Survey
* Your assessment is very important for improving the workof artificial intelligence, which forms the content of this project
* Your assessment is very important for improving the workof artificial intelligence, which forms the content of this project
Genetic algorithm wikipedia , lookup
Computational complexity theory wikipedia , lookup
Knapsack problem wikipedia , lookup
Inverse problem wikipedia , lookup
Dirac bracket wikipedia , lookup
Financial economics wikipedia , lookup
Multi-objective optimization wikipedia , lookup
Perfect competition wikipedia , lookup
Simplex algorithm wikipedia , lookup
Multiple-criteria decision analysis wikipedia , lookup
PRICE-BASED MARKET CLEARING UNDER MARGINAL PRICING: A BILEVEL PROGRAMMING APPROACH Ricardo Fernández-Blanco José M. Arroyo Natalia Alguacil Universidad de Castilla – La Mancha Ciudad Real, Spain [email protected] Universidad de Castilla – La Mancha Ciudad Real, Spain [email protected] Universidad de Castilla – La Mancha Ciudad Real, Spain [email protected] Pedro J. Muñoz Universidad de Castilla – La Mancha Ciudad Real, Spain [email protected] Abstract – Market clearing in restructured power systems is mostly implemented through an offer-based setting with the goal of maximizing social welfare. This approach leads to sound results from an economic viewpoint when generation offers reflect true production costs. However, offers may significantly differ from actual costs in practice, thus yielding undesired distortion. Under a marginal pricing scheme, this paper presents a general bilevel programming formulation for alternative market-clearing procedures dependent on market-clearing prices rather than on offers. The resulting nonlinear mixed-integer bilevel programming formulation is transformed into an equivalent single-level mixed-integer linear program suitable for efficient off-the-shell software. The bilevel formulation is investigated through a particular instance of price-based market clearing driven by consumer payment minimization. This problem has recently received considerable attention due to the open challenges posed from both modeling and computational perspectives. Numerical results are provided to illustrate the performance of the proposed approach. Keywords: Offer cost minimization, bilevel programming, marginal pricing, consumer payment minimization, price-based market-clearing procedure 1 INTRODUCTION In the framework of the nowadays competitive power industry this paper considers a pool-based electricity market for energy. The Independent System Operator (ISO) receives energy offers from producers and energy bids from consumers, and determines, for every hour, the market-clearing price, the power productions, and the consumption levels. The market-clearing procedure used by the ISO is typically formulated as a unit commitment problem [1] in which generation offers replace production costs and consumption bids are taken into account. In this problem, the objective function to be maximized is referred to as the declared social welfare. When the demand is inelastic, the objective function only includes generation offer information and the resulting problem becomes an offer cost minimization. Market-clearing prices resulting from the unit commitment problem are used for market settlement. 17th Power Systems Computation Conference Under the assumption of perfect competition, the optimal market-clearing solution is equal to the equilibrium solution where the marginal value to consumers is equal to the marginal cost to producers. This equilibrium maximizes the social welfare, which is the sum of the consumer surplus and the producer surplus. As a consequence, suppliers have no incentive to offer different from their production costs. Thus, the declared social welfare based on generation offers reflects the true social welfare and maximizing social welfare is commonly accepted as the right goal in a market setting. However, market-clearing procedures based on declared social welfare maximization are characterized by several practical shortcomings [2]: - The assumption of perfect competition does not hold in real-life power markets. Consequently, generation offers may not reflect actual costs and the optimization might not maximize the true social welfare or accurately reflect the natural behavior of market participants. - The scoring rule, which is used to select the accepted offers and bids, and the payment rule are different. While the former is based on offer cost minimization, the latter is determined in terms of market-clearing prices. This discrepancy leads to distortion since the solution does not necessarily constitute an equilibrium where the objectives of market participants are maximized. This paper analyzes an alternative type of marketclearing procedure for energy trading in an electricity pool. The alternative market clearing allows considering the payment rule through the incorporation of marketclearing prices in the problem formulation. This procedure is hereinafter referred to as price-based market clearing. Examples of price-based market clearing are (i) procedures based on consumer payment minimization [3]-[6], where market-clearing prices appear in the objective function, and (ii) revenue-constrained marketclearing procedures [3], [4], where market-clearing prices appear in the constraints. Moreover, new procedures explicitly including generation and consumer surplus maximization [7] would fall within this class of problems. Stockholm Sweden - August 22-26, 2011 The presence of market-clearing prices in the problem formulation complicates the solution of the optimization problem since market-clearing prices may themselves result from an optimization process. Several definitions for the market-clearing price are available [1], being the two most commonly adopted (i) the price of the highest accepted generation offer, which guarantees revenue adequacy for suppliers, and (ii) the marginal price, which provides appropriate economic signals. While the former has received considerable attention in price-based market clearing [4], [5], and references therein, few works have considered marginal pricing [3], [6], [7]. In [3], payment minimization under marginal pricing was addressed by a heuristic approach. In [6], Zhao et al. first formulated consumer payment minimization under marginal pricing as a bilevel programming problem, which was solved by augmented Lagrangian relaxation and surrogate optimization. In [7], agent-based simulation was used to analyze an auction driven by consumer and generation surplus maximization based on marginal market-clearing prices. Within the framework of marginal pricing, the main purpose of this paper is to formulate a general pricebased market-clearing procedure as a bilevel programming problem [8], of which the consumer payment minimization reported in [6] is a special case. Bilevel programming is appropriate to model problems where one agent, the leader, optimizes its objective function (upper-level problem) considering that a second agent, the follower, will react by optimizing its own objective function (lower-level problem). These models are relevant in those situations where the actions of the follower affect the decision making of the leader. This is the case in price-based market clearing: the selection of accepted bids and offers (upper-level problem) depends on market-clearing prices (lower-level problem), which are in turn determined based on the set of accepted bids and offers. Under marginal pricing, market-clearing prices are the Lagrange multipliers or dual variables associated with the power balance equations in an economic dispatch problem [1]. Another salient feature of this paper with respect to [6] is the proposal of an equivalent single-level mixedinteger linear formulation based on the Karush-KuhnTucker (KKT) optimality conditions, duality theory, and integer algebra results. The main advantages of expressing the original bilevel optimization as an equivalent single-level mixed-integer linear program is the guaranteed convergence to the optimal solution in a finite number of steps and the ready availability of efficient commercial branch-and-cut software. The main contributions of this paper are threefold: - The ISO is provided with an optimization-based framework for market clearing explicitly accounting for marginal market-clearing prices in the problem formulation. - Bilevel programming is proposed as a suitable solution methodology. 17th Power Systems Computation Conference - The applicability of bilevel programming is illustrated with the solution of an instance of pricebased market clearing, namely the consumer payment minimization problem. The remainder of the paper is organized as follows. Section 2 provides a general formulation for price-based market clearing under marginal pricing. Section 3 presents the solution methodology. Section 4 describes the application of the proposed framework for consumer payment minimization. In Section 5, numerical results illustrate the performance of the proposed approach. Finally, relevant conclusions are drawn in Section 6. 2 PROBLEM FORMULATION The price-based market-clearing procedure can be formulated in a compact way as: Min F(x, y, λ ) (1) x , y ,λ subject to: x∈X G e (x, y ) = 0 (2) (3) G i (x , y ) ≥ 0 (4) (5) H(x, y, λ ) ≥ 0 , where x represents the vector of scheduling variables (status of generating units, start-up costs, etc.), y is the vector of dispatching variables (generation levels, network-related variables, etc.), λ is the vector of marketclearing prices, F(x, y, λ ) is the objective function (declared social welfare, consumer payment, consumer and generation surplus, etc.), X is the feasibility set for scheduling variables x (integrality constraints, start-up cost constraints, minimum up and down times, etc.), G e (x , y ) denotes the equalities associated with power balance equations, G i (x, y ) models all inequality constraints involving variables x and y (power limits, ramping limits, etc.), and H(x , y, λ ) represents all remaining inequality constraints involving market-clearing prices such as those imposing revenue adequacy. Under marginal pricing, the vector of marketclearing prices λ is defined as the vector of marginal costs. Mathematically, λ can be derived from the following economic dispatch problem for the optimum vector of scheduling variables x * : ( Min f x * , y y ) (6) subject to: ( ) G (x , y ) ≥ 0 : (μ ) , G e x * , y = 0 : (λ ) i * (7) (8) Stockholm Sweden - August 22-26, 2011 where f (⋅) is the objective function, which may be different from that of the price-based market-clearing procedure F(⋅) ; the vector of market-clearing prices λ is equal to the vector of Lagrange multipliers or dual variables associated with the power balance equations (7); and µ is equal to the vector of Lagrange multipliers or dual variables associated with constraints (8). Note that under marginal pricing, the definition of λ is an optimization problem itself. Thus, the price-based market-clearing procedure can be recast as the following bilevel programming problem: ( * * Min F x , y , λ x ) (9) subject to: x∈X ( ) H x , y * , λ* ≥ 0 , * (10) (11) * where y and λ are obtained from: ( Min f x * , y y ) ( ) G (x , y ) ≥ 0 : (μ ) . i * ( ) ( ) ( G (x , y ). L x * , y, λ, μ = f x * , y − λT G e x * , y −μ T i ) * (15) The optimal solution to (12)-(14) must satisfy the KKT necessary optimality conditions: ( ) ( ) ( ∂L x * , y, λ, μ ∂f x * , y ⎛⎜ ∂G e x * , y = − ⎜ ∂y ∂y ∂y ⎝ ⎛ ∂G i x * , y −⎜ ⎜ ∂y ⎝ (12) )⎞⎟ ⎟ ⎠ )⎞⎟ ⎟ ⎠ T λ T μ=0 (16) μ≥0 (17) e * (18) i * T i ( ) G (x , y ) ≥ 0 μ G (x , y ) = 0 , G x ,y = 0 (13) (14) The above bilevel problem consists of an upper-level optimization (9)-(11) and a lower-level optimization (12)-(14). The upper level controls the vector of scheduling variables x. Lower-level decision variables comprise the vector of dispatching variables y whereas the vector of market-clearing prices λ is also associated with the lower-level problem. The goal of the upper level is to minimize the pricebased market-clearing objective function (9) evaluated at the optimal values of the lower-level variables, y * and λ* , subject to expressions exclusively constraining upper-level decision variables x (10), to a set of constrained functions with y and λ as parameters (11), and to the lower-level optimization (12)-(14). The lower-level optimization is identical to the economic dispatch problem (6)-(8) associated with the optimal upper-level decision variables x * . In general, problem (9)-(14) is a mixed-integer nonlinear bilevel programming problem. It is worth noting that the lower-level objective function (12) is typically piecewise linear whereas constraints (13)-(14), including power balance equations as well as generation and network limits, are usually linear. As a consequence, the lower-level problem (12)-(14) is parameterized in terms of the upper-level decision vector x in such a way that the lower-level problem is linear and thus convex. As 17th Power Systems Computation Conference 3 SOLUTION APPROACH To convert the original bilevel formulation (9)-(14) into an equivalent single-level problem, the lower-level optimization is first replaced by its KKT conditions. Consider the Lagrangian function associated with the lower-level problem for a given upper-level vector x * : ( subject to: G e x * , y = 0 : (λ ) described next, this feature allows the transformation of problem (9)-(14) into an equivalent single-level mixedinteger problem. (19) * (20) where (16)-(17) are the dual feasibility constraints, (18)-(19) are the primal feasibility constraints, and (20) express the complementary slackness conditions. Hence, problem (9)-(14) can be recast as the following single-level equivalent: Min F(x , y, λ ) (21) x , y , λ ,μ subject to: x∈X H(x, y, λ ) ≥ 0 ∂f (x , y ) ∂y μ≥0 (22) (23) T T i ⎛ ∂G e (x, y ) ⎞ ⎛ ⎞ ⎟ λ − ⎜ ∂G (x, y ) ⎟ μ = 0 −⎜ ⎜ ⎟ ⎜ ⎟ ∂y ∂y ⎝ ⎠ ⎝ ⎠ (24) (25) G (x , y ) = 0 (26) G (x, y ) ≥ 0 (27) μ G (x , y ) = 0 . (28) e i T i Problem (21)-(28) falls into the category of what is known in the literature as a mathematical program with equilibrium constraints (MPEC) [8]. As can be seen, the vectors of Lagrange multipliers λ and µ become decision variables of the resulting single-level equivalent. In Stockholm Sweden - August 22-26, 2011 addition, the following types of nonlinearities are present in problem (21)-(28): - Nonlinear expressions (23) typically involving products of market-clearing prices and power outputs. These sets of nonlinear products can be transformed into equivalent mixed-integer linear programming expressions based on the KKT conditions as described in [9]. - Products of scheduling variables x, typically binary, and continuous variables λ and µ in (24). These nonlinearities can be equivalently formulated as linear expressions using well-known integer algebra results [10]. - Products of Lagrange multipliers and lower-level decision variables y in the complementary slackness conditions (28). As shown by Fortuny-Amat and McCarl [11], complementary slackness conditions can also be formulated as mixed-integer linear programming expressions. It is worth mentioning that under the assumption of linearity of the lower-level problem, complementary slackness conditions (28) can be replaced by the equality associated with the strong duality theorem [10], thereby leading to a more effective linearization in terms of computational performance. Thus, after some algebra, problem (21)-(28) is recast as a mixed-integer linear programming problem suitable for commercially available branch-and-cut software. 4 APPLICATION To illustrate the above bilevel programming framework, we now consider an instance of price-based market clearing, namely the payment minimization problem presented in [5], hereinafter referred to as PM. As done in [5], we assume that (i) the demand is inelastic, (ii) generation offers comprise a single block, and (iii) minimum up and down times, ramping limits, transmission network, and ancillary services are ignored. Notwithstanding, these simplifications do not alter our main conclusion, and results could be extended to include all of them at the expense of an increased notational complexity. For quick reference, PM is formulated as: Minimize p jk , s jk , v jk ,λ k ∑λ k Dk + k∈K ∑∑ s jk (29) j∈J k∈K subject to: ⎛ ∂⎜ ⎜ ⎝ λk = ∑∑ j∈J ) is the start-up cost of unit j in period k; v jk is a binary variable that is equal to 1 if unit j is scheduled on in period k, being 0 otherwise; λ k is the market-clearing price in period k; K is the index set of time periods; D k is the demand in period k; J is the index set of generatis the coefficient of the start-up cost of ing units; C SU j unit j; P jk and Pjk are the lower and upper bounds for the power output of unit j in period k, respectively; and C Pjk represents the offer cost coefficient of unit j in period k. The objective function (29) represents the total payment by consumers and comprises two terms: the payment for energy and the payment for generation startups. Constraints (30) and (31) model the start-up costs. Constraints (32) impose the integrality of variables v jk . Constraints (33) are the power balance equations. Constraints (34) set the lower and upper bounds for the power outputs. Finally, constraints (35) model the marginal pricing setting. For the sake of unit consistency, hourly time periods are considered. According to Section 2, the bilevel programming formulation for problem (29)-(35) is: Minimize s jk , v jk ∑∑ s + jk (36) j∈J k∈K ( ) s jk ≥ C SU v jk − v j,k −1 ; ∀j ∈ J, ∀k ∈ K j (37) s jk ≥ 0; ∀j ∈ J, ∀k ∈ K (38) v jk ∈ {0,1}; ∀j ∈ J, ∀k ∈ K , (39) where market-clearing prices λ k are obtained from: p jk (31) subject to: v jk ∈ {0,1}; ∀j ∈ J, ∀k ∈ K (32) ∑ (33) ∑p j∈J jk ∑∑ p P jk C jk (40) j∈J k∈K s jk ≥ 0; ∀j ∈ J, ∀k ∈ K = D k : (λ k ) ; ∀k ∈ K ( (41) ) v jk P jk ≤ p jk ≤ v jk Pjk : γ jk , θ jk ; ∀j ∈ J, ∀k ∈ K, (42) j∈J 17th Power Systems Computation Conference k Dk subject to: (30) v jk P jk ≤ p jk ≤ v jk Pjk ; ∀j ∈ J, ∀k ∈ K ∑λ k∈K s jk ≥ C SU v jk − v j,k −1 ; ∀j ∈ J, ∀k ∈ K j p jk = D k ; ∀k ∈ K (35) where p jk is the power output of unit j in period k; s jk Minimize ( ⎞ p jk C Pjk ⎟ ⎟ k∈K ⎠ ; ∀k ∈ K , ∂D k (34) Stockholm Sweden - August 22-26, 2011 and where γ jk and θ jk are the Lagrange multipliers per bound for the dual variable γ jk , and θ jk is the associated with the lower and upper bounds for the power output of unit j in period k, respectively. The upper-level optimization (36)-(39) determines the scheduling variables v jk and the corresponding lower bound for the dual variable θ jk . start-up costs s jk that minimize the consumer payment. In contrast, the lower-level problem (40)-(42) determines generation power outputs p jk and the marketclearing prices λ k associated with the upper-level scheduling variables v jk by solving a multiperiod economic dispatch based on the minimization of the generation offer cost. Applying the methodology described in Section 3, the bilevel programming problem (36)-(42) is transformed into the following equivalent single-level mixed-integer linear program: ∑ λ D + ∑∑ s Minimize k a jk ,a Ajk , b jk , b Ajk , p jk , k∈K s jk , v jk , γ jk ,θ jk ,λ k k jk (43) j∈J k∈K subject to: s jk ≥ C SU v jk − v j,k −1 ; ∀j ∈ J, ∀k ∈ K j ( ) (44) s jk ≥ 0; ∀j ∈ J, ∀k ∈ K (45) v jk ∈ {0,1}; ∀j ∈ J, ∀k ∈ K γ jk + θ jk + λ k = C Pjk ; (46) ∀j ∈ J, ∀k ∈ K (47) γ jk ≥ 0; ∀j ∈ J, ∀k ∈ K (48) θ jk ≤ 0; ∀j ∈ J, ∀k ∈ K (49) ∑p (50) jk = D k ; ∀k ∈ K j∈J v jk P jk ≤ p jk ≤ v jk Pjk ; ∀j ∈ J, ∀k ∈ K ∑λ k Dk + k∈K ∑∑ a j∈J k∈K ∑∑b (51) ∑∑p (52) a Ajk = γ jk − a jk ; ∀j ∈ J, ∀k ∈ K (53) 0 ≤ a jk ≤ v jk γ jk ; ∀j ∈ J, ∀k ∈ K (54) 0 ≤ a Ajk ≤ 1 − v jk γ jk ; ∀j ∈ J, ∀k ∈ K (55) b Ajk = θ jk − b jk ; ∀j ∈ J, ∀k ∈ K (56) v jk θ jk ≤ b jk ≤ 0; ∀j ∈ J, ∀k ∈ K (57) jk Pjk j∈J k∈K ( = tively. 5 NUMERICAL RESULTS The bilevel programming framework for PM has been tested on two case studies described in [5]. For both cases, the results of PM are compared with those achieved by a conventional market-clearing procedure based on the minimization of the sum of the generation offer costs and start-up costs. This cost minimization problem is denoted as CM. The model has been implemented on an Intel Core i7, 1.73-GHz processor with 8 GB of RAM using Cplex 12.0 under GAMS. 5.1 Illustrative Example The first example considers two hours and four generating units which are initially scheduled off. Table 1 shows the data for both periods. This small example is useful to highlight the differences in the results yielded by PM and CM. D1 = 100 MW Hour 1 jk P jk P jk C jk + Expressions (43)-(46) correspond to the upper-level problem while constraints (47)-(58) equivalently replace the lower-level problem. Constraints (47)-(49) are the dual feasibility constraints, (50)-(51) represent the primal feasibility constraints, and (52)-(58) are related to the linearization of the strong duality equality. Constraint (52) is the linear expression where the objective functions of the primal and dual problems are equated. Constraints (53)-(55) and (56)-(58) model the linearization of the product terms v jk γ jk and v jk θ jk , respec- Unit 1 2 3 4 P j1 [MW] 5 5 0 5 j∈J k∈K ) (1 − v jk ) θ jk ≤ b Ajk ≤ 0; ∀j ∈ J, ∀k ∈ K , (58) where a jk represents the nonlinear product v jk γ jk , b jk represents the nonlinear product v jk θ jk , a Ajk and b Ajk are auxiliary continuous variables used in the lin- earization of the above nonlinear terms, γ jk is the up- 17th Power Systems Computation Conference Hour 2 Unit 1 2 3 4 C Pj1 Pj1 [MW] [$/MWh] 50 10 40 20 10 65 60 30 D 2 = 150 MW C SU j [$] 0000 0000 0050 1800 P j2 P j2 C Pj2 C SU j [MW] 5 5 0 5 [MW] 060 060 030 100 [$/MWh] 15 20 65 30 [$] 0000 0000 0050 1800 Table 1: Data for the illustrative example. For this illustrative example, the computing time required to achieve the optimal solution to PM using the single-level equivalent was less than 1 s. The optimal solutions to CM and PM are summarized in Table 2. Under conventional market clearing, generators 1, 2, and 3 are dispatched at maximum capacity in both periods whereas generator 4 is not sched- Stockholm Sweden - August 22-26, 2011 uled. Generator 3 is the marginal unit in both periods thereby setting the corresponding market-clearing prices equal to its offer cost, i.e., $65/MWh. This solution costs $6050 and yields a payment equal to $16300. Unit p jk [MW] 1 2 3 4 λ k [$/MWh] Offer Cost + Start-up Cost [$] Payment [$] CM Hour 1 2 50 60 40 60 10 30 00 00 65 65 6050 16300 PM Hour 1 2 50 60 40 60 00 00 10 30 30 30 6400 9300 tion in payment is achieved by slightly increasing the cost by 2.6%. Hour 01 02 03 04 05 06 07 08 09 10 11 12 D k [MW] 2500 2550 2570 2530 2650 2700 2680 2740 2850 2900 3500 3700 Hour 13 14 15 16 17 18 19 20 21 22 23 24 D k [MW] 3800 3900 4200 4250 4400 4500 4100 3850 3400 2700 2300 2350 Table 2: Results for the illustrative example. In contrast, PM results in a different schedule in both periods. While generators 1 and 2 do not experience any change in their generation levels with respect to the solution to CM, generators 3 and 4 exchange their schedules and power dispatches. As a consequence, generator 4 becomes the marginal unit in both periods, setting both market-clearing prices at $30/MWh. The optimal payment is equal to $9300 and the associated cost is equal to $6400. In other words, a 42.9% reduction in payment is attained at the expense of a 5.8% increase in cost. In this case, the market-clearing price in each period under marginal pricing is identical to the corresponding highest accepted offer. Note, however, that both pricing schemes lead in general to different market-clearing prices irrespective of the objective function being minimized. As an example, let the demand at hour 1 be reduced so that it belongs to the interval (50, 55] MW. At the optimal solution to either CM or PM, generator 2 would be dispatched at its minimum power output while generator 1 would be the marginal unit by supplying the remaining demand. Under marginal pricing, the marketclearing price would be the offer cost of generator 1, i.e., $10/MWh, whereas the highest accepted offer is that of generator 2, i.e., $20/MWh. 5.2 Medium-Sized Test Case The second case study considers 25 generating units and 24 hours. The hourly system demand is shown in Table 3. Data for generators are presented in Table 4. It is assumed that generators do not modify their respective supply offers over the time span. In addition, all units are initially scheduled off except units 1-8. The execution of Cplex was stopped when the value of the payment was below a specified threshold or when this number appeared to have reached a lower bound. With these stopping criteria, Table 5 shows the results attained by the proposed approach for PM. The computing time required by this solution was equal to 78 s. Table 5 also lists the results corresponding to the optimal solution to CM. As can be seen, a 7.0% reduc- 17th Power Systems Computation Conference Table 3: System demand for the medium-sized test case. Unit 01 02 03 04 05 06 07 08 09 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 P jk Pjk C Pjk C SU j [MW] 100 060 050 030 040 040 030 030 020 020 020 020 020 020 020 020 020 020 020 020 020 020 020 015 010 [MW] 455 350 300 200 350 320 240 200 190 180 170 160 150 140 130 125 120 115 110 120 115 110 100 090 080 [$/MWh] 30 34 35 37 40 42 45 47 55 57 58 59 60 62 63 65 66 68 70 75 78 80 90 93 95 [$] 1200 1150 1100 1000 0350 0350 0300 0300 0180 0180 0175 0160 0250 0200 0150 0180 0140 0140 0160 0250 0250 0300 0050 0050 0040 Table 4: Generation data for the medium-sized test case. 0Payment [$] Offer Cost + Start-up Cost [$] PM 4764845 3482645 CM 5122905 3394415 Table 5: Results for the medium-sized test case. Stockholm Sweden - August 22-26, 2011 100 8 90 7 80 6 70 5 60 4 50 System demand (GW) Market-clearing price ($/MWh) Hourly market-clearing prices associated with the solutions presented in Table 5 are depicted in Fig. 1. This figure also shows the hourly system demand. As expected, market-clearing prices resulting from either PM or CM follow the shape of the demand curve. Note, however, that market-clearing prices corresponding to PM are less than or equal to those resulting from CM for all periods. 3 40 30 2 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 Hour Market-clearing price for CM Market-clearing price for PM System demand Figure 1: Market-clearing prices and system demand for the medium-sized test case. 6 CONCLUSION This paper has formulated the price-based marketclearing problem under marginal pricing as a general bilevel programming problem, offering flexibility in the problem definition. The resulting bilevel programming formulation is transformed into an equivalent singlelevel mixed-integer linear program. This transformation comprises two steps. First, the lower-level optimization is replaced by a set of constraints based on the KKT optimality conditions and duality theory. Subsequently, a number of nonlinearities are converted to linear equivalents using some well-known integer algebra results. The ultimate goal of this paper is to provide the ISO with a tool that can be used for comparative analysis of different market-clearing procedures, so that informed decisions can be made. This general bilevel formulation and its single-level equivalent have been applied to an instance of pricebased market clearing in which consumer payment is minimized. Numerical results reveal that bilevel programming is an effective approach to address pricebased market-clearing procedures. Research is currently underway to solve other instances of price-based market clearing such as those explicitly maximizing the surplus of consumers and producers, as well as market-clearing procedures with revenue adequacy constraints. Finally, further research will also be devoted to the analysis of joint energy and reserve electricity markets. 17th Power Systems Computation Conference Powered by TCPDF (www.tcpdf.org) ACKNOWLEDGMENT The authors acknowledge the support from the Ministry of Science of Spain under project ENE200907836, and from the Junta de Comunidades de Castilla – La Mancha under project PAI08-0077-6243. REFERENCES [1] M. Shahidehpour, H. Yamin and Z. Li, “Market Operations in Electric Power Systems”, New York, Wiley, 2002, ISBN 0-471-44337-9 [2] J. M. Jacobs, “Artificial Power Markets and Unintended Consequences”, IEEE Transactions on Power Systems, vol. 12, no. 2, pp 968-972, May 1997 [3] J. Alonso, A. Trías, V. Gaitan and J. J. Alba, “Thermal Plant Bids and Market Clearing in an Electricity Pool. Minimization of Costs vs. Minimization of Consumer Payments”, IEEE Transactions on Power Systems, vol. 14, no. 4, pp 1327-1334, November 1999 [4] C. Vázquez, M. Rivier and I. J. Pérez-Arriaga, “Production Cost Minimization versus Consumer Payment Minimization in Electricity Pools”, IEEE Transactions on Power Systems, vol. 17, no. 1, pp 119-127, February 2002 [5] P. B. Luh, W. E. Blankson, Y. Chen, J. H. Yan, G. A. Stern, S.-C. Chang and F. Zhao, “Payment Cost Minimization Auction for Deregulated Electricity Markets Using Surrogate Optimization”, IEEE Transactions on Power Systems, vol. 21, no. 2, pp 568-578, May 2006 [6] F. Zhao, P. B. Luh, J. H. Yan, G. A. Stern and S.-C. Chang, “Payment Cost Minimization Auction for Deregulated Electricity Markets with Transmission Capacity Constraints”, IEEE Transactions on Power Systems, vol. 23, no. 2, pp 532-544, May 2008 [7] A. Somani and L. Tesfatsion, “An Agent-Based Test Bed Study on Wholesale Power Market Performance Measures”, IEEE Computational Intelligence Magazine, vol. 3, no. 4, pp 56-72, November 2008 [8] S. Dempe, “Foundations of Bilevel Programming”, Norwell, Kluwer Academic Publishers, 2002, ISBN 1-4020-0631-4 [9] C. Ruiz and A. J. Conejo, “Pool Strategy of a Producer with Endogenous Formation of Locational Marginal Prices”, IEEE Transactions on Power Systems, vol. 24, no. 4, pp 1855-1866, November 2009 [10]C. A. Floudas, “Nonlinear and Mixed-Integer Optimization: Fundamentals and Applications”, New York, Oxford University Press, 1995, ISBN 0-19510056-3 [11]J. Fortuny and B. McCarl, “A Representation and Economic Interpretation of a Two-Level Programming Problem”, Journal of the Operational Research Society, vol. 32, no. 9, pp 783-792, September 1981 Stockholm Sweden - August 22-26, 2011