IEEE TRANSACTIONS ON SUSTAINABLE ENERGY, VOL. 2, NO.
3, JULY 2011 277
Coordinated Trading of Wind and Thermal Energy Ali T. Al-Awami, Student Member, IEEE, and Mohamed A. El-Sharkawi, Fellow, IEEE AbstractTrading wind energy in short-term electricity mar- kets has high associated risks due to the uncertainties in hourly available wind, energy prices, and imbalance penalties. Coordi- nated trading of wind and thermal energy is proposed to miti- gate risks due to those uncertainties. The problem of wind-thermal coordinated trading is formulated as a mixed-integer stochastic linear program. The objective is to obtain the optimal trade-off bidding strategy that maximizes the total expected prots while controlling trading risks. For risk control, a weighted term of the conditional value at risk (CVaR) is included in the objective func- tion. The CVaR aims to maximize the expected prots of the least protable scenarios, thus improving trading risk control. A case study comparing coordinated with uncoordinated bidding strate- gies depending on the traders risk attitude is included. Simulation results show that coordinated bidding can improve the expected prots while signicantly improving the CVaR. Index TermsCoordinated bidding strategy, mixed integer sto- chastic programming (SP), risk, thermal power production, wind power production. NOMENCLATURE The most important notations used throughout the paper are listed below for quick reference. Indices: Bidding period. Scenario. Thermal unit. Wind plant. SegmentB. Decision Variables: Optimal bid of a thermal unit. Optimal bid of a wind plant. Actual, or realized, thermal power output. Thermal unit state; 1 means ON; 0 means OFF. Thermal power output corresponding to a segment of a piecewise linear thermal heat rate curve. and Auxiliary variables for computing . Manuscript received August 06, 2010; revised December 15, 2010; accepted January 27, 2011. Date of publication February 04, 2011; date of current ver- sion June 22, 2011. The work of A. T. Al-Awami was supported by King Fahd University of Petroleum & Minerals (KFUPM), Dhahran, Saudi Arabia. The authors are with the Department of Electrical Engineering, University of Washington, Seattle, WA 98195 USA (e-mail: aliawami@u.washington.edu; melshark@u.washington.edu). Color versions of one or more of the gures in this paper are available online at http://ieeexplore.ieee.org. Digital Object Identier 10.1109/TSTE.2011.2111467 Stochastic Variables: Actual, or realized, wind power output. Spot market energy price. , Under- and over-generation imbalance penalties as multipliers of the energy price. Combined imbalance penalty . Other Variables: Total optimal combined bid of wind and thermal power outputs. Total expected prots. Conditional value at risk at the condence interval. Condence level. Risk-aversion parameter. Prots associated with a scenario. Imbalance-up, or total over-generated energy in excess of combined schedule. Imbalance-down, or total under-generated energy in decit of combined scheduleE. Parameters and Constants: Start-up cost of a thermal unit. Minimum up-time of a thermal unit. Minimum down-time of a thermal unit. Initial minimum up-time of a thermal unit. Initial minimum down-time of a thermal unit. Thermal ramp-up rate [megawatt per hour (MW/h)]. Thermal ramp-down rate (MW/h). Thermal fuel cost. , , , Thermal heat rate curve parameters. Offset of a piecewise linear thermal heat rate curve. Slope of a segment of the piecewise linear thermal heat rate curve. Break point of a segment of the piecewise linear thermal heat rate curve. Initial state of a thermal units. 1949-3029/$26.00 2011 IEEE 278 IEEE TRANSACTIONS ON SUSTAINABLE ENERGY, VOL. 2, NO. 3, JULY 2011 , Minimum and maximum thermal power output. Rated wind power output. Number of periods. Number of scenarios. Number of thermal units. Number of wind plants. Number of segments. Probability of a scenario. I. INTRODUCTION B ECAUSE of the negative environmental impacts of fossil fuel coupled with the desire to achieve energy indepen- dence, power system researchers are working diligently on de- vising the tools necessary to accommodate renewable energy sources into a smarter power grid. Wind power has shown to be one of the most promising renewable sources. Several regions in the world have already set targets for the level of wind energy penetration into their power grids. The target set by the U.S. De- partment of Energy is 20% by 2030 [1]. The European Unions target is 14%17% by 2020 and 26%34% by 2030 [2]. In order to achieve these targets, several countries are sub- sidizing wind energy [3]. However, as wind technologies ad- vance, wind energy is becoming more competitive and subsi- dies will eventually be signicantly reduced or even halted. As such, wind producers are becoming more interested in partici- pating in electricity markets to maximize their prots. Most short-term electricity markets require that energy pro- ducers submit day-ahead schedules. In addition, system opera- tors apply imbalance penalties for producers not meeting their energy schedules. Because wind is stochastic and day-ahead wind forecast is highly uncertain, wind traders are exposed to high risks due to generation-schedule mismatch. In addition, en- ergy prices and imbalance penalties are uncertain as well. There- fore, wind traders need to mitigate the risks associated with these three sources of uncertainty. Researchers have suggested several approaches to mitigate wind trading risks. In [4][7], stochastic models have been de- veloped to optimally bid into the electricity market. One of the most comprehensive works is presented in [7], in which the un- certainty of energy spot market prices, imbalance penalties, and wind power output are considered. The objective is to maximize wind trading prots while controlling its risks by maximizing the conditional value at risk (CVaR). Another risk mitigation approach reported in the literature is coordinating wind with hydro power or pumped storage [8][12], other storage media [13], or thermal units [14]. In [8][12], coordination between wind and hydro is suggested and comparison between coordinated and uncoordinated bid- ding strategies is carried out. In [11], risk analysis is carried out through examining the standard deviation of the expected prots before and after coordination. In [12], wind-hydro coor- dination is formulated with transmission congestion constraints. It is assumed that the wind and hydro plants are separately owned. The use of Shapley value is suggested to share the extra prots due to coordination. However, these papers do not include risk control terms in the objective function or the constraints. Coordination between wind and other storage media is exam- ined in [13]. The added value of coordination is analyzed and a sensitivity analysis of the added value due to energy prices, im- balance prices, wind variations, and market closure lead-time is carried out. Again, no risk control is included in the optimiza- tion problem itself. One promising way to deal with wind trading risks is via coordination with thermal generators. Some thermal units have relatively high production costs. Hence, they might not be com- petitive to bid individually in the energy market during periods of low expected energy prices. Nevertheless, some of these expensive thermal units are exible; they have fast-ramping, near-zero minimum output power and very short minimum up/down time [15]. These features make them very suitable for balancing wind. In fact, the coordination with the larger, less exible thermal units can also improve the total expected prots for a producer participating in the day-ahead energy market and substantially reduce its risks, as discussed in this paper. In [14], a simplied model of coordinated bidding between a wind plant and a thermal unit is presented. An insight of the interaction between the wind plant and the thermal unit due to coordination is provided. It was observed that the thermal unit operated whenever the energy price or the imbalance charges were expected to be higher than the units incremental cost. However, some thermal constraints, such as minimum up/down times and ramping rate, are not considered and no risk analysis is carried out. In this paper, a wind-thermal coordinated bidding strategy is proposed for the day-ahead energy market. The problem is formulated as a mixed integer linear program (MIP) that deter- mines the optimal bidding curves for wind and thermal units as well as the optimal thermal commitment schedules. The un- certainties of wind power outputs, energy prices, and imbal- ance prices are included. The objective is to obtain the optimal trade-off between the expected prots and risk, depending on the producers risk aversion attitude. The CVaR for a selected condence level is used as a risk control metric. This tends to maximize the expected prots of the least protable scenarios. Thermal constraints, such as minimum up/down times, ramping rates, minimum and maximum power outputs, and startup costs are considered. A case study for a producer with a wind plant and several thermal units is carried out. The results demonstrate the coordination benets for the producer in terms of improving the expected prots while signicantly reducing both wind and thermal bidding risks. The results also show the impact of co- ordination and risk aversion attitude on both wind and thermal optimal scheduling decisions. II. PROBLEM DESCRIPTION This section gives a description of the market framework, coordination benets, stochastic programming (SP) approach, and the data preprocessing required for the optimization. AL-AWAMI AND EL-SHARKAWI: COORDINATED TRADING OF WIND AND THERMAL ENERGY 279 A. Market Framework Consider a pool-based electricity market at which suppliers and consumers submit day-ahead energy supply and demand bidding curves (bidding offers) for each hour. The hourly bid- ding curves represent the bid volumes in MW as functions of the energy spot market prices per MWh. The market operator aggregates these bidding curves to determine the hourly market clearing prices (MCPs) and the cleared energy volumes assigned to each participant by identifying the point of intersection of the aggregated supply and demand bids. Afterwards, balancing markets are carried out closer to real-time in order to help the market operator maintain the supply-demand balance. For each hour, each supplier is paid the MCP times the cleared energy volume and penalized for any imbalance between the cleared energy volume and the actual energy produced. Dif- ferent markets apply different imbalance penalty rules. For ex- ample, in the Spanish market, the imbalance charges depend on the direction of imbalance of the supply with respect to the di- rection of the overall system imbalance. Therefore, when the system is running short 1 , then those suppliers who are running short 2 are penalized, while those who are running long 3 are not penalized. The opposite holds true when the system is running long. These rules are set up such that the suppliers that degrade the system balance are penalized, while those that help maintain the system balance are rewarded. B. Coordination Benets Most electricity markets require that market participants submit their schedules for the next day several hours before the start of that day. In the Spanish market, for example, the market participants are required to submit their hourly bids for hours 124 of day at 10:00 A.M. of day . That is, the gap between the time the bids are submitted and real-time is 1438 hours. This leaves wind producers very vulnerable to high imbalance penalties due to the highly uncertain wind forecasts. In addition, energy prices and imbalance penalties are uncertain as well. Therefore, wind producers seek to nd ways to mitigate risks associated with these uncertainties. Unlike wind turbines, thermal units are highly predictable and controllable. However, they are still exposed to the uncer- tainty in energy prices. In addition, most thermal units are re- strained by such constraints as minimum up/down times and minimumpower outputs. Therefore, a thermal unit is exposed to the risk of operating at low-protability or even at loss at some periods because of these constraints. Other types of thermal units, such as reciprocating engines [15], are exible enough to operate with no minimum up/down time and zero minimum power output. However, these units usually have relatively high operating costs making them often uncompetitive in the spot market, especially at periods of low expected energy prices. Wind-thermal coordination, especially at periods of thermal low protability, is expected to be very benecial for both wind and thermal units. Coordination offers wind facilities the opportu- nity to avoid high imbalance penalties utilizing these thermal 1 Generation is less than demand. 2 Under-generating. 3 Over-generating. units. For the thermal units, instead of operating at low-prof- itability due to their various constraints, the coordination allows them to participate in balancing wind energy more economi- cally, thus more protable operation is achieved. Wind-thermal coordination is benecial for the independent system operator as well. For stability purposes, the ISO is re- quired to maintain the generation-load balance at all times. Co- ordination results in a reduction in the overall imbalance due to wind forecast uncertainty, thus helping the ISO achieve this ob- jective. C. Two-Stage SP Approach The problem of obtaining the optimal wind-thermal coordi- nated bidding curves that maximize the expected total prots while reducing trading risks is formulated as a two-stage mixed integer stochastic program. SP is commonly used for han- dling uncertainties in optimization [16]. It involves generating scenarios for each stochastic variable that represent plausible realizations of that variable. The generated scenarios of each stochastic variable need to reect the statistical properties of that variable. Then, using the generated scenarios of all the stochastic variables, a scenario tree is built. The models most commonly used in SP involve two stages. The decision variables of the rst stage must be decided upon before the stochastic variables are realized. These are called here-and-now decisions. The decision variables of the second stage depend on which of the scenarios is realized. These are called wait-and-see decisions. Note that the second stage de- cisions are affected by those of the rst stage. In this study, the stochastic variables are the wind power out- puts, the spot market energy prices, and the imbalance prices. Each branch of the scenario tree consists of a set of three strings: one for the hourly wind power outputs, one for the hourly en- ergy prices, and one for the imbalance prices. The here-and-now decision variables are the hourly bids of the wind plants and thermal units and the thermal unit commitment schedules. The wait-and-see decision variables are the thermal units actual power outputs since they depend on the realized scenarios of wind power, energy prices, and imbalance prices. D. Scenario Generation Several scenario generation techniques have been reported in the literature. In [17], Monte Carlo simulations were used to generate energy price scenarios given the hourly expected en- ergy prices and their associated variances. Scenarios for wind power output were generated in a similar fashion in [8]. The problem with the technique used in [8] and [17] is that coupling in the stochastic variable among consecutive hours is not taken into account. In [7] and [18], time series models were used to generate energy price scenarios, thus coupling among consec- utive hours is considered. Nonlinear optimization is suggested in [19] to generate scenarios that maintain a given set of statis- tical properties. In this work, a simple, yet effective, technique to generate wind power scenarios is developed. This technique takes into account the expected wind forecast, its variance, and the hour-to-hour coupling of wind power output. To generate wind scenarios, the knowledge of the following two statistical properties is assumed: 280 IEEE TRANSACTIONS ON SUSTAINABLE ENERGY, VOL. 2, NO. 3, JULY 2011 1) the distributions of the hourly wind power outputs given the forecasts; and 2) the distribution of the wind power ramping rate. The day-ahead hourly wind power outputs given the forecasts are assumed to follow normal distributions. However, any other distribution can be used. Both the expected value and the stan- dard deviation of the day-ahead hourly wind power forecasts are assumed to be available. Similarly, the wind power ramping rate is assumed to follow normal distribution whose mean and standard deviation are as- sumed to be known from historical wind data. The wind power ramping rate reects the correlation between the wind power output at hour and that of the proceeding hour . Using the wind forecasts and ramping rate, a set of wind power scenarios is generated. The way the scenarios are gener- ated is explained as follows: 1) For any given hour, use the expected value and standard deviation of the wind power forecast to generate scenarios. Do the same for all hours in the forecast time horizon. , i.e., . That guarantees satisfying the rst property. 2) Rearrange the scenarios generated in (1) to ensure that they obey the ramping rate statistical properties. This is done as follows: a) Given the wind power output of period , scenario , , generate a possible wind power output for period , . This is done by generating a number at random that follows the ramping rate sta- tistical properties and adding it to , i.e., , where and are ramping rates mean and standard deviation. b) Among the scenarios generated in (1) for period , search for the scenario that is closest to . That is, nd such that c) Swap with . d) Do (a)(c) for all scenarios in period . e) Do (a)(d) for all periods. f) Note that as gets larger for a given , there are fewer elements in to compare with , i.e., the chances that does not have an acceptable t in are higher. Therefore, after processing (a)(e), the last of the rearranged scenarios are eliminated. The distribution of the hourly wind power output should not deviate signicantly as long as is kept small. In general, as increases, the ramp rate criterion improves at the expense of a slight devi- ation in hourly distribution of the conditional wind power output. Hence, the choice of is determined heuristically based on the above properties. This ensures that the generated scenarios meet both of the sta- tistical properties of the wind power forecast. Note that swap- ping scenarios within a given one-hour period does not affect the distribution characteristics of the generated wind power sce- narios at that period. For the energy spot market prices and imbalance prices, sea- sonal ARIMA models [7] are commonly used. The same tech- nique is used in this paper. Note that the structure of imbalance penalties described in Section II-A is utilized to combine under-generation and over-generation penalty multipliers in one stochastic variable, as shown in [7]. This can be further explained as follows: 1) From the supplier perspective, when the supplier is run- ning short, it is penalized by paying to the market operator , where . When the supplier is running long, it is penalized by being paid by the market operator , where . Note that the suppliers running long are penalized by being paid less than or equal to MCP for the energy they supply in excess of their cleared energy volumes. 2) From the system perspective, when the system is running short, the imbalance penalty multipliers applied to sup- pliers are and . This way, the suppliers running short are penalized while those running long are not. However, when the systemis running long, the penalty multipliers are and . This way, the sup- pliers running long are penalized while those running short are not. 3) From the above, it is observed that and . Also, at least one of the two penalty multipliers is exactly equal to one at any given period and scenario. 4) Hence, for scenario generation purposes, the two penalty multipliers can be combined in one stochastic variable such that: a) When the system is running short, . b) When the system is running long, . E. Scenario Reduction The size of the scenario tree might render the optimization problem intractable. Hence, a reduction technique is applied to obtain a set of reduced scenarios that maintains the statistical properties of the original set to acceptable levels. This reduced scenario tree is then used to obtain the optimal solution. The forward selection algorithm proposed in [20] is used in this paper to reduce wind, energy price, and imbalance price scenarios. This algorithm is based on selecting a subset of scenarios from the original set such that the sum of minimum distances between all the members of the two sets is minimized. The algorithm starts with an empty subset . The rst member in is the scenario with the minimum sum of distances from the rest of the scenarios in . This, in essence, is approximately the centroid of all scenarios. The second member of is the sce- nario that minimizes the sumof minimumdistances between the members of and those of . The algorithm goes on until the required size of is reached or the required sumof minimum distances between the members of the two sets is achieved. III. MATHEMATICAL FORMULATION This section gives a detailed mathematical formulation of the wind-thermal coordinated bidding problem. AL-AWAMI AND EL-SHARKAWI: COORDINATED TRADING OF WIND AND THERMAL ENERGY 281 A. Risk-Neutral Wind-Thermal Coordinated Bidding The objective of maximizing the expected prots for a pro- ducer that owns both thermal generating units and wind plants can be expressed as follows: (1) where (2) The three terms between the brackets in (2) are the per-sce- nario prots due to the thermal units, wind plants and imbal- ances, respectively. These three terms can be expressed as fol- lows: (3) (4) (5) Equation (3) is the thermal prots per scenario. It constitutes the thermal energy revenues, thermal production costs, and the startup costs per period and per generating unit. Equation (4) shows the wind prots per scenario, which is a function of wind revenues. It is assumed that wind production, operating, and maintenance costs are negligible. If some or all of these costs are considerable, they can be incorporated easily in the wind prot. Equation (5) is the imbalance term, which consists of the imbalance-up prots and the imbalance-down penalty. Only one of these two terms can be nonzero for any given period and sce- nario. Imbalance-up is nonzero whenever the producer is run- ning long. Imbalance-down is nonzero, however, whenever the producer is running short. Mathematically, Imbalance-up and Imbalance-down conditions can be expressed as if otherwise (6) if otherwise (7) where (8) and (9) (10) (11) 1) Modeling Imbalance: To model the discontinuities in the and functions, an auxiliary binary variable is introduced in [6], [11], and [12] for each period and each sce- nario. In [7], the mutual relationship of the two terms is utilized to avoid using any additional binary variables. That is done by adding three more constraints for each period and each scenario. This method is more computationally efcient [7]. A similar ap- proach is used in this work. Awind-thermal coordinated version of the constraints used in [7] is shown as follows: (12) (13) (14) 2) Wind Operating Constraints: At each period and each scenario, the scheduled wind power of each wind plant must be within its operating limits (15) 3) Thermal Operating Constraints: At each period and each scenario, the actual, or realized, thermal power of each com- mitted thermal unit must be within its operating limits (16) At each period and each scenario, the realized thermal power of each committed thermal unit must not exceed its maximum ramping limits (17) As shown in [21] and [22], a computationally efcient way to model the minimum-up time constraint is as follows: (18) (19) (20) The rst condition ensures that the thermal unit will not be shut off unless it has been on initially for a sufcient number of periods . The second condition enforces the min- imum-up time constraint from to . The last condition enforces the minimum-up time 282 IEEE TRANSACTIONS ON SUSTAINABLE ENERGY, VOL. 2, NO. 3, JULY 2011 constraint for the rest of the planning horizon such that if the unit started at any of these periods, it remains on until the end of the planning horizon. Similarly, the minimum-down time constraint can be en- forced as follows: (21) (22) (23) 4) Piecewise Linearization of Thermal Production Cost: Production cost of thermal units is usually expressed as a quadratic function of the thermal power output [23] (24) As shown in [17] and [22], a computationally efcient piece- wise linear version of (24) can be used (25) (26) (27) (28) 5) Nondecreasing Bidding Curves: Most energy markets re- quire hourly bidding offers that are nondecreasing with respect to energy spot market prices. This requirement can be enforced in the optimization constraints as follows: (29) (30) (31) (32) Note that (29) and (30) correspond to the wind bidding offers while (31) and (32) correspond to the thermal bidding offers. To enforce nondecreasing offers, (29) and (31) are used. As stated earlier, there are three different stochastic variables in the problem under consideration: energy prices, imbalance prices, and wind power outputs. Therefore, it is very likely that two or more different scenarios can have the same energy price. To enforce the fact that scenarios with the same energy price must have the same bidding offer, (30) and (32) are included. A similar set of constraints is used by [7] and [17] to enforce nondecreasing bidding offers for wind energy [7] and thermal energy [17]. In each of (29)(32), a set of constraints is included for each period and each generating unit. However, for , the constraints are already satised. In addition, by symmetry of each set of constraints with respect to scenario pairs, if any of the constraints is satised for and , it is also satised for and . In this work, these two observations are utilized in order to reduce the number of constraints to (about half). This is done by considering only in (29)(32). B. Incorporating Risk Control The optimization problem described so far aims to maximize the expected prots with no consideration to risk. Note that one of the main incentives of wind-thermal coordination is to miti- gate trading risks for both wind and thermal plants. Therefore, a risk metric is necessary both to assess the benets of coordi- nation and to avoid risky bidding strategies. In nance, the value-at-risk (VaR) is commonly used to quan- tify a portfolio exposure to risk [24]. In SP terms, the VaR at a certain condence level represents the upper bound of the prot for the least protable scenarios. For example, for a condence level and scenarios considered and arranged in a descending order based on prof- itability, VaR is the prot associated with the 96th scenario (i.e., the fth least protable scenario). Therefore, VaR does not reect the extent of risk associated with the scenarios whose prots fall below the threshold indi- cated by VaR . In addition, VaR is difcult to handle when the prots are not normally distributed [25]. As an alternative measure, the CVaR is known to have better properties than VaR [25]. The CVaR at the condence level gives the expected prots for the least protable scenarios, i.e., the scenarios whose prots are less than or equal to VaR . Hence, for the example above, is the expected value of the prots associated with scenarios 96100 (i.e., the ve least protable scenarios). In this work, CVaR is used to control risk. The CVaR can be expressed as a linear term in the objective function [7], [25]. To include the CVaR, the objective function needs to be modied as follows: (33) where (34) In addition, the following constraints need to be included: (35) (36) is a weighting parameter to be set by the producer before the optimization algorithm is run in accordance with its risk- aversion attitude. Therefore, a risk-neutral attitude would corre- spond to , while a risk-averse attitude would correspond to . A more comprehensive mathematical description of the CVaR and the rationale behind (34)(36) is given in [25]. C. Individual Uncoordinated Bidding The mathematical formulation discussed above results in co- ordinated wind-thermal bidding. For comparison purposes, in- dividual thermal and individual wind bidding results are also AL-AWAMI AND EL-SHARKAWI: COORDINATED TRADING OF WIND AND THERMAL ENERGY 283 needed. In this paper, individual thermal optimization is per- formed by setting all the realized wind scenarios to zero for all , , and . To run individual wind optimization, the fuel costs of all thermal units are set to extremely high values. IV. TEST SYSTEM A system with one wind plant and ve thermal units is con- sidered. The wind plant installed capacity is 200 MW, while the thermal plant total installed capacity is 340 MW. The thermal units operating characteristics are given in the Appendix. Both the expected value and the standard deviation of the day- ahead hourly wind power forecast are assumed to be available. The values used here are given in the Appendix. The hourly standard deviations are selected such that the mean-absolute er- rors of the hourly forecasts are within the ranges reported in the literature [11], [26]. Similarly, the wind power ramping rate is assumed to follow normal distribution whose mean and standard deviation are as- sumed to be known from historical wind data. In this work, the expected ramp rate is assumed to be zero and the standard de- viation is assumed to be 12.5% of installed capacity (25 MW). These statistics in percentage were derived from publicly avail- able wind plant data in the Pacic Northwest [27]. Using the wind forecast and ramping rate, a set of wind power scenarios is generated. The sce- narios are generated and processed using the technique explained in Section II. The last 3% of the rearranged scenarios scenarios are eliminated. For the energy spot market prices and imbalance prices, the ARIMA models used in this work are identical to those sug- gested in [7]. The data used to t these models are taken from the electricity market of the Iberian Peninsula for the period Jan- uaryMarch of 2008 [28]. The rst four days of January and the weekends are excluded; the former to avoid the fact that the new years data might be outliers, while the latter to avoid any pos- sible weekly seasonality. Using those ARIMA models, a set of 4000 scenarios was generated for energy prices and another set of 4000 scenarios was generated for imbalance prices. The total number of scenarios in the scenario tree is . The problem as such is too large to be solved using any of the existing mixed integer programming solvers. Hence, the number of scenarios must be reduced. The method proposed in [20] is used to reduce wind, energy price, and imbalance price scenarios to 5 each, i.e., the reduced scenario tree comprises scenarios. For risk control, the condence level used to calculate is . This is a commonly used value [7]. The model is implemented using CPLEX 12.1, which is called from the OPL modeling language [29]. The average computational time of the optimal risk-averse coordinated bidding strategy is 306 s on a PC with an AMD Phenom Quad-Core 2.20-GHz CPU and 6 GB of RAM. V. SIMULATION RESULTS Several tests are performed on the proposed algorithm. These include risk-neutral and risk-averse optimization. For compar- ison, the coordinated and uncoordinated bidding strategies are shown. TABLE I EXPECTED PROFITS OF RISK-NEUTRAL COORDINATED VERSUS UNCOORDINATED BIDDING STRATEGIES TABLE II COMMITMENT SCHEDULE FOR UNCOORDINATED THERMAL BIDDING, TABLE III COMMITMENT SCHEDULE FOR COORDINATED THERMAL BIDDING, A. Risk-Neutral Optimization In this case, the optimization problem is solved with for uncoordinated wind, uncoordinated thermal, and coordi- nated wind-thermal bidding strategies. A comparison on the ex- pected prots of each case is shown in Table I. These results show a coordination gain as high as 1.81%. Tables II and III show the thermal unit commitment results for both uncoordinated and coordinated cases. For comparison, the differing commitment status due to coordination is shown in bold. These tables show that coordination results in having Unit 1 committed more often than in the uncoordinated case. Although Unit 1 is a relatively expensive unit, it is very ex- ible; it can ramp from 0 MW to full capacity in one hour and its . Therefore, it is suitable for wind balancing. However, Unit 1 is not the only unit affected by coordination. Figs. 1 and 2 show the different bidding curves for hours 14 and 21, respec- tively, for the uncoordinated and coordinated cases for the wind and thermal plants. Note that in hours 14 and 21, coordination did not result in any changes in the commitment schedule of any unit as seen in Tables II and III. As can be observed in both g- ures, coordination changed the bid thermal quantities substan- tially. Another interesting observation is that uncoordinated wind bid volume is not a function of the energy price. This can be seen by considering the objective function given in (33). If the thermal prot term (PFT) is excluded from the objective func- tion and is set to zero, the energy price can be eliminated from all the remaining terms; i.e., the optimal solution is not a function of the energy price. This has been reported in [7] as well. 284 IEEE TRANSACTIONS ON SUSTAINABLE ENERGY, VOL. 2, NO. 3, JULY 2011 Fig. 1. Bidding curves for hour 14, . Fig. 2. Bidding curves for hour 21, . TABLE IV COMMITMENT SCHEDULE FOR UNCOORDINATED THERMAL BIDDING, TABLE V COMMITMENT SCHEDULE FOR COORDINATED THERMAL BIDDING, B. Incorporating Risk Management In this case, the uncoordinated and coordinated optimization problems are run with . Table IVshows the commitment schedules of the thermal units without coordination. Bold digits in Table IV represent the cases where unit commitment status differs from that of risk-neutral uncoordinated thermal bidding strategies (Table II). These results show that the thermal units tend to be decommitted more often to avoid risky scenarios, thus improving CVaR. Table IV shows the commitment schedules of the thermal units with coordination. Underlined digits in Table V repre- sent cases where the unit commitment status differs from that of risk-averse uncoordinated thermal (Table IV). Fig. 3. Bidding curves for hour 14, . Fig. 4. Bidding curves for hour 21, . Comparison of risk-averse uncoordinated and coordinated thermal commitment schedules shows again that coordination causes thermal units to be committed more often to balance wind mismatch. Also, comparing Table V with Table III (risk-neutral coordinated), it is observed that the thermal units tend to be decommitted more often to reduce risk. Figs. 3 and 4 show the risk-averse bidding curves for hours 14 and 21, respectively, for the uncoordinated and co- ordinated cases for the wind and thermal plants. As can be seen in both gures, coordination changed the bid thermal quanti- ties substantially. Here, however, uncoordinated wind bid is a function of the energy price. This is due to the fact that with the CVaR term, energy price can no longer be eliminated from all the terms in the objective function. To study the effect of risk-aversion on the optimal bids more closely, Figs. 5 and 6 show the different bids with and without risk aversion for hours 14 and 21, respectively. These two g- ures indicate that risk-averse bid volumes are always less than or equal to risk-neutral bid volumes. This is expected as the wind plant tends to reduce its bid to avoid running short and the thermal units tend to be decommitted more often to avoid periods of nonprotable operation. To investigate further the impact of risk aversion, the ex- pected prots and CVaR for different values of are shown in Figs. 7 and 8 for individual wind and individual thermal bid- ding strategies, respectively. As expected, when increases, the CVaR increases and the expected prots decreases. For in- dividual wind bidding, increasing from 0 to 0.7 increases the CVaR by more than 53% for a reduction in expected prots of only 3%. For individual thermal bidding and the same range of , CVaR increases by as much as 190% for a reduction of only 2% in expected prots. AL-AWAMI AND EL-SHARKAWI: COORDINATED TRADING OF WIND AND THERMAL ENERGY 285 Fig. 5. Bidding curves for hour 14, and . Fig. 6. Bidding curves for hour 21, and . Fig. 7. Expected prots and CVaR for uncoordinated wind bidding. Fig. 8. Expected prots and CVaR for uncoordinated thermal bidding. Next, a comparison in prots and CVaR between the coordi- nated bids and the sum of uncoordinated bids is carried out for different values of . As shown in Fig. 9, the set of points be- longing to coordinated bids are higher and mostly to the right of the uncoordinated bids. Fig. 10 shows coordination gains in ex- pected prots and CVaRas percentages of uncoordinated prots and CVaR for different values of . The value of coordination is very noticeable in the gure. At , for example, prot gains are 1.8% and CVaR gains are higher than 30%. In addition, co- ordination gains in prots can be higher than 4% for almost the same CVaR of the sum of uncoordinated bids (see the case of ). These results are among the main contributions of this paper. Fig. 9. Expected prots and CVaR for coordinated and uncoordinated bidding. Fig. 10. Gains on expected prots and CVaR due to coordination. VI. DISCUSSION In this work, it is assumed that both the wind and thermal plants are connected to the same bus, exposed to the same en- ergy price, and owned by one producer. However, if some or all of these units are connected to different buses and exposed to different market prices, then coordination can still be possible should the market regulations allow. Also, if the generators are owned by more than one producer, these producers can form a coalition if market regulations allow, as discussed in [12]. In that case, coordination benets should be distributed among the different participants in the coalition in a fair and transparent manner, as discussed in [12]. Performing a longer-term study to examine the benets of coordination in terms of both expected prots and risk control seems to be an important future research direction. A cost-ben- et feasibility analysis of installing new thermal units for an existing wind plant participating in the spot market to mitigate risks associated with wind uncertainty is another relevant re- search point. The formulation presented in this paper forms the basis for these studies. Note that such applications would require 286 IEEE TRANSACTIONS ON SUSTAINABLE ENERGY, VOL. 2, NO. 3, JULY 2011 TABLE VI THERMAL UNITS DATA TABLE VII HOURLY EXPECTED VALUE AND STANDARD DEVIATION OF DAY-AHEAD WIND POWER FORECAST substantially more processing time than the single-day simula- tion presented in this work. This should not be considered a lim- itation, though, as these planning studies are conducted ofine. In addition, some systems have intraday markets, which give market participants the opportunity to adjust their day-ahead schedules a few hours before real-time. Adjustment markets help wind plants mitigate wind uncertainty risks [7]. Studying the benets of wind-thermal coordination in a system having an adjustment market can be another interesting research direction. VII. CONCLUSION Coordinated bidding among wind plants and thermal units has been proposed as a risk control mechanism. The objective is to maximize the expected prots while avoiding the risks asso- ciated with periods of high wind imbalance or lowwind/thermal revenues. The CVaR is used as a risk management metric. Simulation results show the following: Wind-thermal coordination can cause the combined ex- pected prots to increase while substantially improving the combined CVaR. Coordination results in the thermal units being committed more often to help balance any possible wind imbalance. Risk-averse bidding volumes are always less than or equal to risk-neutral bidding volumes. This is because wind plants are encouraged to reduce their bids to avoid high under-generation penalties and thermal units are decommitted more often to avoid periods of nonprotable operation. APPENDIX The data corresponding to the thermal units considered in this paper are given in Table VI. The hourly wind power forecast data are given in Table VII. REFERENCES [1] U.S. Department of Energy: 20% Wind Energy by 2030: Increasing Wind Energys Contribution to U.S. Electricity Supply Jan. 28, 2010 [Online]. Available: http://www1.eere.energy.gov/windand- hydro/pdfs/41869.pdf [2] European Wind Energy Association: Pure PowerWind Energy Tar- gets for 2020 and 2030 [Online]. Available: http://ewea.org/ [3] European Wind Energy Association: Economics of Wind Energy [On- line]. Available: http://ewea.org/ [4] G. N. Bathurst, J. Weatherill, and G. Strbac, Trading wind generation in short term energy markets, IEEE Trans. Power Syst., vol. 17, no. 3, pp. 782789, Aug. 2002. [5] P. Pinson, C. Chevallier, and G. N. Kariniotakis, Trading wind gen- eration from short-term probabilistic forecasts of wind power, IEEE Trans. Power Syst., vol. 22, no. 3, pp. 11481156, Aug. 2007. [6] J. Matevosyan and L. Sder, Minimization of imbalance cost trading wind power on the short-termpower market, IEEETrans. Power Syst., vol. 21, no. 3, pp. 13961404, Aug. 2006. [7] J. M. Morales, A. J. Conejo, and J. Perez-Ruiz, Short-term trading for a wind power producer, IEEE Trans. Power Syst., vol. 25, no. 1, pp. 554564, Feb. 2010. [8] E. D. Castronuovo and J. A. P. Lopes, On the optimization of the daily operation of wind-hydro power plant, IEEE Trans. Power Syst., vol. 19, no. 3, pp. 15991606, Aug. 2004. [9] J. M. Angarita and J. G. Usoala, Combining hydro-generation and wind energy. Bidding and operation on electricity spot markets, Elect. Power Syst. Res., vol. 77, pp. 393400, Apr. 2007. [10] J. M. Angarita and J. G. Usoala, Combined hydro-wind generation bids in a pool-based electricity market, Elect. Power Syst. Res., vol. 79, pp. 10381046, Jul. 2009. [11] J. Garca-Gonzlez, R. M. R. de la Muela, L. M. Santos, and A. M. Gonzlez, Stochastic joint optimization of wind generation and pumped-storage units in an electricity market, IEEE Trans. Power Syst., vol. 23, no. 2, pp. 460468, May 2008. [12] M. Zima-Bockarjova, J. Matevosyan, M. Zima, and L. Soder, Sharing of prot from coordinated operation planning and bidding of hydro and wind power, IEEE Trans. Power Syst., vol. 25, no. 3, pp. 16631673, Aug. 2010. [13] G. N. Bathurst and G. Strbac, Value of combining energy storage and wind in short-term energy and balancing markets, Electr. Power Syst. Res., vol. 67, pp. 18, 2003. [14] M. Gibescu, E. W. van Zwet, W. L. Kling, and R. D. Christie, Op- timal bidding strategy for mixed-portfolio producers in a dual imbal- ance pricing system, in Proc. 16th Power Syst. Computation Conf., Glasgow, Scotland, 2008. [15] Wartsila [Online]. Available: http://wartsila.com/ [16] P. Kall and S. W. Wallace, Stochastic Programming. Chichester, U.K.: Wiley, 1994. AL-AWAMI AND EL-SHARKAWI: COORDINATED TRADING OF WIND AND THERMAL ENERGY 287 [17] T. Li, M. Shahidehpour, and Z. Li, Risk-constrained bidding strategy with stochastic unit commitment, IEEE Trans. Power Syst., vol. 22, no. 1, pp. 449458, Feb. 2007. [18] M. A. Plazas, A. J. Conejo, and F. Prieto, Multimarket optimal bidding for a power producer, IEEE Trans. Power Syst., vol. 20, no. 4, pp. 20412050, Nov. 2005. [19] K. Hoyland and S. W. Wallace, Generating scenario trees for multi- stage decision problems, Manage. Sci., vol. 47, no. 2, pp. 295307, Feb. 2001. [20] H. Heitsch and W. Rmisch, Scenario reduction algorithms in sto- chastic programming, Comput. Optim. Appl., vol. 24, no. 2/3, pp. 187206, Feb. 2003. [21] A. Tuohy, P. Meibom, E. Denny, and M. OMalley, Unit commitment for systems with signicant wind penetration, IEEE Trans. Power Syst., vol. 24, no. 2, pp. 592601, May 2009. [22] M. Carrion and J. M. Arroyo, A computationally efcient mixed-in- teger linear formulation for the thermal unit commitment problem, IEEE Trans. Power Syst., vol. 21, no. 3, pp. 13711378, May 2006. [23] A. J. Wood and B. F. Wollenberg, Power Operation, Operation and Contol, 2nd ed. New York: Wiley, 1996. [24] W. Sharpe, Portfolio Theory and Capital Markets. New York: Mc- Graw-Hill, 1970. [25] R. T. Rockafellar and S. Uryasev, Optimization of conditional value-at-risk, J. Risk., vol. 2, pp. 2141, 2000. [26] R. Kelly, New modeling techniques: From wind assessment and fore- casting to wind resource management, in Proc. Bonneville Power Ad- ministration & California ISOs Int. Wind Forecast Techniques and Methodologies Workshop, Portland, OR, Jul. 2008. [27] Wind Integration Study Group: Revised 2004 10 Minute Data. North- west Power Pool Website 2006 [Online]. Available: http://www.nwpp. org/ntac/publications.html?CommitteeID=29 [28] Red Elctrica de Espaa, S. A. Sistema de Informacin del Operador del Sistema [Online]. Available: http://www.esios.ree.es [29] IBM ILOG [Online]. Available: http://www-01.ibm.com/soft- ware/websphere/products/optimization/ Ali T. Al-Awami (S08) received the B.Sc. and M.Sc. degrees in electrical engineering from King Fahd University of Petroleum and Minerals, Saudi Arabia, in 2000 and 2005, respectively. He is now working towards the Ph.D. degree at the University of Washington, Seattle, WA. In 2000, he joined the Saudi Electricity Company as a control engineer in the System Operation Department. In 2002, he joined the Electrical En- gineering Department at KFUPM as a Graduate Assistant, where he is currently a Lecturer. He authored and coauthored several papers and book chapter in his research areas. His research interests include power system operation and optimization and the integration of renewable energy sources into the power grid. Mohamed A. El-Sharkawi (S76M80SM83 F95) received the Ph.D. degree in electrical engi- neering from the University of British Columbia in 1980. In 1980, he joined the University of Washington, Seattle, WA, as a faculty member where he is presently a Professor of Electrical Engineering. He also served as the Associate Chair and the Chairman of Graduate Studies and Research. He has been the Founding Chairman of numerous IEEE task forces and working groups and subcommittees and published over 200 papers and book chapters in his research areas. He authored two textbooks on Fundamentals of Electric Drives and Electric Energy: An Introduction. He also authored and coauthored ve research books in the area of intelligent systems and power systems. He holds ve licensed patents in the area of renewable energy VAR management and minimum arc sequential circuit breaker switching.