Electromagnetic Transients Simulation As An Objective Function Evaluator For Optimization of Power System Performance
Electromagnetic Transients Simulation As An Objective Function Evaluator For Optimization of Power System Performance
Electromagnetic Transients Simulation As An Objective Function Evaluator For Optimization of Power System Performance
Abstract This paper presents a new method for power systems performance optimization using Electromagnetic Transients Simulation. In this method, an emtp-type program
(PSCAD/EMTDC) becomes the inner objective function calculation loop of a non-linear optimization program. The procedure is conducted in a Graphical Environment and with the
use of specially designed blocks, allows for the selection of
any arbitrary objective function. This approach of mixing
non-linear optimization with emtp-type simulation is very
powerful in addressing complex optimization problems in
modern power networks. The usefulness of this approach is
demonstrated with the selection of optimal controller gain
settings for two different Power-Electronic Systems: a dc-dc
converter and an HVDC transmission system.
Keywords nonlinear optimization, electromagnetic transients
simulation, SIMPLEX, HVDC Transmission, dc-dc Converter
I. INTRODUCTION
Electromagnetic Transients Simulation programs (referred to as emtp-type programs) are useful tools for the
analysis of transients in large power networks. They allow
for a very precise representation of the power network,
which is far more detailed as compared to representations
used in load flow and stability programs. If the power network contains Flexible Ac Transmission (FACTS) or other
power electronic apparatus, the emtp-type program is able
to model the operation of individual semiconductor
switches, and is also able to represent the controls in detail.
Similarly, transmission line, cable and machine models
used in emtp-type programs are very accurate and valid to
high frequencies. Emtp-type programs, are thus very useful in a range of applications such as the determination of
equipment stresses, controller tuning of FACTS devices
and so on. Their drawback is that they are relatively slow
and it takes a long time to complete the studies.
Optimization problems form an important category of
power system studies that are often very hard or even impossible to solve using analytical techniques. As a bruteforce approach to solving such problems, transients simulation programs usually host a multiple run feature in
which a series of runs is conducted, with the optimization
parameters being successively varied over their corresponding feasible ranges, with a view to determining the
best (or worst) case [1]. In this approach, the parameters
to be varied are changed in a well-defined manner such as
with linear, logarithmic or pseudo-random increments.
This approach is wasteful in simulation time because no
intelligence is used in determining the parameters for the
International Conference on Power Systems Transients IPST 2003 in New Orleans, USA
Network
Dynamics
(DSDYN)
Network
Solution
Output
Generation
Final
Time
No
Yes
Additional
Runs
Stop
No
Yes
3.5
(9.3)
(12.4)
(1)
2.5
(5.5) (I)
(7.3)
(III)
(IV)
(10.3)
(4.0)
(II)
1.5
0.5
(6.3)
(V)
-0.5
-1 -1
-0.5
0.5
1.5
2.5
3.5
International Conference on Power Systems Transients IPST 2003 in New Orleans, USA
Fig. 3 shows a schematic diagram of the interface between the two parts. As shown, at the beginning of each
run, the optimization algorithm generates a new point in
the optimization search space and submits it to EMTDC.
The coordinates of this point represent the values of the
parameters that are being optimized. For example, in a
controller design study, these would represent the controller setting values. In other studies these coordinates could
represent network parameters such as resistance or inductance values. PSCAD/EMTDC conducts the next run using
these coordinate values. From the computed transient, the
desired objective function is extracted. For example, in a
controller optimization study, this objective function could
be the peak overshoot or the integral squared error (the
integral of the square of the deviation of the response from
the desired reference). This value is returned to the optimization program, which then uses it to determine the next
point to be investigated.
A certain amount of bookkeeping is necessary in order
to make the interface work properly. For example, the
PSCAD/EMTDC run can be initiated from different points
within the optimization routine. After the PSCAD/EMTDC
run, the flow must return to this point. This requires the
setting of a flag to indicate the point of re-entry.
D. Software Implementation
The optimization routine is associated with a graphical
component block within PSCAD/EMTDC. Placing this
Initialization
Vd = D E , where D =
TON
TON + TOFF
(2)
Effecting a change in the dc voltage Vd, via the duty cycle D, can regulate the load current. The controller in Fig.
5 compares the measured current with the reference and
passes the error through the proportional-integral block
and produces a correction D to the steady state duty cycle
D0 so that the error is eventually (ideally) reduced to zero.
The gain K and time constant T of this controller play an
important part in ensuring a good response. If set incorrectly, they can produce an oscillatory, or even unstable
response.
In order to set these gains to their optimal value, an objective function is selected as in Eqn. 3. This is the wellknown Weighted Integral Square Error (WISE) popular
in controls literature. Here, T0 is the time at which the reference is changed and TF is the total length of the simulation run. T1 is a suitably selected intermediate point that
permits the assignment of different weighting factors K1
and K2 to the initial and later portions of the response. This
function is generated within the PSCAD/EMTDC environment and is made available to the optimization routine
at the end of any given run. As is evident, the objective
function has a zero value if the current faithfully reproduces the reference. Note that positive and negative deviations from the reference are penalized equally through the
squaring function.
Bookkeeping
and Generation of
a New Point
PSCAD/EMTDC
Simulation Run
and Objective Function
Evaluation (at the end of
the run)
Converged?
No
Yes
International Conference on Power Systems Transients IPST 2003 in New Orleans, USA
T
Parameter
Initial Value
Optimal Value
K1
8.0
100
95
90
WISE
85
80
75
70
65
In this exercise six different gain and time constant settings are simultaneously optimized in order to provide a
rapid startup response for the dc current in a high voltage
DC transmission system.
The model for the system is the CIGRE HVDC benchmark model [6] shown in Fig. 8, with the controller shown
in Fig 9. The dc system is rated at 500 kV (dc), 1000 MW.
The ac side systems are fairly weak; with short circuit ratios on the sending and receiving end systems of 2.5 and
2.35 with damping angles of 84o and 76 o. Additional details are provided in the appendix.
The controllers on the rectifier and inverter side are
identical in structure, but have different set points. There
are two principal control loops; one for controlling the dc
current, and the other for controlling the extinction angle .
At any given operating point, only one of these loops is
enabled via the maximum select block as shown in Fig. 9.
Although at any given instant, only one of the loops is
active, during the startup transient, each of these loops may
interchangeably come in and out of operation.
60
55
0
IL
10
15
Run Number
20
25
30
60
IL (A)
40
20
(a) Initial Parameter Setting
0
0.05
0.1
0.15
0.2
0.25
Time (sec)
0.05
0.1
0.15
0.2
0.25
Time (sec)
0.3
0.35
0.4
0.45
0.5
0.45
0.5
50
IL (A)
40
30
20
10
0
0.05
T0 (sec)
0.1
K2
5.0
105
Isrc
T (sec)
0.01
0.122
T1 (sec)
0.2
K
1.6
0.86
0.02
0.35
0.4
0.01
140.63
Iref
+
Id
K+1/sT
++
Firing
Logic
D0
International Conference on Power Systems Transients IPST 2003 in New Orleans, USA
ISE =
( I ref I d ) 2 dt
Imag
(4)
Idref
where T is the final simulation time and Iref and Id are the
reference and actual dc line current. Obviously the faster
and smoother the current reaches its steady-state value
asserted by Iref, the smaller the value of ISE.
The startup of the system is simulated by ramping up the
current order from 0 kA to 2 kA over 100 ms. As in the
previous example, the control parameters within the transients simulation program are modified judiciously in each
subsequent run by the non-linear Simplex algorithm; with
the ISE objective function evaluation being carried out
within the transients program. The initial and optimized
values for the three pairs of gain and time constant are
listed in Table II. Fig. 10 shows reference current and the
actual current flowing through the line with initial and
optimized settings, showing clearly the marked improvement in the response. The long duration of the simulation
is deliberately chosen so as to allow the presence of any
slowly growing unstable modes to be observed. In case
such modes exist, they will increase the objective function
value and thus be rejected.
The total number of simulation runs (objective function
evaluations) for this case was 170. It is important to note
that this is orders of magnitude smaller than what would be
obtained with blind search methods. For example, if each
of the variables were varied over 10 increments, the total
number of runs would have been 106. Even then, the
granularity of the solution would have been excessive,
with a potential error of 10%. Also, the proper selection of
the search intervals for the parameters requires some apriori knowledge of the location of the minimum (maximum); and for an improper choice, the minimum could lie
outside the search ranges. Using the optimization approach
requires no such knowledge, as the method will automatically find at least a local minimum (or maximum).
AC 1
Rectifier
Inverter
Id
Parameter
K1
T1
K2
T2
K3
T3
Initial
1.0
0.015
0.36
0.012
0.42
0.085
Final
1.37
0.004
0.81
0.01
0.33
0.065
VDCL
180
K2+1/sT2
ref
2
1
0
(a) Initial Parameter Setting
-1
0
0.05
0.1
0.15
0.2
0.25
0.3
Time (sec)
0.05
0.1
0.15
0.2
0.25
0.3
Time (sec)
0.35
0.4
0.45
0.5
0.45
0.5
3
Id(kA)
2
1
0
(b) Optimal Parameter Setting
-1
0
0.35
0.4
IV. CONCLUSIONS
A novel method for integrating optimization algorithms
in transient simulation programs is introduced in this paper. Transient simulation is used to evaluate the optimization objective function, while the optimization algorithm
serves as the engine for producing new candidate points in
the optimization space. Using the proposed approach, a
number of successive runs each with a new point in the
optimization space are performed to find the maximum or
minimum of the objective function. The approach significantly reduces the computational burden and time for the
solution of complicated multi-variable non-linear optimization problems.
The power of this approach was demonstrated through
determining the solution to two optimization problems
concerning power electronic controller settings. The results obtained show considerable savings in computation
time over the currently used multiple run approach,
where the optimization variables are varied over prespecified ranges. It is also seen that unlike the blind search
methods, an a-priori knowledge of the minimum is less
important as the method is able to find the local minimum
by varying its increments automatically.
Select
Max
Curr. Error
AC 2
Inverter (I)
Id
Vdc
Rectifier
K1+1/sT1
International Conference on Power Systems Transients IPST 2003 in New Orleans, USA
ACKNOWLEDGMENTS
REFERENCES
[1] PSCAD/EMTDC Users Manual, Manitoba HVDC Research
Center,, 1998.
Table III CIGRE HVDC model specifications
[2] D.A. Woodford, A.M. Gole and R.W. Menzies, "Digital modeling of dc links and synchronous machines, IEEE Trans., Power
App. and Syst., Vol. PAS-102 (6), pp. 1616-1623, 1983.
AC System
378 kV
SCR: 2.5 at 84
AC System
218.2 kV
SCR: 2.35 at 76
Rectifier Side
Filters and Capacitors (MVAR)
11th harmonic: 252
13th harmonic: 252
Capacitors: 125
Inverter Side
Filters and Capacitors (MVAR)
11th harmonic:252
13th harmonic: 252
Capacitors: 125
Transformers
(each)
345/211.4 kV
597.8 MVA
Xs = 0.18 p.u.
Transformers
(each)
207.2/230 kV
585.9 MVA
Xs = 0.18 p.u.