THE APPLICATION OF OPTIMIZATION TECHNOLOGY FOR
ELECTRICITY MARKET OPERATION
David Sun, Xingwang Ma, Kwok Cheung
AREVA T&D Inc.
Bellevue, Washington
Abstract – Mathematical optimization provides a formal
framework that enables systematic and transparent decision
making. Significant progress has been made in recent years in
applications of formal optimization techniques for competitive
market based resource commitment, scheduling, pricing and
dispatch. This paper describes experiences with the application of
optimization technology for electricity market operation.
Experiences with and the status of the latest development in the
security-constrained unit commitment and economic dispatch
algorithms are described with references to actual market practices.
key words – Electricity market operation, optimal power flow,
security-constrained unit commitment and economic dispatch,
locational marginal pricing, formal optimization.
I. INTRODUCTION
A fundamental attribute of successful electricity market is its
ability to satisfy and balance diverse set of business requirements.
Examples of business process requirements are maintaining secure
and reliable operation of the physical system, providing correct
economic signals to enable competitive market activities, etc..
Mathematical optimization provides a formal framework that
enables systematic and transparent decision making. This paper
describes experiences with and the status of the development of
optimization-based applications for electricity market operation.
The unified framework for competitive electricity market and
grid reliability is founded on the duality theory that is concerned
with the outcome in a competitive market and the mathematical
solution to a constrained optimization problem. This duality result
states that a competitive market solution consists of both the
market clearing quantities and the associated market clearing
prices that equate market demand to market supply in all
interrelated parts of the market. Only when the market clearing
quantities are priced at the logically associated prices is the overall
solution incentive compatible. When prices are incentive
compatible, participants will provide bids that reflect its actual cost
or value, and produce or consume the efficient quantities.
Participants’ production or consumption of the efficient quantities
leads to grid reliability. The duality theory clearly suggests that
formal optimization techniques be utilized for electricity market
clearing so that market based dispatches, when executed in realtime grid operation, may lead to harmonious unification of market
efficiency and grid reliability. This fundamental principle of the
duality theory has been reflected in most existing market designs.
Most market rules require that RTO
• Provide a reliable grid and coordinate grid operation to
minimize costs consistent with secure operation of the grid;
• Schedule and dispatch generation to satisfy market demand for
electricity at minimum price while taking into account the
security of the grid;
• Purchase for the benefit of grid reliability the ancillary services
and resources that are necessary for economic and secure
delivery of electricity;
X. Ma, D. Sun and K. Cheung are with AREVA T&D Inc., 11120 NE
33rd Place, Bellevue, WA 98004.
•
Provide information to participants in an open, nondiscriminatory manner to help them make consumption and
production decisions.
These competitive market rules are also in line with the longheld industry principle about reliable and economic grid operation.
Optimization based economic dispatch and unit commitment has
been practiced for decades to coordinate grid’s economic and
reliable operation. But traditional optimization methods for
economic dispatch and unit commitment are less rigorous and the
underlying grid models are over-simplified. The key to competitive
market success lies in the dealings with modeling details. The
constrained optimization problem behind electricity market
dispatch and pricing must therefore model grid details in order to
realize the market efficiency as promised in the theory of
economics and mathematics. As a result, significant progress has
been made in recent years in applications of formal optimization
techniques for competitive market based resource commitment,
scheduling, pricing and dispatch. Two important optimization
based applications for market clearing, security constrained
economic dispatch (SCED) and unit commitment (SCUC) are
discussed in this paper.
The paper is organized as follows. The SCED algorithm and its
evolution are presented in Section II. The SCUC algorithm is
discussed in Section III. Section IV concludes the paper.
II. SCED FOR MARKET CLEARING
The market clearing function determines the quantities that
market participants produce or consume, and the prices that are
used to settle the quantities of electricity services produced or
consumed. Correctly determining spot electricity prices is critical
not only to successful competitive market operation, but also to the
reliability of the grid. The optimal power flow (OPF) based
framework [1] has been proposed for spot electricity pricing in
many academic publications [2,3]. Due to practical limitations of
non-linear optimization methods and intrinsic properties of grid
operations, the DC model SCED algorithm, which is a variant of
the AC OPF with many market related extensions, has been
commonly accepted as the market clearing method [4,5]. Below,
traditional OPF methods are described first, followed with
discussion of the SCED method for market clearing.
A. OPF Evolution for Electricity Pricing
OPF is aimed to find a power flow solution that leads to both
economic and reliable grid operation. It is common practice that
the control variable values determined in the OPF solution are used
as guidance to achieve optimal grid operation. The dual solution,
that defines the marginal prices of relevant quantities, has been
largely ignored until the advent of competitive markets. According
to the duality theory, the dual solution in the form of prices is
equivalent to the primal solution in the form of quantities in
achieving the optimal objective.
The OPF problem is stated in mathematical terms as follows:
min f (u P , uQ )
subject to
(λ )
(µ )
g (u , x ) = 0
h(u, x) ≤ 0
u min ≤ u ≤ u max
where
u = [u p , uQ ]T : Control variables, such as generator real and
reactive power output and transformer tap positions;
f (u , x) : Objective function typically represented with total
production cost;
x = [ xθ , xV ]T : State variables, such as bus voltage angles and
magnitudes;
g (u , x) = [ g P (u , x ), g Q (u , x)]T : Bus real and reactive power
balance equality constraints;
h(u, x ) = [hP (u , x ), hQ (u, x )]T : Inequality constraints, such as
transmission line and transformer thermal limits and bus
voltage magnitude limit constraints;
λ = [λ P , λQ ]T : Shadow prices of bus real and reactive power
balance constraints;
µ = [ µ P , µ Q ]T : Shadow prices of grid security constraints;
u min = [u Pmin , uQmin ]T : Minimum limits for control variables;
u max = [u Pmax , uQmax ]T : Maximum limits for Control variables.
In the OPF formulation, each of the equality or inequality
constraints has a corresponding shadow price at its optima. The
shadow price, µ , of a binding grid security constraint defines the
marginal cost of maintaining the specific grid security constraint.
The shadow price, λ P of bus real power balance constraint can be
used to define the short-run marginal prices of
producing/consuming real power at a bus. The shadow price, λQ ,
of bus reactive power balance constraint may be interpreted as the
marginal cost of producing/consuming reactive power at a bus.
When generators or loads are allowed to respond to the spot
price signals, the prices can induce generator responses to achieve
economic and reliable grid operation as effectively as the MW
dispatches. In order for price signals to be able to convey effective
control information for economic and reliable grid operation, price
signals must be unambiguous, transparent and explainable. While
the OPF formulation provides the theoretical basis for calculating
spot prices of electricity, finding a global optimal solution to a
realistic non-linear optimization problem is still artful, and thus it
has not been widely utilized in practice. However, real and reactive
power problems are only loosely coupled. While the reactive
power problem is highly non-linear, the real power problem is very
linear. Therefore, real and reactive power problems can be solved
in a decoupled fashion.
The decoupled OPF formulation [6] for real power can be
described as follows:
min f (u P )
subject to
(λ P )
g P (u P , xθ ) = 0
(µ P )
hP (u P , xθ ) ≤ 0
u Pmin ≤ u P ≤ u Pmax
This formulation is very similar to the previous full AC OPF
formulation, except that reactive power related control variables,
voltage magnitude variables and voltage security constraints are
reduced to parameters with pre-determined values. This is
currently the most commonly used optimization framework in
electricity market clearing and price calculation.
B. SCED for Market Clearing
The security constrained economic dispatch, which is
essentially a reformulation of the decoupled OPF, is the commonly
used algorithm for market clearing. The SCED algorithm has been
successfully applied to solve market clearing problem with large
scale grid models, such as 2000 bus ISO-NE market, 15,000 bus
PJM market, and 30,000 bus MISO market in the US.
The SCED algorithm provides several advantages over the
traditional optimal power flow:
• The SCED algorithm uses a detailed, linearized grid model,
and is solved with state-of-the-art linear programming (LP)
techniques.
• With the use of standard LP methods, the SCED algorithm
provides almost unlimited capability to handle as many linear
constraints as necessary.
• Both base and contingency grid security constraints can be
effectively handled.
• Ancillary services can be incorporated in the LP based
optimization framework.
• Participant bid data models can be accurately reflected in the
formulation. This modeling flexibility makes it possible for an
electricity market to offer as many participant choices as
needed for them to manage risks.
• The same framework can be applied for energy, ancillary
service and transmission right market clearing.
• Below we discuss the applications of the SCED algorithm to
solve the joint optimization of energy and ancillary services.
- Joint optimization of energy and ancillary services
In many of the existing electricity markets, energy and
ancillary services (AS) are both allowed for competitive bidding.
The fact that the same resource and same capacity may be used to
provide multiple products dictates that AS market operation should
be closely coordinated with the energy market. This close
coordination is best achieved through joint optimization of AS and
energy markets, which minimizes the total market cost of meeting
system demand and AS requirements while satisfying network
security constraints.
In the development of AS markets, different market designs
have been attempted. The designs can be categorized as three
approaches: independent merit-order stack approach, sequential
market clearing, and joint optimization.
• Independent Merit Order Dispatch: Independent merit order
based market clearing ignores the capacity coupling between
energy production and supply of ancillary services. Each
product is cleared separately from other products based on a
separate merit order stack. This approach is simple, but it easily
leads to solutions that are physically infeasible.
• Sequential Market Clearing: The sequential approach
recognizes that energy and reserves compete for the same
generating capacity. In essence, a priority order is defined for
each product. Available capacity of a resource (e.g. generating
unit) is progressively reduced as higher priority products are
dispatched from that resource. The degree of sophistication of
recognizing the coupling varies from market to market.
While the sequential dispatch is an improvement over
purely independent merit-order dispatch, it needs further
improvements for handling inter-dependencies among the
coupled products. Explicit evaluation of costs associated with
lost opportunity, production cost impacts, etc. are mechanisms
that are used to provide quantitative indices for analyzing
tradeoff decisions. They also help reduce the arbitrariness in
the dispatch sequence.
•
Joint Optimization: In the joint optimization approach, the
objective is to minimize the total cost of providing ancillary
services along with energy offers to meet forecast demands as
well as AS requirements. The allocation of limited capacity
among energy and ancillary services for a resource is
determined in terms of its total cost of providing all the
products relative to other resources. The effective cost for a
resource to provide multiple products depends on its offer
prices as well as the product substitution cost. Product
substitution cost arises when a resource has to reduce its use of
capacity for one product so that the capacity can be used for a
different product (leading to an overall lower cost solution).
The product substitution cost is determined internally as part of
the joint optimization. This product substitution cost plus its
bid price reflects the marginal value of a specific product on
the market. The marginal value, which is typically the market
clearing price, represents the price for an extra unit of the
product that is consistent with the marginal pricing principle
for the energy product.
Market clearing prices for ancillary services reflecting product
substitution costs create price equity among the multiple products.
Price equity refers to the fact that a market participant can expect
to receive equivalent amount of profits no matter which kind of
service the participant is assigned to provide. Price equity incents
participants to follow dispatch instructions. Without the price
equity, participants would tend to provide those services that
produce the most profits, which could deviate significantly from
dispatch instructions. Significant deviations from dispatch
instructions could seriously degrade dispatcher’s ability to
maintain grid reliability. Therefore, joint optimization of energy
and AS market has been accepted in many electricity markets,
including PJM-RTO, New York-ISO, ISO-New England, New
Zealand and Australian markets.
- An SCED formulation example
Assume that a joint market allows bidding for multiple
products, generation energy offer, price-responsive demand bid,
and two types of ancillary service - ten-minute reserve, and thirtyminute reserve. With unit commitment schedules given, an
exemplar SCED formulation for this joint market may be described
in mathematical terms as follows:
min c( P) + cs( S ) + co(O) − cd ( D)
subject to
• System power balance constraint:
(λ )
∑ ( Pi − Di ) −FD − PL = 0
| Pi − Pi0 |≤ RRimax
(φi )
where Greek labels are corresponding constraint shadow prices;
c, cs, co, cd: Cost or benefit associated with unit production,tenminute reserve, thirty-minute reserve and demand bids;
Pi , Di , S i , Oi : Decision variables for generation dispatch, load
dispatch, spinning reserve dispatch and operating reserve dispatch
of resource i;
Pimin , Pimax : Min and max dispatch limit for generator i;
Pi0 : Initial generation output for generator i;
Dimax : Maximum dispatch limit for load i;
FD , PL : Fixed demand and transmission losses;
d i : Load distribution factor for fixed demand;
S max , O max : Ten- and thirty-minute reserve requirement;
: Grid security limit for constraint l;
Lmax
l
al , i : Sensitivity of constraint l with respect to bus i.
The optimal solution to this SCED problem determines the
market clearing quantities and market clearing prices by location.
The market clearing prices by location, called locational marginal
prices, and AS market clearing prices are by-products of the
optimization solution. Under the joint optimization based market
for energy and multiple ancillary services, the market clearing
prices for the multiple products have the following important
characteristics:
• Locational
marginal
prices
for
energy:
∂PL
LMPi = λ − λ
− ∑ al ,i µ l
∂Pi
l
These locational price results give precise representation of the
cause-effect relationship that is consistent with grid reliability
management.
• Higher prices for higher quality of ancillary services
o Market clearing pricing for ten-minute reserve:
ρS = γ S + γ O
o Market clearing pricing for thirty-minute operating
reserves: ρO = γ O ≤ ρ S
• Marginal equity between energy and reserve prices:
∂C ( P )
o LMPi −
= ρ S , if S>0
∂Pi
i
•
Ten-minute reserve requirement constraint:
(γ S )
•
o
max
∑ Si ≥ S
i
Thirty-minute operating reserve requirement constraint:
(γ O )
max
∑ ( Si + Oi ) ≥ O
i
• Grid base-case and contingency constraint:
(µl )
∑ al ,i ( Pi − Di − d i × FD) ≤ Ll
i
•
Generator minimum generation limit constraint:
(τ imin )
•
Generator joint maximum generation limit constraint:
(τ imax )
•
Pi + Si + Oi ≤ Pimax
Price-responsive load dispatch range constraint:
(η imax )
•
Pi ≥ Pimin
0 ≤ Di ≤
Dimax
Generator ramp-rate limit constraint:
LMPi −
∂C ( P)
= ρ O , if S=0 and O>0
∂Pi
•
LMPs for energy are always higher than market clearing prices
for ancillary services except that there exists ancillary service
capacity shortage. This can be seen from the above equations.
These price relationships are essential to encouraging rational
participants’ responses to market signals. But these price
relationships are true only under rigorous mathematical terms.
Therefore, adopting a formal optimization technology in market
clearing is central to market efficiency and grid reliability.
III. OPTIMIZATION-BASED UNIT COMMITMENT
A. Market Oriented Unit Commitment Problem
Unit commitment determines unit startup and shutdown
schedules that meet forecast demand at minimum startup cost and
energy production cost. Traditional unit commitment problems in
general did not explicitly take into account grid security constraints
in the commitment process. Even with the simplified modeling
assumption, finding an optimal solution to unit commitment has
been a challenging task due to its nature of curse of dimensionality.
Many existing electricity markets, such as PJM, ISO-New
England, New York ISO, and Mid West ISO, allow three-part bids
including startup cost, no load cost, and energy cost components.
These markets also adopt a two-settlement market structure that
consists of a day-ahead market and a real-time market. The dayahead market is a financial market that determines the minimum
cost commitment schedules based on participants’ bid-in fixed
demand, price-sensitive demands, virtual transactions, external
interchange schedules and transmission security constraints.
Because many trading activities of the day-ahead market are used
as mechanism to hedge financial risks in the real-time market,
participants’ bids for demands and virtual transactions can deviate
significantly from the physical grid operation. To guarantee grid
reliability, a reliability commitment is required after day-ahead
market is cleared. In the reliability commitment, the units
committed in day-ahead market remain committed. Additional
units required to meet forecasted demand and reserve capacity
requirements are committed in the form of reserve capacity by
minimizing the startup and no load cost components of available
units that are not commitment in the day-ahead market.
The market oriented unit commitment problem becomes more
complex compared to the traditional unit commitment problem.
The unit commitment schedules are applied in the SCED based
market clearing for determination of day-ahead settlement
quantities and prices. From a market efficiency perspective, those
units committed in the day-ahead market process should be able to
collect sufficient financial proceeds from day-ahead market
settlement to cover their production cost based on the day-ahead
market clearing quantities and prices. With this result, uplift
charges resulting from cost compensation for market committed
units are reduced, cost socialization is minimized, and market
efficiency is improved. More importantly, any units that would be
able to make a profit in terms of the day-ahead market clearing
prices, if they were committed, should not be left uncommitted in
the day-ahead market solution. When a generator that could be
profitable is not committed, a rational explanation must be
presented to the generator. This is important for creating a fair and
transparent market that encourages rational bidding behavior
complying with grid reliability requirements.
The key to fair and transparent market that involves day-ahead
unit commitment is to include grid security constraints in the
commitment solution. As such, security constrained unit
commitment (SCUC) has become an integral part of many dayahead markets. Successful solution to the SCUC problem presents
great technical challenges, due to its requirement on global
optimality and the need to deal with large-scale commitment
problems of 1000 to 3000 units over a 2 to 7 day time horizon with
grid models of over 20,000 buses.
In addition to the requirements of commitment optimality and
the ability to deal with large scale problems, requirements on new
functional capabilities increase. Hydro-thermal coordination needs
to be handled more effectively and consistently. Combined-cycle
units need to be modeled to reflect the coupling effect of gas
turbines and steam components.
In meeting the challenges, great efforts have been made and
significant progress has been achieved in advancing unit
commitment techniques.
B. Advances in Unit Commitment Methods
The Unit Commitment problem is a large-scale combinatorial
scheduling problem. For N units over M time periods there are a
maximum of (2N-1)M combinations to evaluate. For a small
problem where N=5 and M=24, there are (s5-1)24 = 6.2x1035
possible combinations! Solving even this small problem by direct
enumeration is clearly an intractable approach.
Historically, unit commitment algorithms have focused on
heuristic methods that intelligently reduce the number of
combinations that need to be evaluated. A popular method is to
assume that units are restricted to some fixed startup/shutdown
order. This assumption reduces the number of possible
combinations to 2N x M, or for our example above to 25 x 24 = 768
combinations. This new problem can be solved optimally using a
dynamic programming algorithm.
More recently, Lagrangian Relaxation (LR) algorithms have
been utilized. The LR approach decomposes the problem by unit.
Each single unit problem can be optimally solved using a small
single unit dynamic programming (SUDP) algorithm. A
coordination module is used to adjust parameters (Lagrange
multipliers) sent to the SUDP to achieve feasibility of system
constraints. This approach implicitly evaluates many more
combinations than is possible using priority-based methods. The
LR decomposition provides a duality gap, which is a measure of
how far the current solution is from the estimated lower bound (the
theoretical minimum).
Early LR algorithms [7-9] had numerous convergence
problems. Later a “sequential bidding” method was developed,
which utilized many of the LR components, but avoided
convergence issues. This method creates a priority order that is
determined dynamically by the algorithm, rather than a priori.
This worked well for non-market-based systems, and provided
additional capabilities for fuel and emissions constraints.
Early experiences with the market oriented unit commitment
made it clear that priority-based methods were not appropriate for
market clearing purposes. Temporal changes in both unit offers
and availability effectively invalidate the assumption of using a
fixed priority over all study periods. The requirements for optimal
hydro-thermal coordination and explicit inclusion of many grid
security constraints created nearly insurmountable difficulties for
the traditional unit commitment algorithms, such as priority-based
method and classic dynamic programming method.
The LR method is the most commonly used algorithm today
for market based unit commitment. In some practical market
applications, LR algorithms provided significantly improved
results that typically have duality gaps of less than 1%, indicating
near-optimum schedules. It is interesting to note that the LR
decomposition is analogous to an economic market simulation.
The LR coordinator can be viewed as the market clearing function,
and the SUDP as the participant bidding mechanism. Based on
current estimates of system prices, the SUDP determines
individual unit bids that maximize each unit’s profit. The LR
coordinator clears the market and commits additional units as
necessary to meet system constraints. The solution iterates to
achieve a minimum cost solution in much the same way as the
market clearing process is repeated several times each day. As a
result, the solutions from the LR tend to satisfy the participants,
since their profits are maximized, while also producing secure and
minimum cost system results.
The LR method has been successfully applied in clearing large
scale market of over 1000 units. But the LR method is frequently
challenged due to its inability to prove that it has found a global
solution. In search of a global optimal solution for the unit
commitment problem, attempts were made decades ago to use
mixed integer programming (MIP) method without success.
In recent years there have been significant improvements in
MIP algorithms. These MIPs utilize a “branch and bound” method
that can implicitly evaluate all combinations, which enables MIP to
ascertain global optimality. This claim of a globally optimal
solution cannot be made for any of the other methods.
Since its development in the early 1970’s, the branch-andbound method had suffered from poor and unpredictable
performance when applied to practical sized mixed integer
problems. However, the situation changed significantly during the
last few years. A major contributor to MIP performance
improvement is the drastic performance improvement of the LP,
which is at the core of any MIP algorithm. In addition to
improvements due to LP, there are a large number of schemes that
have been developed over time for improving the intelligence of
the branch and bound search logic. Although there was not a single
monumental breakthrough in branch-and-bound, these different
schemes collectively provide a tool-kit approach for tailoring to
specific MIP problems. The use of MIP solvers is appealing in that
they eliminate numerous heuristics utilized in other approaches,
and theoretically allow for inclusion of complex constraints that
are difficult to deal with using other methods. Intensive research
on application of the MIP method is being conducted and initial
success with MIP application to market based unit commitments
with over 1000 units has been achieved [10].
C. An SCUC Example Problem: Reliability Commitment
In this section, the reliability commitment problem in a multisettlement market system is used to demonstrate that successful
market operation depends on not only the use of the right
optimization methods, but the correct problem formulation and
solution performance.
- Solve the right problem
While the Day-Ahead (DA) market produces financially
binding schedules for the next operating day, these schedules may
not provide an adequate resource plan for physical system
operation. Reserve adequacy (RA) analysis must be performed to
generate resource operating plans that ensure secure and reliable
operation of the power system. Additional generation capacity
may have to be committed for reliability purposes to bridge the gap
between the capacities as seen from financially binding day-ahead
schedules and the capacities needed to meet RTO’s forecast of the
physical demand.
Although the RA and DA market clearing processes share
many common characteristics, such as the conventional UC
problem definition, these are two fundamentally different business
processes and have important differences in their respective
problem definitions. The RA process not only needs to respect
physical system requirements at least cost, but also must minimize
any interference or distortion to the real-time economic price
signals. The latter is particularly affected by the physical resource
commitment schedules produced by the RA process. If the demand
bids cleared in the day-ahead market should fall short of the
forecast/actual demand, then real-time energy prices must be
allowed to rise. The rise of real-time energy price would be the
result of additional energy produced by units that were committed
by RA. These are low cost providers of operating reserve but not
necessarily energy. The decision criteria in RA commitment is
based on start-up and cost of minimum generation, but not on
incremental energy cost.
The transition from 3-part energy bids in the DA market to the
2-part energy bid in the RA process is a relatively simple function
from the application software and optimization method
perspective. However, this fundamental transition in problem
definition from DA to RA leads to additional important and related
aspects of RA definition. The principal areas of continued RA
problem definition centers on the issues of: (a) modeling and
analysis of transmission security and incremental capacity
commitment; and (b) RA input data preparation (interpretation and
extrapolation). These are indicative of the challenge for the RA
analysis; i.e. the transition between market operation and physical
operation, which is essential for reliable operations. Thus, it is
naturally subjected to significant re-definition as the market
continues to evolve.
The early and principal focus of the RA process was to ensure
that the RTO system would have sufficient incremental capacities
committed, in the form of operating reserves, to meet forecast
demand. In that model, the energy balance in the SCUC was
maintained at the level of cleared DA demand bids and thus the
resulting power flows on the transmission system does not reflect
the level commensurate with the energy demand forecast.
The growing need for further improvement over the earlier RA
process clearly points in the direction of more accurate modeling
and analysis of transmission security constraints. This direction
implies potentially fundamental augmentation of the RA business
requirements. More than system-wide incremental capacity
commitment process, the RA commitment decision will need to
better recognize the impacts of transmission security. This requires
increased emphasis on energy scheduling, including possible redispatching MW from the reference DA schedules, which in turn
requires modeling energy cost and energy demand. On the other
hand, it is also fundamental that committing for reserves (based on
2-part bids of startup and min-generation cost but not energy cost)
remains the primary business objective for RA. How to navigate
through seemingly conflicting requirements to produce useful and
consistent definition of the RA problem is important and
challenging. Close co-operation with RTO customers is key to
resolving the challenges confronting the market operators and
security coordinators.
- Solve the problem fast
All markets have definitive time-lines for specific market
events. For example, several day-ahead markets close at noon and
publish their results by 4PM. To provide sufficient flexibility and
margin, the market-clearing applications need to be completed in
less than half of the total available time. This can be quite
challenging given the size and complexity of the problem. The
adopted system architecture for solving the SCUC-RA problem
shall be able to withstand the challenges. Using the concept of
Benders Decomposition [11], the problem is decomposed into a
unit commitment module and a simultaneous feasibility test (SFT)
module. Such a system design is shown in Figure 1. Some of the
key system features include the following:
• Increase speed by plugging additional processors – For the
seven-day SCUC RA analysis, 168 SFT solutions must be
conducted (one for each hour) and each SFT solution involves
the analysis of thousands of power flows depending on the
number of contingencies. Furthermore several iterations
between SFT and SCUC may be needed to reach a secure
commitment solution. The amount of SFT computation is
tremendous. Since the MW dispatches for each of the 168
hours are available simultaneously from the SCUC solution,
the hundreds of thousands of SFT power flow analyses may be
performed in parallel to reduce the SCUC-RA problem
solution time. As system size increases and requirements
change, this system architecture would provide the capability
to scale the system by plugging in more processors to meet
performance requirements.
• Improve performance through efficient data management – For
the seven day SCUC RA analysis, the volume of data flowing
between SFT and SCUC is huge when hundreds of security
constraints may be detected. Generating the vast volume of
topology and security constraint sensitivity data not only takes
up computer disk resources and system support personnel’s
burden to manage the data, it also decreases the performance of
the analytical engines. To alleviate this problem, network
topology for each of the hours is identified in advance so that
hours with the same topology are grouped and one network
topology is produced for each group of the same topology
hours. This typically reduces the data volume dramatically.
As many RTO markets may grow in size and expand in market
products, performance, integration and manageability are the
critical system issues for the continual growth in size and
complexity of the markets. This SCUC-RA system architecture can
better meet these challenges.
With the integration of the transmission network security
analysis in the RA unit commitment, the SCUC commitment and
dispatch MW solutions from one iteration are fed to the SFT
application for security analysis. Network security violations for
each hour detected by SFT are formulated as sensitivity-based
constraints to be enforced in the next pass SCUC commitment
solution. This iteration continues until no new violation is found in
a commitment solution for any of the study hours. The process
guarantees a reliable and secure commitment solution with
minimum heuristics and operator’s intervention.
This SCUC schema can handle the circumstances where
detailed SFT analysis for all 168 hours in the RA study is required
and parallel SFT processing may become necessary. However,
flexibility will be built into the SCUC-RA system that allows the
operator to perform the RA commitment task as practically
needed:
•
•
•
Predefine security constraints for preemptive enforcement;
Perform SFT analysis for selected hours;
Control the iteration process between SFT and SCUC.
MIP SCUC
SFT
SFT
SFT
SFT
SFT
SFT
SFT
SFT
SFT
Hour 1
Hour 2
Hour n
Figure 1 - Parallel processing based SCUC architecture
IV. CONCLUSIONS
The unified framework for competitive electricity market and
grid reliability is founded on the duality theory that is concerned
with the outcome in a competitive market and the mathematical
solution to a constrained optimization problem. Experiences with
the implementation of several electricity markets suggest adoption
of formal optimization technology as one of the key factors for a
successful market. Significant progress has been made in recent
years in applications of formal optimization techniques for
competitive market based resource commitment, scheduling,
pricing and dispatch.
The SCUC and SCED algorithms are key components of a
successful electricity market operation system. Traditional
methods for the SCUC, such as the dynamic programming and the
more recent Lagrangian Relaxation methods are being frequently
challenged from ever-growing market demands for systematic and
transparent decision-makings in market operations. The recent
success in the application of formal optimization method, such as
the MIP, is significant in improving market efficiency and
enhancing grid reliability. The SCED algorithm is presently built
on the LP based optimization method. It provides the capability to
incorporate as many grid security constraints of linear format as
needed and guarantees the consistency between market clearing
quantities and prices. While the current SCED algorithm for
View publication stats
electricity market operation is mostly based on the DC grid model,
there is a clear trend that the voltage/var impacts on market
clearing quantities (in the form of MW-Mvar coupling) and prices
be appropriately considered. Extension of the existing
optimization-based SCED framework will be required.
V. REFERENCES
[1] D.I. Sun, B. Ashley, B. Brewer, B.A. Hughes, and W.F.
Tinney, "Optimal Power Flow by Newton Approach", IEEE
Transactions Vol. PAS-103, pp. 2864-2880, Oct. 1984.
[2] F.C. Schweppe, M.C. Caramanis, R.D. Tabors and R.E. Bohn,
Spot Pricing of Electricity, Kluwer Academic Publishers,
1988.
[3] M.L. Baughman and S.N. Siddiqi, “Real-Time Pricing of
Reactive Power: Theory and Case Study Results,” IEEE
Trans. on Power Syst., vol. 6, no. 1, pp.23-29, Feb. 1991.
[4] Check http://www.pjm.com for PJM market.
[5] Check http://www.iso-ne.com for ISO-NE market.
[6] A.A. El-Keib and X. Ma, “Calculating Short-Run Marginal
Costs of Active and Reactive Power Production,” IEEE
Trans. on Power Syst., vol. 12, no. 2, pp. 559-565, May 1997.
[7] F. Zhuang, F. Galiana, “Towards a More Rigorous and
Practical Unit Commitment by Lagrangian Relaxation”, IEEE
Transactions on Power Syst., vol. 3, no. 2, pp. 763-773, May
1988.
[8] S. Wang, S. Shahidepour, D. Kirschen, S. Mokhtari, “ShortTerm Generation Scheduling with Transmission and
Environmental Constraints using Augmented Lagrangian
Relaxation”, IEEE Transactions on Power Syst., vol. 10, no.
3, pp. 1294-1301, Aug. 1995.
[9] F. Lee, “A Fuel-Constrained Unit Commitment Method”,
IEEE Trans. on Power Syst., vol. 4, no. 3, pp. 1208-1218,
Aug. 1989.
[10] D. Streiffert, R. Philbrick, Andrew Ott “Mixed Integer
Programming Solution for Market Clearing and Reliability
Analysis”, to be presented at IEEE PES General Meeting,
June 12-16, 2005.
[11] H. Ma, S. Shahidehpour, “Unit Commitment with
Transmission Security and Voltage Constraints” IEEE Trans.
on Power Syst., vol. 14, no. 2, pp. 757-764, May 1999.
VI. BIOGRAPHIES
David I. Sun joined AREVA T&D Inc. in June 1980. He received
his B.S. and M.S. from Rensselaer Polytechnic Institute, and his
Ph.D. from University of Texas at Arlington, in 1974, 1976, and
1980 respectively, all in Electrical Engineering. His current focus
is on the planning and development of deregulation applications.
Xingwang Ma received his B.S. from Hefei University of
Technology, China and his M.S. from the Graduate School, EPRI,
Beijing, China in 1983 and 1985 respectively. He also received a
M.S. in Electrical Engineering from The University of Alabama in
1995. He has been associated with EPRI(China) and ABB Systems
Control. He joined AREVA T&D Inc. in 1996. He has been
involved in the design and implementation of several electricity
market systems.
Kwok W. Cheung received the BS degree from National Cheng
Kung University, Taiwan, in 1986, the MS degree from University
of Texas at Arlington, in 1988, and the Ph.D. degree from
Rensselaer Polytechnic Institute, Troy, NY in 1991. He joined
AREVA T&D Inc. in 1991. His current focus is on the
development of deregulation applications and market systems.
Formatted: English (U.S.)
Formatted: English (U.S.)
Formatted: English (U.S.)