimum γ : 0.9240, average γ : 1.0058
increase unit 4 by 137.2 MW
Example
iteration #1
1 350 50 2.205 1.0454
2 500 150 4.568 1.0267
3 500 300 8.455 0.9240
4 312.8 187.2 5.493 0.9279
5 150 50 2.205 0.9610
6 250 50 2.205 1.0032

Example
iteration #2
1 242.1 157.9 4.762 0.9204
2 500 150 4.568 1.0267
3 500 300 8.455 0.9240
4 312.8 187.2 5.493 0.9279
5 150 50 2.205 0.9610
6 250 50 2.205 1.0032

Example
iteration #3
1 242.1 157.9 4.762 0.9204
2 360.6 289.4 8.167 0.8811
3 500 300 8.455 0.9240
4 312.8 187.2 5.493 0.9279
5 150 50 2.205 0.9610
6 250 50 2.205 1.0032

Example
iteration #17
1 199.1 200.9 5.841 0.8760
2 353.7 296.3 8.355 0.8746
3 446.0 354.0 9.954 0.8735
4 260.0 240.0 6.851 0.8745
5 71.6 128.4 4.042 0.8731
6 135.3 164.7 4.932 0.8745

Example
iteration #18
1 197.4 202.6 5.884 0.8743
2 353.7 296.3 8.355 0.8746
3 446.0 354.0 9.954 0.8735
4 260.0 240.0 6.851 0.8745
5 71.6 128.4 4.042 0.8731
6 135.3 164.7 4.932 0.8745

maximum γ : 0.8746, minimum γ : 0.8731, average γ : 0. 8739
increase unit 5 by -0.8 MW
Example
iteration #19
1 197.4 202.6 5.884 0.8743
2 353.7 296.3 8.355 0.8746
3 446.0 354.0 9.954 0.8735
4 260.0 240.0 6.851 0.8745
5 72.4 127.6 4.022 0.8740
6 135.3 164.7 4.932 0.8745

end search
EEL 6266
Chapter 6
Hard limits and slack variables
take account of the limits on the take-or-pay generating unit
PT min ≤ PT ≤ PT max
this may be added to the Lagrangian by using two new
functions and two new variables, called slack variables
ψ 1 j = PT j − PT max + S12j and ψ 2 j = PT min − PT j + S22 j
where S1j and S2j are the slack variables
they may take on any real value including zero
the new Lagrangian
  jmax 
L = ∑ n j ∑ Fi j + ∑ λ j  Pload , j − ∑ Pi j − PT j  + γ  ∑ (n j qT j ) − qtotal 
jmax N
 jmax N
j =1 i =1 j =1  i =1   j =1 
+ ∑α1 j (PT j − PT max + S ) + ∑α (P 2j)

jmax jmax
2
1j 2j T min − PT j + S 2
j =1 j =1
α1j and α2j are Lagrange multipliers
the first partial derivatives for the kth interval are
∂L dF
= 0 = nk i k − λk
∂Pi k dPi k
∂L dq
= 0 = γ nk T k − λk + α1k − α 2 k
∂PT k dPT k
∂L
= 0 = 2α1k S1k
∂α1k
∂L
= 0 = 2α 2 k S2 k
∂α 2 k
when the constrained variable PTk is within bounds, α1j = α2j = 0
and S1j and S2j are nonzero; when PTk is limited, either S1j or S2j
is zero and the associated Lagrange multiplier is nonzero
consider some interval k where PTk = Pmax
S1k = 0 and α1k ≠ 0, then
dqT k
− λk + α1k + γ nk =0
dPT k
if
dqT k
λk > γ nk
dPT k
the value of α1k takes on the value just sufficient to make the
equality true
Example
reconsider the 6-interval 24-hour fuel scheduling
maximum generation on PT is reduced to 300 MW
in the original optimal schedule, PT = 353.3 in the third time-
period
when the limit is reduced to 300 MW, the gas-fired unit burns
more fuel in other time periods to meet the 40 M ft3 gas
consumption constraint
Example
resulting optimal schedule with Ptmax = 300 MW
period PS PT λj γj (∂qTj /∂PTj) α1j
1 183.4 216.6 5.54 5.54 0.
2 350.0 300.0 5.94 5.86 0.08
3 500.0 300.0 6.30 5.86 0.44
4 245.4 254.6 5.69 5.69 0.
5 59.5 140.5 5.24 5.24 0.
6 121.4 178.6 5.39 5.39 0.
shadow price, γ : 0.8603

Fuel scheduling
the major elements in the fuel supply and delivery system
raw-fuel suppliers
coal, oil, and gas companies
long-term contracts with min./max. limits on fuel per time-period
prices may change subject to renegotiation provisions in contract
transportation
railroads, river barges, gas-pipeline companies
represent problems in scheduling of fuel deliveries
fuel storage farms - fuel inventory
coal piles, oil storage tanks, underground gas storage facilities
keeping proper inventory levels to forestall fuel shortages
• load levels exceed forecast, suppliers or shippers unable to deliver
generation stations
Fuel scheduling
fuel scheduling problem
the total time period is broken into discrete time increments
constraint functions
some constrain functions will span two or more time steps
solution technique
linear programming
procedure that minimizes a linear objective function
variables are also subject to linear constraints
any non-linear functions (objective or constrain equations) must be
approximated by linearization about an operating point or by
piecewise linear functions
Linear programming
optimization function
Z = c1 x1 + c2 x2 + K + cN xN
linear constraints
a11 x1 + a12 x2 + K + a1N xN ≤ b1
a21 x1 + a22 x2 + K + a2 N xN ≤ b2
M
aM 1 x1 + aM 2 x2 + K + aMN xN ≤ bM
and the variables may having specified upper and lower limits
x1min ≤ x1 ≤ x1max
x2min ≤ x2 ≤ x2max
M
xNmin ≤ xN ≤ xNmax
One of many Linear programming (LP) techniques
upper-bound linear programming algorithm
variable limits are handled implicitly
requires a slack variable with each constraint
the slack variables
used primarily in inequality constraints
equals the difference or slack between a constraint and its limit
transforms an inequality constraint into an equality constraint
4 x1 + 6 x2 ≤ 28
→ 4 x1 + 6 x2 + xS = 28 0 ≤ xS < ∞
if x1 = 3 and x2 = 2, then xS = 4
for a “greater than or equal to” constraint, the bounds on the slack
variable are changed
equality constraints may contain a slack variable with bounds at 0
Linear programming
linear equations structure (canonical form)
the objective function and constraints are arranged in a matrix
 a11 a12 L a1N  1   xS 1   b1 
a   x1    x  b 
a22 L a2 N   1
 21  x2    S2   2 
 M M O M   + O  M = M 
  M      
a N 1 a12 L a NN     1   xSN  bN 
 xN 
 c1 c2 L cN    − 1  Z   0 
there is at least one variable in each constraint whose coefficient

is zero in all the other constraints
this set of variables is called the basis variables
the remaining variables are called non-basis variables
Linear programming
the solution procedure centers on conducting pivot operations
a pivot exchanges a non-basis variable for a basis variable
pivoting
consider the operation for pivoting variable x1 for xS2
that is equivalent to pivoting column 1 with row 2
steps
multiply the elements in row 2 by: 1 / a21
a2′ j = a2 j a21 j = 1K N b2′ = b2 a21
for each row i (i ≠ 2) subtract from row i the product of row 2 and
ai1 a′ = a − a a′
ij ij j = 1K N
i1 2 j
bi′ = bi − ai1b2′ i = 1K N , i ≠ 2
c′j = c j − c1a2′ j j = 1K N
Linear programming
the result of the pivoting operation
0 a12 ′ L a1′N  1 a1′S 2   xS 1   b1′ 
1 a′ L a′   x1   a′   x   b′ 
2N  
 x2     S2   2 
22 2S 2
M M O M    +  M O  M = M 
 ′ ′  M   ′    ′
 0 a 12 L a NN     a NS 2 1   xSN  bN 
x
0 c2′ L c′N   N   cS′ 2 − 1  Z   z′ 
x1 is now one of the basis variables
Linear programming
the dual upper-bounding algorithm proceeds in simple steps
variables that are in the basis set are exchanged for variables in
the non-basis set using the pivoting operation
the non-basis variables are held equal to either their upper or
lower bounds
the basis variables are allowed to take any value without respect
to their bounds
the solution terminates when all the basis variables are within
their respective limits
The LP algorithm
starting steps
for each variable that has a non-zero coefficient cj in the
objective function (the cost row) must be set according to the
following rule
if cj > 0, set xj = xjmin; else if cj < 0, set xj = xjmax
if cj = 0, xj can be set to any value; for convenience it is set to xjmin
add a slack variable xSj to each constraint in the problem
using the values of xj assigned in the above step, set the sleck
variables to make each constraint equal to its limit
variable exchange
row selection: find the basis variable with the greatest violation
this is the row to pivot on
identified as row R
The LP algorithm
variable exchange using R with the worst violation
column selection 1: use if constrain row R is below its minimum
pick column S so that cS / (-aR,S) is minimum for all S that satisfy:
• S is not in the current basis
• aR,S is not equal to zero
• if xS is at its minimum, then aR,S < 0 and cS ≥ 0
• if xS is at its maximum, then aR,S > 0 and cS ≤ 0
column selection 2: use if constrain row R is above its maximum
pick column S so that cS / (aR,S) is minimum for all S that satisfy:
• S is not in the current basis
• aR,S is not equal to zero
• if xS is at its minimum, then aR,S > 0 and cS ≤ 0
• if xS is at its maximum, then aR,S < 0 and cS ≥ 0
The LP algorithm
variable exchange using row R and column S
pivot at the selected row R and column S
the pivot column’s variable, xS, goes into the basis
if no column fits the column selection criteria, the problem has
an infeasible solution
no values for x1…xN will satisfy all constraints simultaneously
setting the variables after pivoting
all non-basis variables, except for xS hold there current values
the most violated variable xR is set to the limit that was violated
all non-basis variables are determined, set each basis variable to
whatever value is required to make the constraints balance
all basis variables may change, many may now violate the limits
The LP algorithm start
search the basis variables
for the one with the worst
violation, designate it as R
any No
violation? end
Yes
most violated variable most violated variable
is above its maximum violation is below its minimum
type
pick column S using pick column S using
the column selection the column selection
procedure for most procedure for most
violated variable above violated variable below
its maximum its minimum
found Yes Yes found

column S? column S?
No No
infeasible infeasible
solution solution
pivot on selected
row and column
LP example
minimize: Z = 2x1 + x2
subject to: x1 + x2 = 20 constraint #1
–1.4x1 + x2 ≤ 2 constraint #2
2 ≤ x1 ≤ 12 limit #1
2 ≤ x2 ≤ 16 limit #2
–1.4x1 + x2 ≤ 2
cost contours
x2 = 16
x2 = 2
x1 = 2 x1 = 12 x1 + x2 = 20
LP example
problem cast into canonic form:
constraint #1: x1 + x2 + x3 = 20
constraint #2: –1.4x1 + x2 + x4 = 2
cost: 2x1 + x2 –Z = 0
0 ≤ x3 ≤ 0
0 ≤ x4 ≤ ∞
minimum: 2 2 0 0
present value: 2 2 16 2.8 6
maximum: 12 16 0 ∞
variable 3 is the worst-violated basis variable, R = 1
exceeds its maximum of 0
using column selection procedure, pivot at column 2, S = 2
s = 1: a11 > 0 x1 = x1min c1 > 0 then c1 / a11 = 2/1 = 2
s = 2: a12 > 0 x2 = x2min c2 > 0 then c2 / a12 = 1/1 = 1
LP example
pivoting results in:
constraint #1: x1 + x2 + x3 = 20
constraint #2: –2.4x1 – x3 + x4 = –18
cost: x1 – x3 – Z = –20
0 ≤ x3 ≤ 0
0 ≤ x4 ≤ ∞
minimum: 2 2 0 0
present value: 2 18 0 –13.2 22
maximum: 12 16 0 ∞
variable 4 is the worst-violated basis variable, R = 2
exceeds its minimum of 0
using column selection procedure, pivot at column 1, S = 1
s = 1: a21 < 0 x1 = x1min c1 > 0 then c1 / –a21 = 1/2.4 = 0.4166
s = 3: a22 < 0 x3 = x3max c2 < 0 then x3 is not eligible
LP example
pivoting results in:
constraint #1: + x2 +.583x3 +.417 x4 = 12.5
constraint #2: x1 +.417x3 +.417 x4 = 7.5
cost: –1.417x3 +.417 x4 – Z = –27.5
0 ≤ x3 ≤ 0
0 ≤ x4 ≤ ∞
minimum: 2 2 0 0
present value: 7.5 12.5 0 0 27.5
maximum: 12 16 0 ∞
no violations among the basis variables
algorithm stops at the optimum
Fuel scheduling by LP
two coal-burning generating units
both must remain online for a 3 week period
the combined output is to supply the following load
week 1: 1200 MW
week 2: 1500 MW
week 3: 800 MW
one coal supplier is under contract to supply 40,000 tons per
week to be split between the two plants
there are existing inventories at the start of the 3-week period
find the following
the operation schedule for each plant for each week
the coal delivery amounts to be made each week
data
coal
heat value: 23 Mbtu/ton
cost: $30/ton or $1.3/Mbtu
inventories: plant #1 = 70,000 tons, plant #2 = 70,000 tons
each plant has a maximum coal storage of 200,000 ton
generators
plant #1: H1(P1) = 380.0 + 8.267 P1 150 ≤ P1 ≤ 600 MW
q1(P1) = 16.52 + 0.3594 P1 [tons/h]
F1(P1) = 495.65 + 10.78 P1 [$/h]
plant #2: H2(P2) = 583.3 + 8.167 P2 400 ≤ P2 ≤ 1000 MW
q2(P2) = 25.36 + 0.3551 P2 [tons/h]
F2(P2) = 760.8 + 10.65 P2 [$/h]
solution process
assume that the units are operated at a constant rate during each
week
coal deliveries are made at the beginning of each week
the problem is set up for 1-week time intervals
then
plant #1: q1(P1) = 2775.4 + 60. 4 P1 [tons/wk]
F1(P1) = 83,269.2 + 1811 P1 [$/wk]
plant #2: q2(P2) = 4260.5 + 59.7 P2 [tons/wk]
F2(P2) = 127,814.4 + 1789 P2 [$/wk]
objective function
minimize: Z = F1[P1(1)] + F2[P2(1)] + F1[P1(2)] + F2[P2(2)]
+ F1[P1(3)] + F2[P2(3)]
constraints
power delivery
volume of coal
P1(1) + P2(1) = 1200
V1(1) + D1(1) – q1(1) = V1(2)
P1(2) + P2(2) = 1500
V2(1) + D2(1) – q2(1) = V2(2)
P1(3) + P2(3) = 800
V1(2) + D1(2) – q1(2) = V1(3)
coal deliveries V2(2) + D2(2) – q2(2) = V2(3)
D1(1) + D2(1) = 44,000
V1(3) + D1(3) – q1(3) = V1(4)
D1(2) + D2(2) = 44,000
V2(3) + D2(3) – q2(3) = V2(4)
D1(3) + D2(3) = 44,000
variables D1(1)
D2(1)
D1(2)
D2(2)
D1(3)
D2(3)
P1(1)
P2(1)
V1(2)
P1(2)
V2(2)
P2(2)
V1(3)
P1(3)
V2(3)
P2(3)
V1(4)
V2(4)
X1 X2 X3 X4 X5 X6 X7 X8 X9 X10 X11 X12 X13 X14 X15 X16 X17 X18 constants
constraint
1 1 1 1200
2 1 1 40000
3 1 -60.4 -1 2775.4-V1(1)
4 1 -59.7 -1 4260.5-V2(1)
5 1 1 1500
6 1 1 40000
7 1 1 -60.4 -1 2775.4
8 1 1 -59.7 -1 4260.5
9 1 1 800
10 1 1 40000
11 1 1 -60.4 -1 2775.4
12 1 1 -59.7 -1 4260.5
variable
0
0
600 150
1000 400
600 150
1000 400
600 150
1000 400
minimum
variable
40000
40000
200000
40000
200000
40000
200000
40000
200000
40000
200000
200000
maximum
EEL 6266
Chapter 8
Production Cost Models
Introduction
Production cost models
are computational models designed to calculate information
for long-range system planning:
generation system production costs
energy import requirements
availability of energy for sale to other systems
fuel consumption
employees models of expected load patterns and simulated
operation of the system’s generation
uncertainty of load forecasts
reliability of generating units
expected need for emergency energy and capacity supplies
uses statistical computational methods for solving problems
Introduction
Stochastic production cost models
used for long range studies
the risk of sudden, random, generating unit failures and
random deviations for the mean forecasted load are
treated as probability distributions
load modeling considers the behaviors of the “expected load”
patterns that cover periods of weeks, months, and/or years
the load duration cover expresses the time period that the
loading is expected to equal or exceed a given power value
generating unit modeling includes fuel costs
usually expressed over a monthly basis
generating unit scheduled maintenance outages
may involve time periods from days to years
Introduction
Types of production cost studies
Interval under Economic Dispatch Long-Range Operation Weekly
Load Model
Consideration Procedure Planning Planning Schedules
total energy or seasons block loading
load duration or years (w/o regard to incr. costs)
load duration months
or load cycles or weeks
incremental loading
load duration months, weeks, incremental loading
or load cycles or days with forced outages
incremental loading
load cycles weeks or days
with losses
Load-duration Curves
Representation of future loads
Load (MW)
in which the impact of capacity
limitations will be studied
Building the load-duration curve Time (h)
that load = L MW
p(L) Probability
consider the expected load probability density
pattern function
build a histogram of load for

a given time period and find
the load density function, p(x) Load L (MW)
P(L) Probability
cumulative
integrate the load density
load ≥ L MW
1.0
distribution
function to obtain the load function
distribution function, Pn(x)

∞
Pn ( x ) = 1 − ∫ p ( x ) dx 0.0
x Load L (MW)
Building the load duration curve
multiply the probability by the
period length to show the number 1500
of hours that load equals or
Load L (MW)
exceeds a given power level, L 1000
common convention has the

load on the vertical axis
500
Block-Loading
simulates the economic dispatch
0
procedure with this type of load model 0 4 8 12
Hours load equals
generating units are ordered by cost or exceed L MW
units are assumed to be fully loaded or
loaded up to the limitation of the load-duration curve
Block-Loading
example of the Niagara-Mohawk system
gas turbines (280 MW out of 400 MW)
Incremental Maximum
Unit
Cost ($/h) Capacity (MW) 1500
Rio Bravo #1, #2, & #3 (120 MW)
2-Mile Point 8.1 800
Mohawk #1 8.5 300 Mohawk #2 (200 MW)
Load L (MW)
Mohawk #2 8.7 200 1000
Mohawk #1 (300 MW)
Rio Bravo #1 9.2 75
Rio Bravo #2 9.6 25
Rio Bravo #3 9.7 20 500
(8) gas 2-Mile Point (800 MW)
9.9 400
turbines
0
Hours load equals or exceed L MW
Example x-Load Duration Energy
(MW) (h) (MWh)
consider a two generating unit
100 20 2000
system that will serve the 80 60 4800
following expected load pattern: 40 20 800
construct a load-duration curve Totals: 100 7600
in tabular and graphic form
T Pn(x), Hours that load equals or exceeds x

T Pn(k), Hours 100
x-Load Exact
that Load Equals
(MW) Duration (h)
or Exceeds x
0 0 100
20 0 100 50
40 20 100
60 0 80
80 60 80
100 20 20 0
0 20 40 60 80 100
100+ 0 0 x-Load (MW)
Example
the two generating units have the following characteristics
Power Fuel Fuel Fuel Cost Incremental Unit Force
Unit Output Input Cost Rate Fuel Cost Outage Rate
(MW) (MBtu/h) ($/MBtu) ($/h) ($/MWh) (per unit)
0 160 1.0 160
1 8.0 0.05
80 800 1.0 800
0 80 2.0 160
2 16.0 0.10
40 400 2.0 800
the fuel cost rate for each unit is a linear function of the power
output
block-load the two units onto the load-duration curve
unit #1 is used first because of its lower average cost per MWh
T Pn(x), Hours that load equals or exceeds x

Example 100
block-loaded
unit #1 is on-line for 100 h
80 MW output for 80 h 50
40 MW output for 20 h Unit 1
unit #2 is on-line for 20 h

Unit 2
20 MW output for 20 h 0
0 20 40 60 80 100
summary of results x-Load (MW)
Load Duration Energy Fuel Used Fuel Cost
Unit
(MW) (h) (MWh) (MBtu) ($)
40 20 800 9600 9600
1 80 80 6400 64000 64000
Subtotal: 7200 73600 73600
20 20 400 4800 9600
2
Total: 7600 78400 83200
Forced Outages
Forced outage of a generator unit
the time that the unit is not available due
to a failure of some sort
represents a random event
taken out of the total time that the unit
should be available for service
the forced outage rate is the ratio of forced outage
time over the total time available
schedule outage times for maintenance are excluded in
both the total time available and the forced outage time
Forced outage rates for all generating units must be
accounted for in the expected production costs
Forced Outages
Example
reconsider the previous example, but now including the effects
of forced outages
evaluate by load levels
Load = 40 MW; duration 20 h
Unit 1: on-line for 20 h, operates for 0.95 × 20 = 19 h
output: 40 MW, energy delivered: 19 × 40 = 760 MWh
Unit 2: on-line for 1 h, operates for 0.90 × 1 = 0.9 h
output: 40 MW, energy delivered: 0.9 × 40 = 36 MWh
load energy = 800 MWh
generation = 796 MWh
unserved energy = 4 MWh
shortage = 40 MW for 0.1 h
Forced Outages
Example
Unit 2: on-line for 3 h, operates for 0.90 × 3 = 2.7 h
load energy = 4800 MWh
generation = 4668 MWh
shortage = 80 MW for 0.3 h (24 MWh) and
40 MW for 2.7 h (108 MWh)
Forced Outages
Example
Unit 2: on-line for 20 h, operates as follows
• Unit 1 is on-line and operating for 19 h
Unit 2: on-line for 0.90 × 19 = 17.1 h
shortage: 20 MW for 1.9 h
• Unit 1 is supposedly on-line, but not operating 1 h
Unit 2: on-line for 0.90 × 1 = 0.9 h
shortage: 100 MW for 0.1 h and 60 MW for 0.9 h
load energy = 2000 MWh; generation = 1898 MWh
• 100 MW for 0.1 h = 10 MWh; 60 MW for 0.9 h = 54 MWh;
20 MW for 1.9 h = 38 MWh
Forced Outages
Comments
it was necessary to make an arbitrary assumption that the
second unit will be on-line for any load level that equals or
exceeds the capacity of the first unit
the enumeration of the possible states is not complete
need to separate the periods when there is excess capability,
exact matching of generation and load, and shortages
when there is an exact matching of generation and load, it is
referred to as a “zero-MW shortage”
there are two such periods in the example
40 MW loading, 20 h duration, unit 2 on: 0.05 × 0.9 × 20 = 0.9 h
80 MW loading, 60 h duration, unit 1 on: 0.95 × 0.1 × 60 = 5.7 h
total zero-reserve expected duration: 6.6 h
Forced Outages
Summary of all possible states
Load Duration Event Unit 1 Unit 2 Combined Event

(MW) (h) No.
Status Power (MW) Status Power (MW) Duration (h) Consequence
1 1 40 1 0 17.1 Load satisfied
40 20
3 0 0 1 40 0.9 0 MW shortage
4 0 0 0 0 0.1 40 MW shortage
6 1 80 0 0 5.7 0 MW shortage
80 60
7 0 0 1 40 2.7 40 MW shortage
8 0 0 0 0 0.3 80 MW shortage
10 1 80 0 0 1.9 20 MW shortage
100 20
11 0 0 1 40 0.9 60 MW shortage
12 0 0 0 0 0.1 100 MW shortage
Forced Outages
Summary of generation cost results
Scheduled Expected Expected Expected Expected
Unit Time Operating Generated Fuel Used Production
On-line (h) Time (h) Energy (MWh) (MBtu) Cost ($)
1 100 95.0 6840 69920 69920
2 81 72.9 522 10008 20016
Total: 7362 79928 89936
unserved load
Unserved Duration of Unserved Duration of Given
Demand (MW) Shortage (h) Energy (MWh) Shortage or More (h)
0 6.6 0 12.6
20 1.9 38 6.0
40 2.8 112 4.1
60 0.9 54 1.3
80 0.3 24 0.4
100 0.1 10 0.1
Total: 12.6 238
EEL 6266
Chapter 8
Production Cost Models
Probabilistic Programming
Production cost analysis
operating experience indicates that the force outage rate of
thermal-generating units tends to increase with unit size
the frequent long-duration outages of the more efficient base-load
units require running the less efficient, more expensive plants at
higher than expected output and the importation of emergency
energy
two measures of system unreliability due to random forced
generator failures:
time period when the load is greater than available generation
capacity
the expected levels of power and energy that must be imported to
satisfy the load demand
sensitivity indicators of the need to add generator or tie-line capacity
Probabilistic production cost computations
mathematical methods that make use of probabilistic models
load models
energy and capacity models for generators
generator models represent the unavailability of basic energy
resources, random forced outages, and the effects of energy
sale/purchase contracts
tie-line models represent the expected cost of emergency energy,
also called the cost of unsupplied energy
techniques for the convolution of the load distributions with
the capacity-probability density functions
numerical convolution of discrete functions
analytical convolution of continuous functions
the unserved load distribution method
the individual unit probability capacity density functions are
convolved one by one with the load distribution function
each convolution result is an unserved load distribution, which will
be used for the next convolution
the overall result is a series of unserved load distributions
the sequence of convolutions is determined by a fixed economic
loading criterion that sets the order of the generator units
the area under the load distributions represent needed energy
a unit’s energy production is found from the difference in the
energy from the unserved load distribution prior to convolution
and the energy from the new unserved load distribution after the
convolution
the expected cost method
the individual unit probability capacity density functions are
convolved with one another
the sequence of convolutions is determined by a fixed economic
criterion that sets the order of the generator units
the convolutions of the generating units produces a set of available
capacity distributions, each with an associated cost curve as a
function of total power generated
the resultant expected cost curve is convolved with the load
distribution function
this produces the expected value of production cost to serve the
given load forecast
The unserved load method
load is modeled as a load-duration curve
a probability distribution that is expressed in hours that the load
equals or exceeds a given power value
monotonically decreasing function with increasing load
can be treated as cumulative probability density function, Pn(x) or
can be expressed in hours, T Pn(x)
for numerical analysis
treat it in terms of regular discrete steps in a recursive algorithm
procedural steps
if there is a segment of capacity with a total of C MW available
for scheduling
let q be the probability that C MW are unavailable
and p = 1 – q (that is the available probability of the segment)
procedural steps
then after the segment is scheduled
the probability of needing x MW or more is now P′n(x)
because the occurrence of loads and unexpected unit outages are
statistically independent events, the new probability distribution
function is a combination of the two mutually exclusive events
Pn′ ( x ) = q Pn ( x ) + p Pn ( x + C )
• q Pn(x) is the probability that new capacity C MW is unavailable and
needing x MW or more
• p Pn(x + C) is the probability that C is available and needing (x + C)
MW or more
procedural steps
first, the generation requirements for any generating segment are
determined by the knowledge of the distribution T Pn(x) that
exists prior to the dispatch of the particular generating segment
the value of T Pn(0) determines the required hours of operation of a
new unit, and the area under the distribution T Pn(x) between zero
and the rating of the unit determines the energy production
if a generation segment is not perfectly reliable, there will be a
residual distribution of demands that cannot be served by this
particular segment because of forced outages
procedural steps
representing the forced outage when the unit is needed for
T Pn(0) hours
on average it is only available for (1 – q) T Pn(0) hours
energy required by the load distribution that the unit could serve
C
E = T ∫ Pn ( x ) dx = T ∑ Pn ( x ) ∆x
C
0
x =0+
the unit can only generate (1 – q) E because of its expected
unavailability
suppose that the unit has a linear input-output cost function
Fi (Pi ) = F0 + F1 ⋅ Pi
the expected production cost for this period
E [Fi ] = F0 (1 − q )T Pn (0 ) + F1 (1 − q )E
procedural steps
there is a residual of unserved demands due to the forced outages
of the unit
the new distribution of the probabilities of unserved load
T Pn′ ( x ) = qT Pn (x ) + (1 − q )T Pn (x + C )
the process may be repeated until all units have been scheduled
a residual distribution remains that gives the final distribution of
unserved demand
Example
use the data of the two generating unit system and the
associated expected load pattern
Power Fuel Cost Incremental Unit Force
Fuel Input Fuel Cost
Unit Output Rate Fuel Cost Outage Rate
(MBtu/h) ($/MBtu)
(MW) ($/h) ($/MWh) (per unit)
0 160 1.0 160
1 8.0 0.05
80 800 1.0 800
0 80 2.0 160
2 16.0 0.10
40 400 2.0 800
T Pn(k), Hours that
x-Load Exact
first analyze the expected (MW) Duration (h)
Load Equals or
Exceeds x
value of production cost 0 0 100
20 0 100
to serve the given load 40 20 100
forecast by ignoring the 60 0 80
80 60 80
forced outages 100 20 20
100+ 0 0
Example
graphical form of the convolution
T Pn(x) (h) T Pn(x) (h)
100 100
original load duration load duration curve

50 50
(distribution) curve after first unit dispatch
-100 -50 0 50 100 -100 -50 0 50 100

x-Unsupplied Load (MW) T Pn(x) (h) x-Unsupplied Load (MW)
100
load duration curve 50

after second unit dispatch
-100 -50 0 50 100

x-Unsupplied Load (MW)
Example
tabular form of the convolution
T P′n(x) T P″n(x)
x-Load T Pn(x)
= T Pn(x + 80) = T P′n(x + 40)
(MW) (h)
(h) (h)
0 100 80 0
20 100 20 0
40 100 0 0
60 80 0 0
80 80 0 0
100 20 0 0
100+ 0 0 0
Energy (for x ≥ 0) 7600 MWh 400 MWh 0 MWh
shows the load probability for unserved loads of 0 to 100+ MW
Example
repeat, now considering the effects of the forced outage rates
for the first dispatched unit, q1 = 0.05
the recursive equation to obtain P′n(x)
T Pn′ ( x ) = 0.05 T Pn ( x ) + 0.95 T Pn (x + 80)

the original and resultant unserved load distributions
x-Load T Pn(x) T Pn(x + 80) T P′n(x) T Pn(x) (h)
(MW) (h) (h) (h)
100
0 100 80 76 + 5 = 81
20 100 20 19 + 5 = 24
40 100 0 0+5= 5 load duration curve
50
60 80 0 0+4= 4 after first unit dispatch
80 80 0 0+4= 4
100 20 0 0+1= 1
100+ 0 0 0 -100 -50 0 50 100
Energy (for x ≥ 0) 7600 MWh 760 MWh x-Unsupplied Load (MW)
Example
the production cost for unit 1
may be loaded to 80 MW for 80 h and 40 MW for 20 h (max)
but the unit is only available for 19 h at 40 MW and for 76 h at
80 MW
cost:
40 MW, 19 h, E = 760 MWh, fuel = 9120 MBtu, cost = $ 9120
80 MW, 76 h, E = 6080 MWh, fuel = 60.8 Gbtu, cost = $ 60800
Example
for the second dispatched unit, q2 = 0.10
the recursive equation to obtain P′n(x)
T Pn′′(′ x ) = 0.10 T Pn′ ( x ) + 0.90 T Pn′ ( x + 40)
the first resultant and second resultant unserved load distributions
load duration curve
x-Load T P′n(x) T P″n(x) after second unit dispatch
(MW) (h) (h)
T Pn(x) (h)
0 81 12.6 100
20 24 6.0
40 5 4.1
60 4 1.3
50
80 4 0.4
100 1 0.1
100+ 0 0
Energy (for x ≥ 0) 760 MWh 238 MWh -100 -50 0 50 100
x-Unsupplied Load (MW)
Example
the production cost for unit 2
the unit is required a total of 81 h
at 0 MW for 81 - 24 = 57 h
at 40 MW for 5 h
at 20 MW for 24 - 5 = 19 h
cost:
0 MW, 51.3 h, E = 0 MWh, fuel = 4104 MBtu, cost = $ 8208
total: 72.9 h, E = 522 MWh, fuel = 10008 MBtu, cost = $20016
Example
final unserved energy distribution
there remains an expected requirement to supply 100 MW
probability for needing this capacity is 0.1% (0.1 h out of 100 h)
the cost of supplying the remaining 238 MWh of unsupplied
load energy
based on an estimate of the cost of emergency energy supply
assume that emergency energy cost is $ 24 / MWh
the resulting cost is then:
238 MWh × $ 24 / MWh = $ 5712
difference between the two examples
ignoring force outages results in a 1.95 % underestimation of fuel
consumption and a complete neglect of emergency supply costs,
and an 8.1 % underestimate of the total production costs
EEL 6266
Chapter 9
Control of Generation
Generation Control
Optimal dispatch and scheduling of generation establishes
the best operation point with respect to economics
The operating point must be implemented via generation
control
local generator control for each individual generator
energy control center for the control of a large utility and the
flow of power across interconnections to other utilities
regional control over several utilities and the Independent
Power Produces, IPP’s
ISO - Independent System Operator
RTO - Regional Transmission System Operator
Overview of Control Problem
Many generators supply power turbine-generator unit
to the transmission system

consumer loads are constantly turbine-generator unit
changing the power level

control
some control means needed signals measurements of
to allocate the load changes generator electric
power output
to the generators power
measurement of system
a governor on each unit generation system frequency
maintains mechanical speed control
measurements of
system
(electrical frequency) tie-line flows to
neighboring systems
supplementary control acts
to allocate generation
influences the power output
control signal usually originates at a remote control center
Generator Model
Definition of important terms
ω = rotational speed
mechanical turbine Tmech Telec generator electrical
α = rotational acceleration energy energy
δ = phase angle of a rotating machine
Tnet = net accelerating torque in a machine
Tmech = mechanical torque exerted on the machine by the turbine
Telec = electrical torque exerted on the machine by the generator
Pnet = net accelerating power
Pmech = mechanical power input
Pelec = electrical power output
I = moment of inertia for the machine
M = angular momentum of the machine
• all quantities, except for the phase angle, are expressed in per unit on the
machine base and/or the standard system frequency base
• steady-state and nominal values have a “0” subscript added
Generator Model
Basic relationships
acceleration principle: Tnet = I α
momentum principle: M =ω I
power equation: Pnet = ω Tnet = ω (I α ) = M α
Phase angle deviation
general shaft equation: ω = ω0 + α t δ = δ 0 + ω0 t + 12 α t 2
deviation from nominal: dtd (∆δ ) = ∆ω = α t
relationship to torque: Tnet = Tmech − Telec = I α = I ddt (∆δ )
2
2
deviation of power: Pnet = Pmech − Pelec = ω Tnet = ω I d2

(∆δ )
dt 2
the resulting swing eq.: Pmech − Pelec = M d2

dt 2
(∆δ ) = M ddt (∆ω )
Generator Model
Laplace transform of the dynamic power equation
Tmech 0 = Telec 0
Tnet = Tmech 0 − Telec 0 + ∆Tmech − ∆Telec = ∆Tmech − ∆Telec
Pnet = ω Tnet = ω ∆Tmech − ω ∆Telec = ∆Pmech − ∆Pelec
∆Pmech − ∆Pelec = M (∆ω ) d
dt
∆Pmech − ∆Pelec = M s (∆ω )
Block diagram model

1
∆Pmech ∆ω
M s
∆Pelec
Load Model
Electrical loads consist of a variety of devices
purely resistive devices
power electronics
motor loads
motor loads dominate the mix of loads
Motors exhibit a variable power-frequency characteristic
model of the effect of a frequency change on net load drawn
∆PL ( freq ) = D ⋅ ∆ω
D is expressed as a percentage change in load per percentage
change in frequency on the motor’s power base
the value of D must be converted to the system power base for
system studies
Load Model
Block diagram modeling
basic frequency dependent load
∆ω D ∆PL(freq)
rotating mass and load as seen by prime mover output

1
∆Pmech ∆ω
M s
1
∆Pmech ∆ω
M s+D
∆Pload D
∆Pload
the net change in the electrical power load, Pelec, is

∆Pelec = ∆PL + D ∆ω
where ∆PL is the non-frequency-sensitive load change
Load Model
Example
consider an isolated power system on a 1000 MVA base
600 MVA generator
M = 7.6 pu MW/pu frequency/sec on the machine base
400 MVA load
the load changes by 2% for a 1% change in frequency
suppose that the load increases by 10 MVA
what is the transient response of the system frequency
M = 7.6 (600/1000) = 4.56 pu and D = 2.0 (400/1000) = 0.80 pu
0.01  
∆PL (s ) = → ∆ω (s ) =
0.01 1
 
s s  4.56s + 0.8 
− 0 .8 t
∆ω (t ) =
0.01 4.56 0.01
e − = 0.0125e −0.175t − 0.0125
0.8 0.8
Load Model
Example
the final value of ∆ω is -0.0125 pu
a drop of 0.75 Hz on a 60-Hz system
Prime-mover Model
The prime mover drives the generating unit rotating
shaft
steam turbine steam valve
hydroturbine steam prime mover
Modeling must account for ∆Pvalve

control system characteristics 1
∆Pvalve ∆Pmech
e.g., boiler and steam supply 1 + s TCH
Model of the non-reheat turbine Prime-mover model
relates the steam valve position
to the output power ∆Pmech
1 1
“charging time” ∆Pvalve ∆ω
1 + s TCH M s+D
time constant, TCH
per unit change in valve ∆Pload
position from nominal, ∆Pvalve Prime-mover-generator-load model
Governor Model
The governor compensates for changes in the shaft speed
changes in load will eventually lead to a change in shaft speed
change in shaft speed is also seen as a change in system
frequency
simplest type of control is the isochronous governor
rotating
steam valve
shaft
steam prime mover tachometer
∆Pvalve ω
+ = open valve
– = close valve ∆Pvalve ∆ω
∫ dt KG –1 ωref
Governor Model
The isochronous governor
to force frequency errors to zero requires the use of an
integration of the speed error
the isochronous governor can not be used when two or more
generators are electrically connected to the same system
fighting between generator governors for system frequency
problems with load distribution between generators
A load reference control provides settings for both the
frequency and the desired output power
a new input, the load reference signal, controls the desired
power output
feedback loop contains a gain R that determines a speed-droop
characteristic
Governor Model
Load reference control
the speed-droop function handles the load sharing between
generators
there will always be a unique frequency at which the system
loading will be shared among the generators
the gain R is equivalent to the per unit change
∆ω

in frequency for a 1.0 p.u. change in power: R = pu
rotating ∆P
steam valve
shaft
steam prime mover tachometer
∆Pvalve ω
+ = open valve
∆Pvalve ∆ω
– = close valve
∫ dt KG ωref
Pref R
Governor Model
Simplification of the block diagram model
_ ∆ω
_
Kg
ωref Σ Σ ∆Pvalve
s +
+ _
ω
R Σ _ Pref
_ ∆ω
_ 1
1
ωref Σ R
Σ 1 + s (K g R )
−1 ∆Pvalve
+ +
ω Pref
_ ∆ω
_ 1
1
ωref Σ R
Σ 1 + s Tg
∆Pvalve
+ + 1
ω where Tg =
Pref Kg R
Governor Characteristics
Speed-droop characteristic
frequency
f0
p.u. output power

0.5 1.0
Allocation of unit outputs with governor droop
frequency
f0
f′
P1 P1′ P2 P2′
Unit #1 output power Unit #2 output power
Governor Characteristics
Speed-changer settings
frequency
f0
p.u. output power

0.5 1.0
load reference for nominal load reference for nominal

speed at no-load speed at 50% loading
load reference for nominal
speed at 50% loading
Complete Generator Model
Block diagram of governor, prime mover, rotating mass, and
loads 1 ∆ω
R
rotating mass
_ governor prime mover & load
+ 1 ∆Pvalve 1 ∆Pmech + 1
Pref Σ 1 + s TG 1 + s TCH Σ M s+D
load reference _
set point
loading input ∆PL
transfer function of generator
 −1 
 
∆ω ( s )  Ms+D 
=
∆PL ( s )  1  1  1  1 
1 +     
 R  1 + s TG  1 + s TCH  M s + D  
Complete Generator Model
Steady state behaviors
final value of the transfer function
using Laplace method
 −1 
  − ∆P
∆ω t = ∞ = lim [s ∆ω ( s )] = ∆PL  D  = −1 L
s →0
1 +  1  1   R + D
  R  D  
for several generators connected within the system

− ∆PL
∆ω =
1 + 1 +L+ D
RG1 RG 2
EEL 6266
Chapter 9
Tie-line Model
The power flow across a tie-line can be modeled using a
linear load flow approach
steady-state or nominal flow quantity: Ptie flow =
1
(δ 1 − δ 2 )
X tie
deviation from the nominal tie-line flow
Ptie flow + ∆Ptie flow =
1
[(δ1 + ∆δ1 ) − (δ 2 + ∆δ 2 )] = 1 (δ1 − δ 2 + (∆δ1 − ∆δ 2 ))
X tie X tie
∆Ptie flow =
1
(∆δ1 − ∆δ 2 )
X tie
where ∆δ must be in radians for ∆Ptie to be in per unit
ω0
using the relationship for speed and ∆δ: ∆δ = ⋅ ∆ω
s
ω0 (∆ω1 − ∆ω 2 )
then ∆Ptie flow = =
T
(∆ω1 − ∆ω 2 )
s X tie s
where T = 377 / Xtie for a 60-Hz system
Tie-line Model
Simplified control for two interconnected areas
1 ∆ω1
R1
_ ∆PL1 _
+ 1 1 ∆Pmech1 + 1
Pref-1 Σ 1 + s TG1 1 + s TCH 1 Σ M1s + D1
_ +
load reference ∆Ptie T
set points
governors prime movers
s
Σ
_
+
+ 1 1 ∆Pmech2 + 1
Pref-2 Σ 1 + s TG 2 1 + s TCH 2 Σ M 2 s + D2
_ _
∆PL2
1 ∆ω2
R2
Tie-line Model
Consider two areas each with a generator
the two areas are connected with a single transmission line
the line flow appears as a load in one area and an equal but
negative load in the other area
the flow is dictated by the relative phase angle across the line,
which is determined by the relative speeds deviations
let there be a load change ∆PL1 in area 1
to analyze the steady-state frequency deviation, the tie-flow
deviation and generator outputs must be examined
Tie-line Model
after the transients have decayed, the frequency will be
constant and equal to the same value in both areas
d (∆ω1 ) d (∆ω 2 )
∆ω1 = ∆ω 2 = ∆ω → = =0
dt dt
− ∆ω
∆Pmech−1 − ∆Ptie − ∆PL1 = ∆ω ⋅ D1 ∆Pmech −1 =
R1
− ∆ω
∆Pmech −2 + ∆Ptie = ∆ω ⋅ D2 ∆Pmech −2 =
R2
applying substitutions yields
 1
− ∆Ptie − ∆PL1 = ∆ω  D1 + 
 R1 
 1 
+ ∆Ptie = ∆ω  D2 + 
 R2 
Tie-line Model
additional simplification yields
− ∆PL1
∆ω =
1 1
+ + D1 + D2
R1 R2
 1 
− ∆PL1  D2 + 
 R2 
∆Ptie =
1 1
+ + D1 + D2
R1 R2
Notes
the new tie flow is determined by the net change in load and
generation in each area, but not influenced by the tie stiffness
the tie stiffness determines the phase angle across the tie
Tie-line Model
Example
consider two areas, each with a generator, motor loads, and a
single tie-line connecting the two areas.
area 1: R1 = 0.01 pu, D1 = 0.8 pu, Base MVA = 500
area 2: R2 = 0.02 pu, D2 = 1.0 pu, Base MVA = 500
a load change of 100 MW (0.2 pu) occurs in area 1
find the new steady-state frequency and net tie flow change
− ∆PL1 − 0.2
∆ω = = = −0.001318
1 1 1 1
+ + D1 + D2 + + 0.8 + 1.0
R1 R2 0.01 0.02
f new = 60 + (− 0.001318)(60) = 59.92 Hz
 1   1 
∆Ptie = ∆ω  + D2  = (− 0.001318) + 1 = 0.06719 pu = 33.6 MW
 R2   0.02 
Tie-line Model
Example
change in the prime movers
∆Pmech1 = − ∆ω = 0.001318 = 0.1318 (65.88 MW)
R1 0.01
∆Pmech 2 = − ∆ω = 0.001318 = 0.0659 (32.94 MW)
R2 0.02
change in the motor loads
∆PL1( motor ) = ∆ω D1 = −0.001318 ⋅ 0.8 = −0.001054 ( −0.527 MW)
∆PL 2 ( motor ) = ∆ω D2 = −0.001318 ⋅ 1.0 = −0.001318 ( −0.659 MW)
change in tie flow
∆Ptie = ∆Pmech 2 − ∆PL 2 ( motor ) = 0.0672 (33.6 MW)
change in apparent loading
∆PL1 = ∆Pmech1 − ∆PL1( motor ) + ∆Ptie = 0.200 (100 MW)
Generation Control
Automatic Generation Control (AGC) is a control system
that has three major objectives
hold the system frequency at or very close to a specified
nominal value (e.g., 60 Hz)
maintain the correct value of interchange power between
control areas
enforce contracts for shipping or receiving power along the tie-
lines to neighboring utilities
maintain each generating unit’s operating point at the most
economic value
Generation Control
Supplementary control action
assume for the moment that a single generating unit supplies
load to an isolated power system
a load change will produce a frequency change
magnitude depends on the droop characteristics of the governor and
the frequency characteristics of the system load
a supplementary control must act to restore the frequency to the
nominal value (60 Hz)
accomplished through an integral control loop to the governor
the control action forces the frequency error to zero by adjusting the
speed reference set point
Generation Control
Supplementary control action
speed (frequency)
droop control
1 ∆ω
R1
∆PL rotating masses
_ _ and frequency-
governor prime mover dependent loads
+ 1 1 ∆Pmech1 + 1
Pref Σ 1 + s TG 1 + s TCH Σ M s+D
generator
loading
frequency
set point
restoration control
K ∆ω
s
Generation Control
Tie-line control
two utilities will interconnect their systems for several reasons
buy and sell power with neighboring systems whose operating
costs make the transactions profitable
improvement to overall reliability for events like the sudden loss
of a generating unit
provide a common frequency reference for frequency restoration
define tie flows and tie flow changes
total actual net interchange: Pnet int
+ for power leaving the system; – for power entering the system
scheduled or desired value of interchange: Pnet int sched
change in tie flow: ∆ Pnet int = Pnet int – Pnet int sched
Generation Control
Tie-line control
interconnections present a challenging control problem
consider a two-area system
both areas have equal load and generator characteristics
(R1 = R2, D1 = D2)
assume that area 1 sends 100 MW to area 2 under an interchange
agreement between the system operators of the two areas
let area 2 experience a sudden load increase of 30 MW, then both
areas see a 15 MW increase in generation (because R1 = R2) and
the tie flow increases from 100 to 115 MW
• the 30 MW load increase is satisfied by a 15 MW increase in
generation #2 plus a 15 MW increase in tie flow into area 2
this is fine, except that area 1 contracted to sell only 100 MW
• generation costs have increased without anyone to bill
a control scheme is needed to hold the system to the contract
Generation Control
Tie-line control
such a control system must use two pieces of information
the system frequency
the net power flow across the tie line
a few possible network conditions include
if the frequency decreases and net interchange power leaving the
system increases, a load increase has occurred outside the area
if the frequency decreases and net interchange power leaving the
system decreases, a load increase has occurred inside the area
define a control area
part of an interconnected system within which the load and
generation will be controlled
all tie-lines that cross a control area boundary must be metered
Generation Control
Tie-line control
control scheme actions
∆ω ∆Pnet int ∆PL1 ∆PL2 Resulting control action
– – + 0 increase Pgen in area 1
+ + – 0 decrease Pgen in area 1
– + 0 + increase Pgen in area 2
+ – 0 – decrease Pgen in area 2
∆Pnet int
1 2
∆PL1 = Load change in area 1

∆PL2 = Load change in area 2
Generation Control
Tie-line control
for the first row of the control response table, it is required that
∆Pgen1 = ∆PL1
∆Pgen 2 = 0
the required change is known as the area control error (ACE)
the equations for the ACE for each area
ACE1 = − ∆Pnet int 1 − B1 ∆ω
ACE2 = − ∆Pnet int 2 − B2 ∆ω
B1 and B2 are the frequency bias factors, and are set accordingly
B1 = 1 + D1
R1
B2 = 1 + D2
R2
Generation Control
Tie-line control
combining the ACE functions with the tie flow equations
results in
 1 
∆PL1  + D2 
 R2  −  1 + D  − ∆PL1
ACE1 =  1 = ∆PL1
1 1  1 + 1 + D1 + D2
+ + D1 + D2  R1
R1 R2 R1 R2
 1 
∆PL1  + D2 
 R2   1  − ∆PL1
ACE2 = −  + D2  =0
1 1 1 1
+ + D1 + D2  R2  + + D1 + D2
R1 R2 R1 R2
Generation Control
Tie-line bias supplementary control for two areas
B1
1 ∆ω1
_ _ R1
ACE1 ∆PL1 _
K + 1 1 + 1
Σ s
Σ 1 + s TG1 1 + s TCH 1 Σ M1s + D1
_ _
∆Pmech1 +
∆Pnet int 1 ∆Ptie T
–∆Pnet int 2 s
Σ
+ ∆Pmech2 _
+
K + 1 1 + 1
Σ s
Σ 1 + s TG 2 1 + s TCH 2 Σ M 2 s + D2
_ ACE2 _ _
∆PL2
governors prime movers 1 ∆ω2
R2
B2
EEL 6266
Chapter 9
Generator Allocation
A typical control area contains many generators
the individual outputs must be set according to economics
the solution of the economic dispatch must be coupled to the
generation control system
the input consist of the total generation required for the area
in order to satisfy the load demand and maintain contractual power
flows across the tie lines
the output is the power distribution across the outputs of all the
generators within the control area
continuously varying system load demand
a particular total generation value will not exist for a very long
time
Economic generator control
it is impossible to simply specify a total generation, calculate
the economic dispatch schedule, and give the control system
the output schedule for each generator
unless such a calculation can be made very quickly
for digital control system it is desirable to perform the economic
dispatch calculation at intervals of 1 to 15 minutes
independent of the calculation schedule
the allocation of generation must be made instantly whenever
the required area total generation changes
the allocation control of generation must run continuously
a rule must be provided to indicate the generation allocation for
values of total generation other than that used in the economic
dispatch
The allocation of individual generators over a range of total
generation values
accomplished using base points and participation factors
for period k, the economic dispatch sets the base-point
generation values for the total generation value measured
at the start of the period
the base-point generation for the ith unit, Pi base is the most
economic output for the particular total generation value
the participation factor, pfi, sets the rate of change of the ith
unit’s power output with respect to a change in total generation
the base points and participation factors are used as follows
Pi scheluded (t ) = Pi base (k ) + pf i ⋅ ∆Ptotal (t )
∆Ptotal (t ) = Pactual (t ) − ∑ P (k ) i base
i∈all gen
Base points and participation factors
participation factors are determined from a generator’s cost
function
dFi/dPi = Fi′
assume that both the first and second
derivatives exist for the cost function
the change in the system’s incremental
cost as a function of the change in ∆λ
power output on the ith generator λ0
( )
∆λi = ∆λsystem ≅ Fi′′ Pi 0 ⋅ ∆Pi
∆Pi
the change in system incremental cost
P i0 Pi
equaling the unit’s incremental cost is
Relationship of ∆λ and ∆Pi
true for all generating units
∆λ ∆λ ∆λ
= ∆P1 , = ∆P2 , L = ∆PN
F1′′ F2′′ FN′′
Base points and participation factors
the total change in generation must equal the change in the
total system demand, and is the sum of all the individual unit
changes
∆PD = ∆P1 + ∆P2 + L + ∆PN
1
= ∆λ ⋅ ∑
i∈all gen Fi ′′
the participation factor for each generating unit is then found

as 1
∆Pi Fi′′
pf i = =
∆PD 1
∑
i∈all gen Fi ′′
Example
Consider a three generator system
the cost functions for the three generators
F1 (P1 ) = 561 + 7.92 P1 + 0.001562 P12
F2 (P2 ) = 310 + 7.85P2 + 0.00194 P22
F3 (P3 ) = 78 + 7.97 P3 + 0.00482 P32
an economic dispatch has been conducted for a
total load demand of 850 MW
the system’s incremental cost is $ 9.148 / MWh
the dispatch is: P1 = 393.2 MW, P2 = 334.6 MW, &
P3 = 122.2 MW
calculate the participation factors for the current dispatch,
and calculate the dispatch for a new total load of 900 MW
Example
participation factors
pf1 =
∆P1
=
(0.003124)
=
320.10
−1
= 0.47
∆PD (0.003124) + (0.00388) + (0.00964)
−1 −1 −1
681.57
∆P2 (0.00388)
−1
pf 2 = = = 0.38
∆PD 681.57
∆P (0.00964)
−1
pf 3 = 3 = = 0.15
∆PD 681.57
new dispatch
∆PD = 900 − 850 = 50
P1 = P1 base + pf1 ⋅ ∆PD = 393.2 + (0.47 )(50) = 416.7
P2 = 334.6 + (0.38)(50) = 353.6
P3 = 122.2 + (0.15)(50) = 129.7
Generator Control
Automatic generator control implementation
the AGC schemes are usually centrally located at a control
center
system measurements, taken at the major substations, other
information and data are telemetered to the control center
unit megawatt power output for each committed generating unit
megawatt power flow over each tie line to neighboring systems
system frequency
control actions are determined in a digital computer
control signals are transmitted to the generation units at remote
generation stations over the same communication channels
raise / lower pulse signals change a generating unit’s load
reference point up or down
Generator Control
the basic reset control loop for a generating unit
consists of an integrator with gain K
the integrator insures that the steady-state
control error goes to zero
the scheduled power value is the control input
a function of the system frequency deviation,
net interchange error, and the unit’s deviation
from its scheduled economic output
Generator Control
the basic generating unit’s power output control loop
+ K 1 1 ∆Pmech
Pi sch Σ load ref 1 + s TGi 1 + s TCHi output
s
_ set point
governor and prime mover
implementation via telemetry

raise/lower request raise/lower request
control control
governor
∆Pmech
Pi sch logic logic
and
telemetry telemetry prime mover output
master remote
station station
at control center at generating plant
Generator Control
The AGC calculation
the input to the AGC combines the inputs of the various tie-
flows errors with the frequency deviation
the resultang is the area control error
fmeasured +
Σ B2 _
_
fstandard
Σ PACE
Ptie-1 _
+
+ +
Ptie-2 Σ Σ
+ _
Ptie-n
Psch. net interchange
Generator Control
The AGC calculation
the control must also drive the generating units to obey the
economic dispatch in addition to pushing the frequency and tie
flow errors to zero
the sum of the unit output errors is added to the ACE to form a
composite error signal
the generation allocation calculation is placed between the ACE
and the governor control / unit control loop
Generator Control
a typical layout
ACE Generation Allocation Logic Unit Control Logic
f
+ + ε unit 1 raise/lower governor
_ std pf1 Σ Σ control logic
and
_ prime mover
+
fmeas Σ + ∆Pgen1
governor
+ + +
Σ Σ Σ
B raise/lower
pf1 control logic
and
_ _ _ prime mover
+ ∆Pgen2
_ Σ
K
s governor
+ +
Σ Σ
raise/lower
_ pf1 control logic
and
Ptie-i + _ prime mover
+ ∆Pgen3
+
Ptie-j Σ Σ Pbase Σ ++ + +
Ptie-k _
+
Psch int
+
Pbase-1 Pbase-2 Pbase-3 Σ
Σ ε unit i
Generator Control
good design requirements
the ACE signal should be kept moderate in size
the ACE is influenced by random load variations
the standard deviation of the ACE should be small
the ACE should not be allowed to drift
the integral of the ACE should span an appropriate,
but small time period
drift has the effect of creating system time errors or
inadvertent interchange errors
the control action should be kept to a minimum
many errors are simple random load changes that
should not cause any control action
chasing these random variations only wears out the
unit’s speed-changing hardware
EEL 6266
Chapter 10
Interchange of Power and Energy
Economy Interchange
Having interconnections between various electric power
utility systems is beneficial
increases reliability
protects against unserved load due to loss of generation
reduces the cost of installing additional generation
more flexible operation
system frequency is more stable due to the increase in rotational
inertia from the summation of all generating units
take advantage of sharing spinning reserve responsibilities
lower operating cost
take advantage of lower cost generation
utilities with excess energy can sell production to utilities with
high load demands
Economy Interchange
Improving operating economics
opportunities may arise whenever two power systems are
operating with different incremental costs
if a sufficient difference in cost exists, then it pays to have
both systems exchange power at an equitable price
consider utility A is generating at a lower incremental cost than
utility B
utility B buys the next megawatt for its load from utility A
instead of generating it
this saves money in supplying the incremental load
utility A benefits economically from the sell of power to utility
B as long as utility B is willing to pay a price that is greater than
the cost of production of that block of power by utility A
Inter-utility Economy Evaluation
Example
consider two utilities connected with tie-lines
1 2 5
3 4 6
Utility Unit Generating Unit Generator Limits Load

Area Num. Cost Function Pmin(MW) Pmax(MW) Demand
1 1122 + 15.84P1 + 0.003124P12 150 600
A 2 620 + 15.70P2 + 0.003880P22 100 400 700
3 156 + 15.94P3 + 0.009640P32 50 200
subtotals: 300 1200
4 950 + 13.41P4 + 0.002641P42 140 590
B 5 560.5 + 14.17P5 + 0.003496P52 110 440 1100
6 560.5 + 14.17P6 + 0.003496P62 110 440
subtotals: 360 1470
Example
first, examine the behaviors when each utility performs an
economic dispatch on its own area without regards to the
neighboring area
results of individual economic dispatches for each area
area A: P1 = 322.7 MW
F1 = 6,559.07 λ = 17.856
P2 = 277.9 MW
F2 = 5,282.50
P3 = 99.4 MW
F3 = 1,835.68
FA = 13,677.25
area B: P4 = 524.7 MW F4 = 8,715.00 λ = 16.185
P5 = 287.7 MW F5 = 4,927.13
P6 = 287.7 MW F6 = 4,927.13
FB = 18,569.26
total operating cost = 32,246.50
Example
now, examine the behaviors when both utility work together
and perform a single economic dispatch for the whole system
results of the economic dispatch
area A: P1 = 184.0 MW
P2 = 166.2 MW
P3 = 54.4 MW
operating cost = 8,530.93
area B: P4 = 590.0 MW
P5 = 402.7 MW
P6 = 402.7 MW
operating cost = 23,453.89
total operating cost = 31,984.82 , λ = 16.990
interchange power = 295.4 MW from area B to area A
Example
area A costs: without interchange: 13,677.21
with interchange: 8,530.93
savings: 5,146.28
area B costs: without interchange: 18,569.23
with interchange: 23,453.89
increased cost: 4,884.66
combined net savings: 261.62
conclusions:
area A can argue that area B had a net operating cost increase of
4884.66 and therefore area A ought to pay area B this cost
area B can argue that area A had a net decrease in operating cost
of 5146.28 and therefore area A ought to pay area B this savings
The problem with the previous approach is that there is no
agreement conserning a mutually acceptable fair price
a common practice is to share the savings equally between the
two parties and set the price for such
Two power system could operate for less money when
interconnected than when operated separately
the dispatch is obtained for the interconnected system by
assuming that all necessary information is available to
everyone
such openness of communication is unrealistic
communication limits and business confidentiality prevents
everyone having all the information
The simplest way to coordinate the operations between two
power systems is to apply comparisons of the incremental
costs
process steps
assume there is no interchange, and each system runs an
economic dispatch for its own system
communicate the incremental cost values and determine who has
the lower incremental cost
the one with the lower cost runs a series of dispatches with
increased total demand
the one with the higher cost runs a series of dispatches with
decreased total demand
compare for equal values of interchange powers the incremental
costs for each system, select the interchange with equal costs
Example
rework the first example by carrying out the steps for
incremental cost comparison between the two utilities
conduct the series of dispatches in steps of 50 MW
steps utility area A: utility area B: Tie flow
1 700 MW 17.856 1100 MW 16.185 0 MW
2 650 17.710 1150 16.291 50
3 600 17.563 1200 16.395 100
4 550 17.416 1250 16.501 150
5 500 17.270 1300 16.656 200
6 450 17.123 1350 16.831 250
7 400 16.976 1400 17.006 300
8 350 16.816 1450 17.181 350
Example
note that at step 6, area A’s incremental cost is just slightly
above area B’s incremental cost, and that this relationship
changes at step 7
for minimum operating cost, the two systems ought to be
interchanging power between 250 MW and 300 MW
the procedure can then be repeated with smaller steps between
250 and 300 MW as needed
Multiple-Utility Interchanges
Regional interconnectivity
most power systems are interconnected with all their
immediate neighboring systems
one system may buy and sell interchange power simultaneously
with several neighbors
the price for interchange must be set while taking account of the
other interchanges
prices are derived from the incremental costs of the various utility
systems
a group of utility power systems may form a power pool
the price might be set by the power and energy pricing policies and
be established by a centralized pool control center
when there is no centralized center, the order in which transaction
agreements are executed is very important in costing the
interchange
Regional interconnectivity
multiple neighboring utilities may engage in “wheeling”
simply stated, wheeling is the transmission of power from one
utility, through one or more intermediate utility systems, to a
third utility system
the intermediate systems’ AGCs keep the net interchange to
specified values, regardless of the power being passed through
the transmitted power changes the transmission losses incurred
in the intermediate systems
the increased losses represent an unfair burden on the intermediate
systems, if the utilities are not part of the interchange agreement
the increased losses are supplied by the intermediate systems’
generation
utilities may impose a wheeling access charge for such power
Example
consider a three utility region with interchange agreements
data for utility areas A & B are the same as in previous examples
area A generation & costs: area B generation & costs:
400 MW @ $ 8452 /h 1100 MW @ $ 18569 /h
500 MW @ $ 10165 /h 1400 MW @ $ 23532 /h
700 MW @ $ 13677 /h
800 MW @ $ 15478 /h
data for utility area C are as follows: (w/ 550 MW load demand)
area C generation, incremental cost, and total cost
450 MW @ $ 18.125 / MWh & $ 8220 /h
550 MW @ $ 18.400 / MWh & $ 10042 /h
let there be two interchange agreements:
BA: area A buys 300 MW from area B
AC: area A sells 100 MW to area C
Example
examine the costs under a split-savings pricing policy
consider the impact with agreement BA first, then AC
Area A Area A Area B Area B Area C Area C
Gen. Cost Gen. Cost Gen. Cost
(MW) ($/h) (MW) ($/h) (MW) ($/h)
Base Case 700 13677 1100 18569 550 10042
With Agreement BA 400 8452 1400 23532 550 10042
Savings / Costs 5225 $/h 4963 $/h 0 $/h
Split Savings Price 5094 $/h from area A to area B
With Agreement AC 500 10165 1400 23532 450 8220
Split Savings Price 1767 $/h from area C to area A
Payment Summary pays a net 3327 $/h receives 5094 $/h pays 1767 $/h
Example
examine the costs under a split-savings pricing policy
consider the impact with agreement AC first, then BA
Area A Area A Area B Area B Area C Area C
Gen. Cost Gen. Cost Gen. Cost
(MW) ($/h) (MW) ($/h) (MW) ($/h)
Base Case 700 13677 1100 18569 550 10042
With Agreement AC 800 15478 1100 18569 450 8220
Split Savings Price 1811 $/h from area C to area A
With Agreement BA 500 10165 1400 23532 450 8220
Split Savings Price 5138 $/h from area A to area B
Payment Summary pays a net 3327 $/h receives 5138 $/h pays 1811 $/h
Example
comparison and results
for utilities B and C, the payments are different depending on the
order in which the agreements were conducted
it is in the seller’s best interest to sell when the buyer’s incremental
cost is high
conversely, it is in the buyer’s best interest to purchase when the
seller’s incremental cost is low
utility A sees no difference based on the order of the agreements
when several two-party agreements are made, the pricing must
follow the proper sequence
selling utilities receive more than its incremental production costs
no matter which transaction is initiated first
the rates paid are different and depend on the order of evaluation
EEL 6266
Chapter 10
Interchange of Power and Energy
Power Pools
When one system is interconnected with many neighbors, all
the transaction setup processes with every neighbor become
time consuming
individual transactions rarely results in the optimum
production cost
one solution is for several utilities to from a power pool
the power pool is administered from a central location
the pool avoids the problem of different prices depending on
the order of agreements
the pool incorporates a central dispatch office
Power Pools
The energy-broker system
the interconnected power systems deal through a broker
the broker sets up sales and purchases of energy
the broker can observe all the buy and sell offers at one time
by accepting quotations to sell and purchase, the central broker
creates an orderly marketplace where prices, supply, and
demand are known simultaneously
the “bulletin board” is the simplest form of a broker scheme
utility members post offers to buy or sell power and energy at
regular intervals
members are free to access the bulletin board at all times
members finding attractive offers are free to contact those posting
the offers and make direct arrangements for the transactions
Power Pools
more complex broker systems make arrangements to match
buyers to sellers directly
the broker may also set the transaction prices
in one scheme, members send the broker hourly buy and sell
offers for energy
each member provides information each hour
• the incremental cost and megawatt-hours it is willing to sell
• the decremental cost and megawatt-hours it is willing to buy
the broker matches buy and sell offers according to certain rules
• the lowest cost seller is matched with the highest cost buyer
• repeat until all offers are processed
• price is set to compensate the seller for the incremental generation
cost and split the savings of the buyer equally with the seller
Power Pools
price setting formula between energy buyer and seller:
Fc = Fs′ + 1 2 (Fb′ − Fs′) = 1
2 (Fb′ + Fs′)
Fs′ = incremental cost of the selling utility ($/MWh)
Fb′ = incremental cost of the buying utility ($/MWh)
Fc = cost rate of the transaction ($/MWh)
the transaction cost rate is the average of the seller’s
incremental cost and the purchaser’s decremental cost
Power Pools
Utilities Incremental Energy Seller’s Total
Example Selling Cost for Sale Cost Increase
four power systems Energy ($/MWh) (MWh) ($)
has submitted the A 25 100 2500

B 30 100 3000
following buy/sell
Utilities Decremental Energy Buyer’s Total
offers to the broker Buying Cost to Buy Cost Decrease
the maximum pool Energy ($/MWh) (MWh) ($)
savings that are C 35 50 1750

D 45 150 6750
possible are:
Fsaving = ($ 1750 + $ 6750) – ($ 2500 + $ 3000)
= $8500 – $ 5500 = $ 3000
the broker sets up the following transactions:
A sells 100 MWh to D
B sells 50 MWh to D and sells 50 MWh to C
Power Pools
Example
the broker’s transactions:
Savings Total Transaction
Transaction Computation Savings ($)
A sells 100 MWh to D (100 MWh)($45 – $25) / MWh 2000

B sells 50 MWh to D ( 50 MWh)($45 – $30) / MWh 750
B sells 50 MWh to C ( 50 MWh)($35 – $30) / MWh 250
Total: 3000
the rates and total payments under a split-savings arrangement

Price Total
Transaction ($/MWh) Cost ($)
A sells 100 MWh to D ($45 + $25) / 2 = 35.0 3500

B sells 50 MWh to D ($45 + $30) / 2 = 37.5 1875
B sells 50 MWh to C ($35 + $30) / 2 = 32.5 1625
Total: 3000
Power Pools
Example
results
utility A receives $ 3500 /h from utility D
utility A receives $ 1000 /h above its costs
utility D saves $ 1375 /h
utility B receives $ 3500 /h from utilities D and C
utility B receives $ 500 /h above its costs
utility C saves $ 125 /h
note that each participant benefits from the transactions
the chief advantage of a broker system is its simplicity
Power Pools
Clearing price policy
economist have encouraged a policy in which the broker
pricing scheme sets one single “clearing price” for energy
in each time period
the market-determined price level should be based on all the
participants’ needs and willingness to buy and sell
the market-determined price removes the absolute need to
quoting cost-based prices
utilities are free to quote offers at whatever price level, but
would be obligated to deliver or purchase energy quoted at the
market clearing price
the transactions market is similar to the stock exchange
in times of shortage, price levels could rise dramatically and
uncontrollably
Power Pools
Allocating pool savings
all saving-allocating methods implemented by central pool
dispatching are based on the premise that no member should
have higher generation production expenses than in could
achieve by dispatching its own generation to meet its own load
the split-savings method meets this criterion
under a central economic dispatching pool, transactions can be
modeled as if the power were sold to the pool by the selling
utilities and then bought from the pool by the buying utilities
under this model, allowances are made for the cost and impact of
using some utilities’ transmission networks over other’s in
carrying out the pool transactions
Power Pools
Example
reconsider the previous example of four utilities
energy interchanges are scheduled by a central dispatching
scheme with 10% of the gross system’s savings are set aside to
compensate those Utilities Incremental Energy Seller’s Total
utilities that provide Selling Cost for Sale Cost Increase
Energy ($/MWh) (MWh) ($)
transmission
A 25 100 2500
facilities to the pool
B 30 100 3000
table of increase, Utilities Decremental Energy Buyer’s Total
decreased costs, and Buying Cost to Buy Cost Decrease
the net saves: Energy ($/MWh) (MWh) ($)
C 35 50 1750
D 45 150 6750
Pool Savings: 3000
Savings Withheld for Compensation: 300
Net Savings: 2700
Power Pools
Example
weighted average incremental costs are computed for selling
and buying power
single clearing prices for buying & selling
F=
∑F × E i i
∑E i
seller’s weighted average incremental cost
FS =
($25 /MWh × 100 MWh ) + ($30 /MWh × 100 MWh )
= $27.50 /MWh
100 MWh+ 100 MWh
buyer’s weighted average incremental cost
FB =
($35 /MWh × 50 MWh ) + ($45 /MWh × 150 MWh )
= $42.50 /MWh
50 MWh + 150 MWh
Power Pools
Example
calculation of individual utility savings
area A sells 100 MWh to the pool
 $42.50 /MWh − $25 /MWh 
S1 = 100 MWh ×   × 0.9 = $787.50
 2 
area B sells 100 MWh to the pool
 $42.50 /MWh − $30 /MWh 

S2 = 100 MWh ×   × 0.9 = $562.50
 2 
area C buys 50 MWh from the pool
 $35 /MWh − $27.50 /MWh 

S3 = 50 MWh ×   × 0.9 = $168.75
 2 
area D buys 150 MWh from the pool
 $45 /MWh − $27.50 /MWh 

S4 = 150 MWh ×   × 0.9 = $1181.25
 2 
Power Pools
Example
total net savings for the four utilities
$ 787.50 + $ 562.50 + $ 168.75 + $ 1181.25 = $ 2700.00 net
the total transfer for this hour are then:

C buys 50 MWh for ($42.5 × 50 – 168.75) = $ 1956.25
D buys 150 MWh for ($42.5 × 150 – 1181.25) = $ 5193.75
subtotal $ 7150.00
A sells 100 MWh for ($27.5 × 100 + 787.50) = $ 3537.50
B sells 100 MWh for ($27.5 × 100 + 562.50) = $ 3312.50
subtotal $ 6850.00
total transmission compensation charge: + $ 300.00
total $ 7150.00
Power Pools
Example
the $ 300 that is set aside for transmission compensation is
split up among the four utility systems according to
agreements reflecting the transmission network contributions
to the pool’s network
Power Pools
Wheeling
wheeling occurs on the AC interconnections for those
networks containing more than two utilities (or parties)
whenever transactions take place
parties represent non-utility organizations
consider a system with six interconnected control areas
suppose areas A and C negotiate the sell of 100 MW by A to C
ignoring losses, A increases net generation by 100 MW and C
decreases net generation by 100 MW
W 50 M
A 50 M A W
E C E C
W
25
M
M
50 MW
25
W
25 MW 50 MW
D F B D F B
Power Pools
Wheeling
executing a power flow study before and after the agreement
shows the change in flows across the whole system
note that not all the transaction flows over the direct
interconnection between the two systems
the other systems are wheeling some amount of the transaction
the wheeled power is referred to as loop flows
the transmission system provides the parallel paths
EEL 6266
Chapter 11
Power System Security
System Security
The main concern in power system operation
minimize the operating cost
An important operating factor
maintain system security
System security
practices designed to keep the system
operating when components fail
example: a generator must be taken off-line
because of auxiliary equipment failure
maintaining proper amounts of spinning reserve,
the remaining generators can make up the deficit
system is served without too low a frequency drop
or the need to shed any load
Security Preparedness
The timing of initiating events that cause
components to fail are unpredictable
the system must be operated so that any credible event
will not put the system in a dangerous condition
power equipment is designed to operate within known limits
protection devices automatically switch out
equipment that exceeds the specified limits
if an event leaves the system operating with limits violated, the
event may be followed by a series of further actions that switch
other equipment out of service
if the process of cascading failures continues, large parts or the
entire system may completely collapse; i.e.; a system blackout
System Security
Three major functions of system security
system monitoring
contingency analysis
security-constrained optimal power flow analysis
System monitoring
provides the operators with up-to-date information
on the condition of the power systems
critical quantities are measured
voltages, currents, power flows, and the
state of circuit breakers and switches
frequency, generator outputs, and transformer tap positions
the measurements are sent to the control central
via the telemetry system
System Security
System monitoring
computers collect the telemetric data, processes and stores
them, and displays information for the operators
the computes check incoming information against
pre-selected limits and annunciate alarms in the
event of overloads or out-of-limit voltages
state estimation combines the telemetric system data with the
system’s network model to produce the “best estimate” of the
current power system condition or state
measurements, metering devices, and the communication
system contain sources of noise and random errors
combining monitoring functions with supervisory
control functions forms the Supervisory Control
And Data Acquisition (SCADA) system
System Security
Contingency analysis
allows the system to be operated defensively
many problems in power systems can cause serious trouble
within a rapid time period and the human operator can not
respond fast enough
cascading failures
models possible system troubles before they arise
using a model of the power system, a computer algorithm
predicts future operating states and gives alarms to any potential
overloads or out-of-voltage limits
System Security
Security constrained optimal power flow
analysis provides a solution to the optimal dispatch of
generation with a large number of system constraints
other adjustments are made to the dispatch schedule so that a
security analysis results in no contingency violations
the power system may operate in one of four states
optimal dispatch - state of the system prior to any contingency
optimal with respect to economic operation, but may not be secure
post contingency - state after contingency has occurred
this condition has a security violation: line overload, voltage limit
secure dispatch - state at no contingency with dispatch schedule
corrections to account for security violations
secure post contingency - state after a contingency with no
resulting security violations
Security States
Illustration of states
optimal dispatch
ignoring losses
maximum line loading of 400 MW
700 MW
500 MW 250 MW
250 MW
unit #2
unit #1
post contingency 1200 MW
consider that a transmission line has opened because of a failure

700 MW
500 MW 0 MW
500 MW
unit #2
unit #1 (overload)
1200 MW
Security States
Illustration of states
the overload line may cause a cascaded outage
secure dispatch
unit #1 generation set to maximum secure line loading of 400 MW
800 MW
400 MW 200 MW
200 MW
unit #2
unit #1
post contingency 1200 MW
consider the same contingency, resulting in no violations
800 MW
400 MW 0 MW
400 MW
unit #2
unit #1
1200 MW
Security States
Remarks
post-contingency line overloading is avoided by adjusting the
generation on unit #1 and unit #2 before an event
essence of security corrections
programming tools that make control adjustments to the pre-
contingency operation to prevent violations in the post-
contingency condition are called “security-constrained optimal
power flows” or SCOPF
Factors Affecting Security
As a consequence of several major widespread blackouts in
interconnected system, operation priorities have evolved
operate the system so that power is delivered reliably
within the constraints placed by reliability considerations,
operate the system most economically
It is highly uneconomical to build a system with sufficient
redundancy to eliminate the possibility of dropping load
systems are designed and operated to has sufficient
redundancy to withstand all major failure events
no guarantee for 100% reliability
Factors Affecting Security
Within the design and economic limitations, the system
operators try to maximize the reliability of the system at any
given time
usually a power system is never operated with all the
equipment in service at one time
occurrence of failures
maintenance
the operators are concern with possible events that cause
trouble on a power systems
focus on two major types of events: transmission line outages
and generation unit failures
line outages and unit failures cause changes in power flows and
bus voltages, which impact the remaining system equipment
Contingency Analysis
Detecting network problems
consider the 6-bus network
a typical constraint of the system is the maximum loading that a
transmission line can carry
expressed in MVA
the flow on a line may increase due to a contingency in the
system
line outage
generating unit failure
other operating factors that are impacted include:
the bus voltages
generation output to make-up the load capacity
• the slack generator may pickup all of the lost generation
• or the generators may share in picking up the lost generation
Example
base case 50.0 MW
Bus 2
241.5 kV / –3.7º
Bus 3
246.1 kV / –4.3º 60.0 MW
74.4 MVAR 2.9 MW 2.9 MW 89.6 MVAR
12.3 MVAR 5.7 MVAR
33.1 MW 26.2 MW
46.1 MVAR 12.4 MVAR 43.8 MW 19.1 MW
27.8 MW 60.7 MVAR 23.2 MVAR
15.5 MW
12.8 MVAR 42.8 MW 70 MW
15.4 MVAR
57.9 MVAR 70 MVAR
107.9 MW Bus 6
16.0 MVAR 28.7 MW 25.7 MW 231.0 kV / –5.9º
15.4 MVAR 16.0 MVAR 1.6 MW
Bus 4 3.9 MVAR
Bus 5
227.6 kV 43.6 MW 35.6 MW 226.7 kV / –5.3º
/ –4.2º 20.1 MVAR Bus 1 11.3 MVAR 15.0 MW 1.6 MW
31.6 MW 18.0 MVAR
241.5 kV / 0º 9.7 MVAR
45.1 MVAR
42.5 MW 34.5 MW 18.0 MW
19.9 MVAR 13.5 MVAR 26.1 MVAR
4.1 MW 4.0 MW 70 MW
70 MW 4.9 MVAR 2.8 MVAR 70 MVAR
70 MVAR
Example
show the impact on the system if the line 3-5 is taken out of
service due to a permanent transmission line fault
notes:
the flow on line 3-6 has increased to 54.9 MW
most of the other transmission lines experience a change in
reactive power flow
the bus voltages at the load buses are impacted
the voltage magnitude at bus 5 is about 5% below normal
Example
line outage50.0 MW Bus 2
241.5 kV / –3.4º
Bus 3
246.1 kV / –3.0º 60.0 MW
91.2 MVAR 5.1 MW 5.1 MW 68.7 MVAR
10.7 MVAR 4.1 MVAR
37.3 MW 22.9 MW
49.1 MVAR 16.2 MVAR 54.9 MW 0 MW
26.0 MW 64.6 MVAR 0 MVAR
20.9 MW
11.8 MVAR 53.6 MW 70 MW
24.7 MVAR
60.3 MVAR 70 MVAR
108.5 MW Bus 6
29.8 MVAR 26.8 MW 22.3 MW 229.8 kV / –5.2º
14.6 MVAR 19.9 MVAR 5.9 MW
Bus 4 10.2 MVAR
Bus 5
226.4 kV 43.2 MW 38.5 MW 219.3 kV / –5.5º
35.5 MW 25.6 MVAR
241.5 kV / 0º 15.3 MVAR
47.6 MVAR
42.1 MW 37.0 MW 0 MW
22.3 MVAR 22.0 MVAR 0 MVAR
7.6 MW 7.4 MW 70 MW
70 MVAR
Example
show the impact on the system when generating unit at bus 3
is taken off-line due to a steam pipe rupture
the loss of generation is pickup completely by the slack bus
generator at node 1
Example
unit failure50.0 MW Bus 2
241.5 kV / –7.0º
Bus 3
221.9 kV / –9.1º 0 MW
74.4 MVAR 21.2 MW 20.5 MW 0 MVAR
28.6 MVAR 31.4 MVAR
20.0 MW 42.9 MW
57.6 MVAR 44.2 MVAR 21.1 MW 0.6 MW
52.4 MW 25.2 MVAR 6.1 MVAR
18.3 MW
28.1 MVAR 20.8 MW 70 MW
29.4 MVAR
25.8 MVAR 70 MVAR
175.0 MW Bus 6
16.6 MVAR 55.7 MW 40.3 MW 214.7 kV / –10.1º
25.9 MVAR 41.8 MVAR 8.8 MW
Bus 4 2.4 MVAR
Bus 5
226.5 kV 64.1 MW 55.2 MW 216.8 kV / –8.4º
18.2 MW 29.7 MVAR
241.5 kV / 0º 2.6 MVAR
56.2 MVAR
62.0 MW 52.5 MW 0.7 MW
15.3 MVAR 18.9 MVAR 10.5 MVAR
10.3 MW 10.0 MW 70 MW
70 MVAR
EEL 6266
Chapter 11
Power System Security
Security Analysis
To be useful to the system operators in any way, a security
analysis study must be executed very quickly
on the order of 5 to 10 minutes
Three approaches
fast algorithms using linear approximating models of the
power system that can cover all possible cases
traditional algorithms for a selection of only the most
important cases
multiple processors / vector processors modifications of
traditional algorithms to gain speed to cover a larger selection
of cases
Sensitivity Factors
Linear sensitivity factors
useful for reaching an approximate analysis of the effect of
each outage
limitations and attributes to the linear (DC) power flow method
only branch active power flows (MW) are calculated with
approximately 5% accuracy
show the approximate change in line flows for changes in
generation and network configuration
two types of sensitivity factors
∆f l
generation shift factors: al ,i =
∆Pi
change on line l w.r.t. gen. i
∆f l
line outage distribution factors: d l ,k = 0
fk
change on line l w.r.t. line k
Sensitivity Factors
Derivation of sensitivity factors
begin with the linear load flow model
θ = [X] P
the incremental changes of the bus voltage angles for
perturbations of power injections
∆θ = [X] ∆P
first consider the generation shift sensitivity for the generator
on bus i
set the perturbation on bus i to +1 and the perturbation on all
other buses to zero
an equal but opposite perturbation (–1) must occur on the
reference bus (that is, the reference bus absorbs any changes and
all other generation buses remain fixed)
Sensitivity Factors
the change in bus phase angles are found using matrix
calculations
 + 1at row i 
∆θ = [ X ]  
− 1
 at ref row 
this is equivalent to a 1 pu power increase at bus i with a
compensating 1 pu power decrease at the reference bus
the ∆θ values are equal to the derivative of the bus angle with
respect to a change in power injection at bus i
the sensitivity factor for the change in power of line l with
respect to a change in generation at bus i is
d 1  1  dθ n dθ m  1
al ,i =
df l
=  (θ n − θ )
m  =  −  = ( X ni − X mi )
dPi dPi  xl  xl  dPi dPi  xl
line l is connected between buses n and m
Sensitivity Factors
a line outage is modeled by adding two power injections, one
at each end of the line to be dropped
the line is not actually dropped, but the effective power change
due to the loss of line flow is added as power injections
suppose line k from bus n to bus m takes an outage
~ ~
if ∆Pn = Pnm and ∆Pm = − Pnm , then line k looks like an outage
Bus n Bus m Bus n Bus m Bus n Bus m

Lines to Lines to Lines to Lines to
remainder remainder remainder remainder
of network of network of network of network
~
Pnm Pnm = 0 Pnm
∆Pn ∆Pm
Line k Line k Line k
Flows in lines before line outage Resulting flows in lines after line outage Equivalent flow in lines after line outage
Sensitivity Factors
beginning with the incremental change in power flows
∆θ = [X] ∆P
then a change in injected powers at nodes n and m produces
 M 
 ∆P  ∆θ n = X nn ∆Pn + X nm ∆Pm
∆P =  n
→
 M  ∆θ m = X mn ∆Pn + X mm ∆Pm
 
∆
 mP
and define the following
θ n ,θ m , Pnm : angles and power flow across line k before the outage
∆θ n , ∆θ m , ∆Pnm : the incremental changes resulting from the outage
~ ~ ~
θ n ,θ m , Pnm : angles and power flow after the outage
Sensitivity Factors
the outage model
~
Pnm = ∆Pn = − ∆Pm =
1 ~ ~
xk
θn − θm ( )
~ ~
θ n = θ n + ∆θ n & θ m = θ m + ∆θ m
∆θ n = ( X nn − X nm )∆Pn & ∆θ m = ( X mm − X mn )∆Pm
Pnm = (θ n − θ m ) + (∆θ n − ∆θ m ) = Pnm + ( X nn + X mm − 2 X nm )∆Pn

~ 1 1 1
xk xk xk
the injection power needed to model the outage

 
 1 
∆Pn =   Pnm
1 − ( X nn + X mm − 2 X nm )
1
 xk 
Sensitivity Factors
define a new sensitivity factor consisting of the change in
phase angle anywhere in the system to the original power Pnm
flow before the outage
∆θ i
δ i ,nm =
Pnm
∆θ i = X in ∆Pn + X im ∆Pm
   
 1   1 
∆θ i = X in   Pnm − X im   Pnm
1− 1 ( X nn + X mm − 2 X nm ) 1− 1 ( X mm + X nn − 2 X nm )
 xk   xk 
xk ( X in − X im )Pnm xk ( X in − X im )
∆θ i = → δ i ,nm =
xk − ( X nn + X mm − 2 X nm ) xk − ( X nn + X mm − 2 X nm )
Sensitivity Factors
if either bus n or bus m is the reference bus, only one injection
is made
xk X in
for m = reference: δ i ,nm =
xk − X nn
− xk X im
for n = reference: δ i ,nm =
xk − X mm
if bus i is the reference bus, then δi,nm = 0 since the reference

bus angle is constant
the line outage distribution factor is expressed as
1
∆f l xl
(∆θ i − ∆θ j )
1  ∆θ i ∆θ j  1
d l ,k = 0 = 0
=  −  = (δ i ,nm − δ j ,nm )
fk fk xl  Pnm Pnm  xl
Sensitivity Factors
if i, j, m, nor n is a reference bus
1  xk ( X in − X im ) − xk (X jn − X jm )
xk (X − X jn − X im + X jm )
xl in
d l ,k =  x − (X + X − 2 X )  =
xl  k nn mm nm  xk − ( X nn + X mm − 2 X nm )
Linear sensitivity factors

the generator shift and line outage distribution factors are
linear, allowing the use of superposition to extend the factors
consider the impact on the generator shift factor of line l and bus
i after line k has been taken out of service
∆f l = al ,i ∆Pi + d l ,k ∆f k ∆f k = ak ,i ∆Pi
∆f l = al ,i ∆Pi + d l ,k (ak ,i ∆Pi ) = (al ,i + d l ,k ak ,i )∆Pi
this is a compensated generation shift sensitivity factor
Sensitivity Factors
Shift factors and generator participation factors
superposition allows one to analyze the effects of
simultaneous changes on several generating buses
the loss of generation on bus i may be compensated by governor
action on the remaining units in the interconnected system
the remaining units pick up in Pjmax
proportion to their maximum MW rating: γ j ,i =
where Pkmax = max. MW rating for unit k ∑
k ,k ≠i
Pk
max
testing for the flow for overload on line l, with all remaining
units taking a proportion of the lost generating unit
fˆl = f l 0 + al ,i ∆Pi − ∑ al , jγ j ,i ∆Pi
j ≠i
Sensitivity Factors
Example
using the [X] matrix for the 6-bus network and line data, find
the generator shift factors and the line distribution factors
Line X (pu) Line X (pu) Line X (pu)
l1, line 1-2 0.20 l5, line 2-4 0.10 l9, line 3-6 0.10
l4, line 2-3 0.25 l8, line 3-5 0.26
0 0 0 0 0 0 
0 0.09412 0.08051 0.06298 0.06435 0.08129
 
0 0.08051 0.16590 0.05897 0.09077 0.12895
[X ] =  
0 0.06298 0.05897 0.10088 0.05422 0.05920
0 0.06435 0.09077 0.05422 0.12215 0.08927
 
0 0.08129 0.12895 0.05920 0.08927 0.16328
Sensitivity Factors
Example
generation shift factors
the factor for a unit 3 outage on line 9 is:
= ( X 3, 3 − X 6 , 3 ) = (0.16590 − 0.12895) = 0.37

df 9 1 1
a9,3 =
dP3 x9 0.10
i=1 i=2 i=3
the complete table of factors: bus 1 bus 2 bus 3
l = 1, line 1-2 0 – 0.47 – 0.40
l = 2, line 1-4 0 – 0.31 – 0.29
l = 3, line 1-5 0 – 0.21 – 0.30
l = 4, line 2-3 0 0.05 – 0.34
l = 5, line 2-4 0 0.31 0.22
l = 6, line 2-5 0 0.10 – 0.03
l = 7, line 2-6 0 0.06 – 0.24
l = 8, line 3-5 0 0.06 0.29
l = 9, line 3-6 0 – 0.01 0.37
l = 10, line 4-5 0 0.00 – 0.08
l = 11, line 5-6 0 – 0.06 – 0.13
Sensitivity Factors
Example
line outage distribution factors
the factor for a line 4 outage on line 9 is:
x4 ( X − X 62 − X 33 + X 63 ) 0.25 (0.08051 − 0.08129 − 0.16590 + 0.12895)
x9 32
d 9,4 = = 0.10 = −0.62
x4 − ( X 22 + X 33 − 2 X 23 ) 0.25 − (0.09412 + 0.16590 − 2 ⋅ 0.08051)
k=1 k=2 k=3 k=4 k=5 k=6 k=7 k=8 k=9 k = 10 k = 11

line 1-2 line 1-4 line 1-5 line 2-3 line 2-4 line 2-5 line 2-6 line 3-5 line 3-6 line 4-5 line 5-6
l = 1, line 1-2 ------- 0.64 0.54 – 0.11 – 0.50 – 0.21 – 0.12 – 0.14 0.01 0.01 0.13
l = 2, line 1-4 0.59 ------- 0.46 – 0.03 0.61 – 0.06 – 0.04 – 0.04 0.00 – 0.33 0.04
l = 3, line 1-5 0.41 0.36 ------- 0.15 – 0.11 0.27 0.16 0.18 – 0.02 0.32 – 0.17
l = 4, line 2-3 – 0.10 – 0.03 0.18 ------- 0.12 0.23 0.47 – 0.40 – 0.53 0.17 0.13
l = 5, line 2-4 – 0.59 0.76 – 0.17 0.16 ------- 0.30 0.17 0.19 – 0.02 – 0.67 – 0.19
l = 6, line 2-5 – 0.19 – 0.06 0.33 0.22 0.23 ------- 0.24 0.27 – 0.03 0.31 – 0.26
l = 7, line 2-6 – 0.12 – 0.04 0.21 0.51 0.15 0.27 ------- – 0.20 0.58 0.20 0.44
l = 8, line 3-5 – 0.12 – 0.04 0.20 – 0.38 0.14 0.27 – 0.17 ------- 0.47 0.19 – 0.42
l = 9, line 3-6 0.01 0.00 – 0.03 – 0.62 – 0.02 – 0.03 0.64 0.60 ------- – 0.02 0.56
l = 10, line 4-5 0.01 – 0.24 0.29 0.13 – 0.39 0.24 0.14 0.15 – 0.02 ------- – 0.15
l = 11, line 5-6 0.11 0.03 – 0.18 0.12 – 0.13 – 0.23 0.36 – 0.40 0.42 – 0.18 -------
Sensitivity Factors
Example
consider the generator outage on bus 3 with all the pickup of
lost generation coming from the generator on bus 1
calculate the post outage flow on line 2, from bus 1 to bus 4
base-case flow on line 2, 1-4 = 43.6 MW
base-case generation at bus 3 = 60.0 MW
generation shift factor, a2,3 = – 0.29
new flow: 43.6 MW + (– 0.29)(– 60.0 MW) = 61.0 MW
consider the line 8 (3-5) outage
calculate the post outage flow on line 9 (3-6)
base-case flow on line 8 = 19.1 MW
base-case flow on line 9 = 43.8 MW
line outage distribution factor, d9,8 = 0.60
new flow: 43.8 MW + (0.60)(19.1 MW) = 55.26 MW
Contingency Selection
AC power flow methods
the approximate methods cannot solve for the apparent or
reactive power flow changes
full ac power flow algorithms are very accurate but are also
relatively slow
because of the way power systems are designed and operated,
only a few outages will actually cause trouble
only a few power flow solutions will conclude that an overload
or voltage violation exist
a solution to the time problem is to find a way of selecting those
contingencies that are likely to result in an violation
these cases are studied in detailed
the remaining cases will go unanalyzed
AC power flow methods
selecting the problematic cases from the full list of cases is not
an exact procedure
two sources of errors can arise
placing too many cases on the short list
conservative approach
leads to longer run times
skipping cases
a case that would have shown a problem is not placed on the list
possibility of having that outage take place and cause trouble
without the operators being warned
Contingency selection
concentric relaxation
bounding
Concentric relaxation
an outage has a limited geographical effect
the loss of a transmission line does not cause much effect at far
distances (i.e., 1000 km)
the power system can be divided into two parts
the affected region
the unaffected region
the regions are found by segmenting the system into layers
layers represent the buses that are n-nodes away from the outage
Concentric relaxation
example of layering the outage effects
Layer 1 Layer 2
Layer 0
p j r
i l m
q outaged line k
s
some arbitrary number of layers is chosen and all buses

included in that layer and all inner layers are solved with the
outage in place
the buses in the outer layers are kept at constant voltage and
phase angle
selecting the optimum layer is not trivial
Bounding
one problem of the concentric relaxation is overcome by using
adjustable region around the outage to solve for the overloads
bounding accomplishes this by using three flexible subsystems
defined as follow:
the subsystem immediately surrounding the outage, N1
the external subsystem that is not solved in detail, N2
the set of boundary buses that separate N1 and N2, N3
the bounding method makes certain assumptions about the
phase angle spread across the lines in region N2
given the injections in N1 and the maximum phase angle
appearing across any two buses in region N3
Bounding
given a transmission line in N2 with flow fpq0
there is a maximum amount that the flow on pq can shift
it can increase to its upper limit, or decrease to its lower limit
∆f pqmax = smaller of [( f pq+ − f pq0 ), ( f pq0 − f pq− )]
the power shift is translated into a maximum change in phase
angle difference
∆f pq =
1
(∆θ p − ∆θ q ) → max (∆θ p − ∆θ q ) = ∆f pqmax x pq
x pq
N3 N2
N1
p q i j
∆Pk k m ∆Pm
Bounding
the maximum change in the phase angle difference across pq
defines an upper limit for angular difference in region N3
∆θ p − ∆θ q < ∆θ i − ∆θ j → ∆f pqmax x pq < ∆θ i − ∆θ j
where i and j are any pair of buses in N3

the right-hand side of the equation provides an upper limit to the
maximum change in angular spread across any circuit in N2,
providing a limit to the maximum change in their flow
a completely safe N2 region
where the maximum shifted phase angle difference is small
enough to be less than all of the biased flow limits
accomplished by expanding the N1 region to encompass the
overloaded circuits
Bounding
expansion rule:
all circuits in N2 are safe from overload if the value of
the upper limit defined by the difference in the change of angles
is less than the smallest value of the biased line flow over all line
pairs pq, where p and q corresponds to the buses at the ends of
circuits in N2:
∆f pqmax x pq < ∆θ i − ∆θ j
if this condition fails, N1 and N3 are expended to cover the
violations
calculate a new upper limit for region N3: ∆θ i − ∆θ j
rerun the test over the newly defined region N2
Example
using the 6-bus system, study the outage of line 9 (3-6)
system data and line upper limits
MW Limit fpq0
Line ∆fpqmax xpq (pu) ∆fpqmax xpq
(pu) (pu)
l1, line 1-2 0.30 0.253 0.047 0.20 0.0094
l2, line 1-4 0.50 0.416 0.084 0.20 0.0168
l3, line 1-5 0.40 0.331 0.069 0.30 0.0207
l4, line 2-3 0.20 0.018 0.182 0.25 0.0455
l5, line 2-4 0.40 0.325 0.075 0.10 0.0075
l6, line 2-5 0.20 0.162 0.038 0.30 0.0114
l7, line 2-6 0.30 0.248 0.052 0.20 0.0104
l8, line 3-5 0.20 0.169 0.031 0.26 0.00806
l9, line 3-6 0.60 0.449 ------- 0.10 -------
l10, line 4-5 0.20 0.041 0.159 0.40 0.0636
l11, line 5-6 0.20 0.003 0.197 0.30 0.0591
Example
Bus 2 Bus 3
50.0 MW 60.0 MW
1.8 MW 1.8 MW
24.8 MW
32.5 MW 16.2 MW 44.9 MW 16.9 MW
25.3 MW N1 & N3
Bus 6
44.9 MW 70 MW
100 MW
N2 24.8 MW
25.3 MW 0.3 MW
33.1 MW
41.6 MW Bus 5
32.5 MW Bus 1 16.2 MW 0.3 MW
Bus 4
33.1 MW 16.9 MW
41.6 MW 4.1 MW 70 MW
70 MW 4.1 MW
Example
to find the maximum change in N3, line 3-6 is examined
x36 ( X 33 − X 36 )P36 0.003695
∆θ 3 = = 0.449 = 0.0577
x36 − ( X 33 + X 66 − 2 X 36 ) 0.02876
x36 ( X 63 − X 66 )P36 − 0.003433
∆θ 6 = = 0.449 = −0.0536
x36 − ( X 33 + X 66 − 2 X 36 ) 0.02876
∆θ 3 − ∆θ 6 = 0.0577 + 0.0536 = 0.1113
checking for violations

the smallest value of the biased line flows is 0.0075 for line 2-4,
which is less than the maximum change
0.0075 < 0.1113
the criterion fails and we need to expand region N3
Example
Bus 2 Bus 3
50.0 MW 60.0 MW
1.8 MW 1.8 MW
24.8 MW
32.5 MW 16.2 MW 44.9 MW 16.9 MW
25.3 MW N1
Bus 6
44.9 MW 70 MW
100 MW
N2 N3 24.8 MW
25.3 MW 0.3 MW
33.1 MW
41.6 MW Bus 5
32.5 MW Bus 1 16.2 MW 0.3 MW
Bus 4
33.1 MW 16.9 MW
41.6 MW 4.1 MW 70 MW
70 MW 4.1 MW
Example
to find the maximum change in N3, line 2-5 is examined
x25 ( X 23 − X 26 )P36
∆θ 2 =
x25 − ( X 33 + X 66 − 2 X 36 )
x25 ( X 53 − X 56 )P36
∆θ 5 =
x25 − ( X 33 + X 66 − 2 X 36 )
∆θ 2 − ∆θ 5 = 0.003564
checking for violations

again, the smallest value of the biased line flows is 0.0075 for
line 2-4, which is greater than the maximum change
0.0075 > 0.003564
the criterion is satisfied
Example
MW Limit fpq0 fpq3-6 out
Line
(pu) (pu) (pu)
l1, line 1-2 0.30 0.253 0.257

l2, line 1-4 0.50 0.416 0.416
l3, line 1-5 0.40 0.331 0.322
l4, line 2-3 0.20 0.018 – 0.220 overloaded
l5, line 2-4 0.40 0.325 0.316
l6, line 2-5 0.20 0.162 0.148
l7, line 2-6 0.30 0.248 0.508 overloaded
l8, line 3-5 0.20 0.169 0.308 overloaded
l9, line 3-6 0.60 0.449 -------
l10, line 4-5 0.20 0.041 0.032
l11, line 5-6 0.20 0.003 0.191

Operation Wood

Uploaded by

Copyright:

Available Formats

Operation Wood

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Operation Wood

Uploaded by

Copyright:

Available Formats

EEL 6266

Power System Operation and Control

= 23.7 + 0.02892(50) = 25.15

is the same Pload

the transmission network loss is a function of the impedances

solve for λ in matrix form

1 400.00 300.00 150.00 15.60 28.54

P1 (MW) P2 (MW) P3 (MW)

as an iterative procedure, a stopping

assume that the fuel cost to be $1/MBtu

the lambda search technique requires the solution of the

to the Lagrange function

the gradient function is: ∇L = = (d dPN )FN (PN ) − λ

Iteration λ Total Generation P1 P2 P3 Cost

note that there is no convergence to a solution

1 800.0 300.0 200.0 300.0 23,751

note that there is a solution convergence by the 6th iteration

if the function g is the gradient vector ∇Lx, then

Iteration λ Total Generation P1 P2 P3 Cost

note the quick convergence to a solution

Pmin Pmax Pmin Pmax

Pmin Pmax Pmin Pmax

1-D Dynamic Programming Example

F1 = $ 810, F2 = $1478, F3 = $ 1998, and Ftotal = $4286

each unit must obey a rate limit such that

the optimization problem can be solved with dynamic

where j is defined for all branches connected to i

Load / Shunt Element, Only impedance

On-Load Secondary side of

common solution techniques

Ykk  Ek j<k j >k 

the matrix representation of the incremental correction

gaussian elimination is often used to solve directly for the

do for all update voltages

solve the power flow Unit 1

the equations in matrix form

the power injection at each node (bus) is:

consider a two generator system P1

F1 (P1 ) = F2 (P2 ) = Fi (Pi ) = 400 + 7 Pi + 0.002 Pi 2

the load would be under served

and generator 1 is set to supply all

the minimum loss solution:

if the losses increase for an increase in power from bus i, the

dF1 dF dF2 dF dF3 dF3

P1′′ P1′ P1 P2′ P2′′ P2 P3′ P3′′ P3

Pf1 = 1.05 Pf2 = 1.00 Pf3 = 0.95

Pi′ = Dispatch ignoring losses

where P is the vector of all generator bus net power injections

the presence of the incremental

the incremental loss for generator bus i assumes

dFi (Pi ) dFref (Pref ) dFi (Pi ) dFref (Pref )

readjust the reference as ∆C = ∑  − βi  ∆Pi

likewise, find the ratio when there is a reactive power change,

the terms ∂Pref∂θi and ∂Pref∂Ei are derived by

what combination of units is best to supply a 550 MW load?

450 MW Min. Load

loading between 600 MW and 1000 MW, run units #1 & #2

loading below 600 MW, only run unit #1

550 MW 1 1000 900 100

q(λ) = 0.0 J* = 40,000 (J* – q) / q = undefined

q(λ) = 14,982 J* = 40,000 (J* – q) / q = 1.67

q(λ) = 18,344 J* = 36,024 (J* – q) / q = 0.965

q(λ) = 19,214 J* = 28,906 (J* – q) / q = 0.502

q(λ) = 19,532 J* = 36,024 (J* – q) / q = 0.844

q(λ) = 19,442 J* = 20,170 (J* – q) / q = 0.037