Tailoring MIP-based Generation Scheduling Program for Korean Electricity Market

Transcription

1 Tailoring MIP-based Generation Scheduling Program for Korean Electricity Market D. HUR *, H. S. JEONG ** Department of Electrical Engineering Kwangwoon University Kwangwoon Rd. 26, Nowon-gu, Seoul, REPUBLIC OF KOREA * dhur@kw.ac.kr, ** econohs@naver.com Abstract: Though the Lagrangian Relaxation (LR)-based Resource Scheduling and Commitment (RSC) was launched to optimize generation scheduling in Korean electricity market, it is chocked up with demerits. Hence, a robust and upgradeable generation scheduling application needs to be utilized for the purpose of serving diverse and challenging business requirements. This paper primarily sets out to provide efficient scheduling and coordination of multiple resources to adapt to new requirements while recognizing the numerous operational constraints such as energy, reserve, and transmission network constraints and allowing for more sophisticated modeling of pumped-storage facilities and combined cycle plants based on Mixed Integer Programming (MIP), which leads to fast and low-cost reliable solutions in the unit commitment problem. Key-Words: Combined cycle plant, Electricity market, Generation scheduling, Lagrangian relaxation, Mixed integer programming, Pumped-storage facility, Reserve 1 Introduction The unit commitment and real-time dispatch of generating facilities are the essential task of Korea Power Exchange (KPX). In recent years, the annual generation costs in Korean electric power system account for approximately 15 thousand million dollars due to rapid load growth and high fuel costs. Accordingly, the role of KPX to promote the efficiency of power systems by adjusting generator outputs and provide market participants with a reasonable price signal by yielding correct market prices becomes very important [1]. Now KPX establishes the generation scheduling and dispatch plan in Fig. 1 using the Resource Scheduling and Commitment (RSC), which was developed by ALSTOM-ESCA (currently, AREVA) and delivered to KPX in 1999 [2]. Since this software was adopted to be used in the generation competition scheme for 2 years or so, neither its functionalities nor the optimization solver has been improved. Simply, the user interface was partly upgraded. As the present market structure is expected to last for the next few years, more advanced generation scheduling program is necessarily needed to support users in the course of rapid changes in market environment and deregulation. Fig. 1. Trading processes in Korean electricity market Since the formerly RSC is based on Lagrangian Relaxation (LR) and dynamic priority order sequential bidding [3]-[4], it has some critical problems in terms of optimization. The water availability of hydro units and the inter-temporal constraints such as energy constraints of generating facilities are not actually considered. There is no pumped-storage model. Typically, pumped-storage units must generate in a high market price while they have to pump in a low market price. Since only one ramp rate is applied, various start-up or shut-down ramp rates are not taken into account. Once transmission line and/or energy constraints are reflected, the optimization results are not reliable. In particular, it would be easy to get sub-optimal unit ISSN: ISBN:

2 commitment solutions when optimizing with the must run or fixed generation constraints. There is no reserve model. The quantity of reserves offered by individual units and the relationship with pre-defined reserves in the market operation rule must be modeled before the optimization. In Korea, the peaker units are, on the whole, combined cycle units. Thus, gas turbines and steam turbines are definitely separated. Furthermore, the combined cycle model must allow combustion turbines to generate when steam turbines are not generating. The power system operations tend to repeat in a weekly basis. In this perspective, the weekly generation scheduling must be executed within a specified simulation time. For all the reasons given previously, it is apparent that a new flexible generation scheduling program should be suggested to aggressively deal with uncertain market situations in Korea. 2 Model and Functionalities This section will describe the functionalities of Korean generation scheduling program (GSP) as well as its models which were customized from AREVA s e-terracommit to fit Korean unique market conditions [5]. 2.1 Optimization model As a resources scheduling and commitment program, Korean GSP aims at: The minimization of the sum of generation costs and reserve costs (co-optimization between energy and reserve costs) less the demand bid revenue, subject to: - Area reserve requirement; - Regional areas power balance and reserve requirements; - Tie-line (between regional areas) flow constraints; - Units physical, operational and economical constraints; - Demand bid optimal scheduling; - Pumped-storage hydro plants and units constraints; - Group of units energy limits constraints; - Generic constraints The Korean GSP problem is formulated as a constrained optimization unit commitment problem [6]. The objective function and the constraints of this optimization problem are formulated as a mix of binary and continuous variables but always linear. Hence, this constrained optimization problem will be solved by the Mixed Integer Programming (MIP)-based method. 2.2 Tie-line model A tie-line is a physical or logical transmission line or a group of transmission lines that connect two zones. Each tie-line has a maximum export tie flow (MW). Flows can be only positive, corresponding to a MW export (physical flow from Zone 1 to Zone 2). As a consequence, for bi-directional inter-zonal interchanges, a tie-line must be declared from Zone 1 to Zone 2, and a second one from Zone 2 to Zone 1. The tie-line maximum flow is defined as the maximum flow between two adjacent zones related by the tie-line. Also, a minimum flow limit can be defined for some hours to force for a minimum power transfer between two zones. A tie-line can be declared as open for some hours to exclude any power transfer between two zones. 2.3 Generation unit model Unit static parameter When committed and not in a start-up or shut-down phase, the unit must be dispatched higher than or equal to Economic Minimum Dispatch (EcoMn) level, in MW. When committed, the unit must be dispatched lower than or equal to Economic Maximum Dispatch (EcoMx) level, in MW. EcoMx must be superior or equal to EcoMn. Load Frequency Control Maximum (LfcMx) level is the maximum generation level that can be reached when regulating up and/or when calling the spinning and non-spinning reserve. LfcMx must be superior or equal to EcoMx. Current Generation Capacity (CurCap) corresponds to the maximum generation level that can be reached in emergency condition, when calling the operating reserve. CurCap must be superior or equal to LfcMx. Unit Penalty Factor is used to penalize unit costs for transmission losses. This penalty factor is incorporated in the objective function (cost with penalty factor is equal to unit cost multiplied by this factor), but is not used in the calculation of actual unit cost or the total system cost. This factor can be either higher or lower than 1. Dispatch Ramp Rate in MW/hour is applied when the unit is committed and not in a start-up or ISSN: ISBN:

3 shut-down phase. The same ramp rate is applied for up and down ramping Unit commitment constraints The following constraints are covered by Korean GSP when solving the unit commitment problem: The minimum up time for a unit is the minimum time between the start-up dispatch command and the next shut-down of the unit (time of unit s circuit breaker opening). The minimum down time for a unit is the minimum time between the shut-down of the unit (time of unit s circuit breaker opening) and the next start-up dispatch command. If maximum number of starts per day and maximum number of unit starts per study are not null, these constraints limit the number of unit starts per day and the number of unit starts over the full study horizon, respectively. To apply correctly these minimum up and down time constraints, unit initial state, and more specifically the number of hours since the last unit commitment or decommitment, must be reported Unit start-up and shut-down ramp rates During the phases of start-up and shut-down, generating units are characterized by four distinct ramp rates: A shut-down ramp rate is applied during the unit shut-down phase (i.e. between EcoMn and 0 MW). Three static start-up ramp rates named hot start-up ramp rate, intermediate start-up ramp rate and cold start-up ramp rate are dependent on the heating condition of the boiler (hot, intermediate, and cold). These ramp rates are applied during the unit start-up phase (i.e. between 0 MW and EcoMn). The heating condition of a unit boiler depends on the number of hours the unit was off line and two other static parameters: Hot to intermediate time (HotToIntTime) and Hot to cold time (HotToCold- Time). The unit is considered in hot condition if the number of hours the unit was off-line is less than or equal to HotToIntTime; The unit is considered in intermediate condition if the number of hours the unit was off-line is higher than HotToIntTime and less than or equal to HotToCold- Time; The unit is considered in cold condition if the number of hours the unit was off-line is higher than HotToColdTime. Note that peaker and hydro units are not subject to the start-up and shut-down models since they have the flexibility to reach immediately the economic range. 2.4 Combined cycle model The combined cycle plant may include multiple combustion turbines (CTs) and steam turbines (STs). Each of the CTs produces MW output based on its incremental cost. Separate input-output curves for the STs are provided which give the total MW output of the ST units as a function of the total MW output of the CTs. CT and ST units of a combined cycle power plant are subject to the following constraints: The CT unit energy cost is modeled through the incremental cost (energy offer prices) as for other thermal units. The ST units do not have an explicit cost (no energy offer prices). ST unit MW is a linear inequality of the total MW output of the CTs as formulated below: Gen ST ( j, t) α GenCT ( i, t) (1) CT s and ST s are dispatched within their economic ranges (EcoMn Unit MW EcoMx). ST cannot be committed unless at least one CT in group is committed. 2.5 Unit energy costs Incremental cost curves are input for each period as part of the unit energy offer data. Energy costs for each study period are computed directly from the unit s incremental cost curve. Energy offer price curves from generators are assumed to be flat price bands as shown in Fig. 2. Piece-wise curves with slopes are not allowed. An energy offer can include up to 10 energy bands. Each band is described by its quantity (MW) and price (KRW/MWh). Offers are submitted through bands of MW breakpoint and price. The MW breakpoints are generation absolute values. Generation offers must be composed of monotonically increasing prices. Unit energy offers is defined between 0 MW and EcoMx and hence is used for the calculation of energy cost even for the start-up phase (between 0 MW and EcoMn) and the shut-down phase (between EcoMn and 0 MW). i ISSN: ISBN:

4 Fig. 2. Energy offer structure To calculate the unit generation cost, we use the unit energy offers as incremental cost curve representing the variation of generation cost with regard to the MW variation. The unit generation cost is then obtained as an integration of this cost curve per MW. To be complete, this calculation needs no load cost. Start-up cost is a generation cost to be added to the energy cost and which incurs each time the unit is synchronized. This cost is calculated according to the heating status of the boiler (hot, intermediate, or cold). 2.6 Demand bid model Ponds Demand bids are submitted on an hourly period basis, where purchasers specify the highest price they are prepared to pay for purchasing energy. When the problem is solved, a demand bid will be only dispatched if it is economic to do so. A demand bid can include up to 5 energy bands. Each band is described by its quantity (MW) and price (KRW/MWh). Demand bids are submitted through bands of MW values and price. The MW values are demand relative values. Demand bids must be composed of monotonically decreasing prices. 2.7 Pumped-storage model Ponds of pumped-storage plant are modeled through an energetic model, where storage data are expressed in MWh. The pumped-storage plant is characterized by: A minimum storage level is supposed to coincide with the water volume available in the pond when the minimum pond level is reached. A maximum storage level is supposed to coincide with the water volume available in the pond when the maximum pond level is reached. An initial storage level at the start of the study is supposed to be between the minimum and maximum storage levels. An ending (target) storage level at the end of the study is supposed to be between the minimum and maximum storage levels. A pumping efficiency factor represents the energy stored in the pond when consuming 1 MWh for pumping purpose. This factor is supposed to be between 0 and 1. A pumping unit gets the same generation limits as any other units (EcoMn, EcoMx, LfcMx, CurCap). In addition, a pumping unit is characterized by: When the unit is pumping, the pump level must be higher than or equal to the minimum pumping level, PumpMn. When the unit is pumping, the pump level must be lower than or equal to the maximum pumping level, PumpMx. Units belonging to pumped-storage plants are also subject to the following constraints: The same unit cannot be simultaneously generating and pumping. When a hydro unit is pumping, no unit from the same plant will generate. When a hydro unit is generating, no unit from the same plant will pump. 2.8 Energy constraints model A generating unit may have limited (maximum) available energy for the whole study period or for a particular day. Also, a unit may have minimum energy to consume again for the whole study period or for a particular day. Also, these minimum and maximum energy limits can apply to a group of units for the whole study period or for a particular day. 2.9 Reserve model Five types of reserves are defined in Korea. Reserve capacity available to satisfy a reserve requirement is also available to satisfy reserve requirements for any lower-quality category of reserves (i.e., higher quality reserves can substitute for lower quality reserves). The quality of reserves is ordered as follows (highest to lowest): Primary up Regulation up Spinning Non-spinning Operating. Each MW of a unit s capacity is cleared to provide only one category of product (i.e. energy, up-primary reserve, up-regulation reserve, spinning reserve, non-spinning reserve, or operating reserve). Thus, area reserve requirements, as well as unit capacities, are specified as incremental requirements. For instance, a unit with zero operating capacity and positive spinning capacity can provide its full spinning capacity to meet operating requirements. ISSN: ISBN:

5 Table 1. Reserve capabilities Resource Primary Regulation Spinning Non-spinning Operating reserve reserve reserve reserve reserve On-line units Off-line fast-start or hydro units Off-line peaker units On-line pumping units Off-line pumping units Date Generation Cost by LR-based RSC [ 10 3 KRW] Table 2. Comparison of generation costs respecting various scenarios Generation Cost by MIP-based generation scheduling program [ 10 3 KRW] Pumped- storage Excluded Hydro Optimized Pumped- storage Optimized Reserve constraints considered T/L constraints considered Jan. 26, ,039,249 55,953,385 55,883,237 55,645,449 56,691,916 56,327,820 Jan. 27, ,089,677 42,998,844 42,945,242 42,732,521 43,629,791 43,680,439 Jan. 28, ,761,964 59,635,992 59,449,901 59,212,170 60,777,578 59,573,423 Jan. 29, ,137,541 67,024,918 66,928,361 66,686,344 68,152,435 66,922,237 Jul. 03, ,594,192 53,426,365 53,336,730 53,045,725 54,260,915 53,733,320 Jul. 04, ,268,583 59,083,019 58,866,608 58,596,118 59,746,046 59,193,614 Jul. 05, ,675,803 49,529,555 49,144,313 48,867,614 49,898,069 49,687,035 Jul. 06, ,843,685 33,747,937 33,652,597 32,896,934 34,379,579 34,054,227 Date Jul. 15, 2008 Sept. 13, 2008 Table 3. Comparison of Generation Cost and SMP Respecting Various Scenarios MIP-based generation scheduling program Item LR-based RSC Pumped- storage Excluded Hydro Optimized Combined Cycle Optimized Generation Cost [ 10 6 KRW] 68,542 68,341 68,280 68,283 System Marginal Price [KRW/kWh] Number of On-line Units 3,456 3,410 3,335 3,332 Generation Cost [ 10 6 KRW] 24,977 24,866 23,219 23,235 System Marginal Price [KRW/kWh] Number of On-line Units 2,409 2,376 2,525 2,520 Reserve substitution ensures cascading prices for reserves so that higher quality reserves have always a price that is equal to or higher than any lower quality reserves. The principle of cascading reserve is illustrated through the following examples: The primary, regulation, and spinning reserves requirements are 500 MW, 500 MW, and 500 MW, respectively. The total primary reserve allocation of 700 MW, the total regulation reserve allocation of 500 MW, and the total spinning reserve allocation of 300 MW are correct. Here, reserve prices are for each MW of reserves assuming that unit will be paid only for satisfying thesingle category of reserves to which it is assigned. In other words, reserve prices are based on incremental model. A reserve energy offer is described by its quantity (MW) and price (KRW per MW per hour of availability = KRW/MWh). Also, different offers (MW/price pair) can be submitted by unit and period. Table 1 provides a synthesis of the reserve capability of each unit depending on its type (hydro, peaker, fast-start) and status Generic constraints Generic constraints provide a simple way for the user to impose supplementary constraints on the optimization problem to express a multitude of additional requirement (security-type constraints, unit additional operational constraints, etc.). There are three sets of variables that can make up a generic constraint: unit generation, demand bid load, and unit pumping load. ISSN: ISBN:

6 The construction of the generic constraints can be formulated mathematically as in (2): SumOverUnits [UnitGen UnitFactor] + SumOverDemandBids [DemandBidLoad DemandBidFactor] + SumOverPumpUnits [UnitPumpLoad UnitPumpLoadFactor] =,, ConstraintLimit (2) (Select one constraint type.) The type of constraint (=,, ) is specified for each constraint. Results of the dispatch create the marginal price of any binding generic constraint. 3 Case Study To begin, total generation costs for Lagrangian Relaxation (LR) and Mixed Integer Programming (MIP) runs in the unconstrained dispatch are compared in Table 2. The unconstrained dispatch is performed for 34 consecutive hours in which both 6 hours before the trading day and 4 hours after the trading day are included. In the unit commitment problem with 5-minute time-lapse simulation, the MIP is producing more excellent solution than the LR, where the tolerance, for the most part, reaches less than 0.1%. Also, total generation costs are significantly reduced in case hydro units as well as pumped-storage units are optimized with other units. In addition, total generation costs under the reserve or transmission line constraints are displayed. According to the market operation rule, primary reserve of 500 MW, regulation reserve of 500 MW, spinning reserve of 500 MW, non-spinning reserve of 1,000 MW, and operating reserve of 1,500 MW are required. The transmission network is, on an hourly basis, constrained by the limit of northward real power flows from the non-metropolitan regions to the metropolitan area through six routes. It is quite clear that total generation costs while meeting reserve or transmission line constraints are somewhat higher than otherwise. In Table 3, simulation results of each day in the winter and the summer season are given for comparative analysis of total generation costs, system marginal prices, and number of on-line units summed over 24 hours in the LR and the MIP. As mentioned earlier, the MIP is more cost effective than the LR since the number of generating units committed in the MIP is obviously less than that in the LR. However, the MIP has a detrimental effect on the system marginal price since there is a good chance that the MIP should cause small-sized but expensive generating units to be committed in the unconstrained dispatch. Obviously, it is no wonder that the MIP optimization with the combined cycle model makes further generation cost savings possible than the LR-based Resources Scheduling & Commitment (RSC). The estimation of energy costs of CT and ST units in the combined cycle model must be carried out with accuracy. 4 Conclusion Since 1999, Korea Power Exchange has been using the RSC which was AREVA s unit commitment and economic dispatch application. However, this LR-based software had some critical drawbacks as may be ascertained from the above discussion. In this context, a new generation scheduling program was specifically customized to ensure the reliable and fair market operations. Subsequently, this paper addressed features of MIP-based generation scheduling program and remarkable improvements over its predecessor designed to meet the needs of users who are operating in evolving Korean electricity market. 5 Acknowledgement This work is the outcome of a Manpower Development Program for Energy & Resources supported by the Ministry of Knowledge and Economy (MKE). References: [1] D. Hur, H. S. Jeong, and H. J. Lee, "A performance review of Lagrangian Relaxation method for unit commitment in Korean electricity market," in 2007 Power Tech, pp [2] Resource Scheduling and Commitment (RSC) Software Specification Document (SSD), CEGELEC ESCA Corporation, [3] F. Zhuang and F. Galiana, "Towards a more rigorous and practical unit commitment by Lagrangian Relaxation," IEEE Trans. on Power Systems, Vol. 3, No. 2, May [4] F. Lee, "Thermal unit commitment by sequential methods," in Application of Optimization Methods for Economy/Security Functions in Power System Operations, IEEE Tutorial Course 90EH PWR, pp [5] E-terracommit Software Specifications Design, AREVA T&D S. A., Tech. Rep. KPX-GSP-SOFT-002, Jul [6] J. A. Momoh, Electric Power System Applications of Optimization, New York, NY: Marcel Dekker, Inc., 2001, pp ISSN: ISBN: