Вы находитесь на странице: 1из 8

2012 International Conference on Future Electrical Power and Energy Systems Lecture Notes in Information Technology, Vol.

Unit Commitment Problem for Hybrid Power System Using Binary/Real Coded PSO K. Chandrasekaran*, Sishaj P Simon
Department of Electrical and Electronics Engineering, National Institute of Technology, Tiruchirappalli, *chansekaran23@gmail.com; sishajpsimon@nitt.edu * Corresponding author Keywords: binary/real coded particle swarm optimization, economic dispatch problem; tanh function, unit commitment problem.

Abstract. The utilization of wind energy sources is increasing throughout the world. It is therefore important to develop the protocol for the integration of wind generation system with conventional thermal unit generation system. High wind penetration can lead to high-risk level in power system reliability. In order to maintain the system reliability, wind power dispatch is usually restricted and energy storage is considered for smoothing out the fluctuations. This paper presents a binary and real coded particle swarm optimization (BRPSO) algorithm for solving the unit commitment problem (UCP) for hybrid power system. On solving UCP, the proposed binary coded PSO finds the ON/OFF status of the generating units while the economic dispatch problem (EDP) is solved using the real coded PSO. Tanh function is proposed to improve the particle searching mechanism in binary PSO. The proposed methodology is tested and validated on 10 and 26 generating unit system for 24 hour scheduling horizon. 1. Introduction Unit commitment (UC) is the most significant optimization task in the operation of the power systems. The UCP schedules the on/off status of thermal generating units while satisfying the physical and operating constraints of the power system [1]. The UCP has commonly been formulated as a nonlinear, large scale, mixed-integer combinatorial optimization problem with constraints. Many methods with various degrees of near-optimality, efficiency, ability to handle difficult constraints and heuristics, have been suggested for UC in the literature [2-3]. These include deterministic and stochastic search approaches. Generally, the deterministic approach is flexible but often prone to the curse of dimensionality. Methods such as genetic algorithms simulated annealing, evolutionary programming; particle swarm optimization (PSO) and ant colony optimization (ACO) are meta-heuristic optimization methods that attract much attention because of their ability to search not only local optimal solutions but also global optimal solution and easily dealing with various difficult nonlinear constraints. However, in this context, an attempt is made to solve UCP using a newly developed binary and real coded PSO algorithm. Here, binary coded PSO algorithm using tanh function is proposed to solve the UCP and the real-coded PSO algorithm is used to solve the EDP. Nomenclature
Fc(P i,k )
Fuel cost ($). Fc (Pi, k ) = ai + bi * Pi, k + ci * Pi, k 2 Cost co-efficient of ith generator unit. Total number of hours considered. Total system demand at kth hour.

ai , bi , ci
H

Load k

978-1-61275-005-7/10/$25.00 2012 IERI

387

ICFEPES2012

LOLPk , spec

Specified LOLP at kth hour. Calculated LOLP at kth hour. Generation power output of unit i at kth hour. Maximum/Minimum power output of unit i. Total start- up cost of unit i during the study period.

LOLPk

Pi , k
Pi,max Pi,min
SC i , k

2. Problem Formulation Generally the unit commitment problem is solved with minimizing the total fuel cost. The fuel cost minimization problem is formulated as: Minimize H N H H N H (1) ((ai + bi .Pi,k + ci .Pi, k 2 )) + SCk Fc = Fc ( Pi , k ) + SCk =

k =1 i =1

k =1

k =1 i =1

k =1

2.1 Problem Constraints 2.1.1 Power balance constraint 2.1.2 Spinning reserve constraint 2.1.3. Capacity limit 2.1.4 Unit minimum ON/OFF durations 2.1.5. Unit ramp constraints 2.1.6 Reliability constraints
LOLP LOLPk spec

(2) (3)

The reliability indices [4] for each hour are calculated by using equation (3).
LOLPk =
j = LC

PR j k

2.1.7 Wind power Forecast Uncertainty In the proposed method, the wind power forecast uncertainty is assumed to be normally distributed .The probability distribution function of the forecast wind power can be described by a four-step model (0,1 , 2 , 3 ) , where the standard deviation is a percentage (usually 5%) of the expected wind power (distribution mean). Such a model will encompass more than 99% of load uncertainty. Based on the above analysis, the LOLP index for each hour including the wind power forecast uncertainty can be calculated by the following equations:
LOLP k =

LOLP k (l ) * PL (l ),
l =1

k [1, H ]

(4)

where PL (l ) - Probability of load step l, LOLPk (l ) -LOLP for the load step l of hour k., 2.1.8. Wind power limit and energy storage operating constraints In this paper, the wind energy dispatch is restricted to a fixed percentage of the system load considering system operating constraints. Here, the hourly wind energy dispatch is restricted to X % of the hourly system load demand in order to maintain system reliability, and the surplus wind energy above X % of the system load is stored in the battery during charging operation. The charging/discharging operation in the battery is subject to the following constraints. a. Initial storage energy limit (5) E ini = w.E max b. Charge/discharge constraints.
Ec ( k ) = . Pinv ( k ) . dk

for ch arge

(6) (7)

Ed (k ) =

inv

(k ) . dk

for disch arge

388

All efficiency is integrated in to , which is the combined efficiency of energy storage system. The discrete representation for storage energy at hour k, E (k ) is given as (8) E ( k ) = E ( k 1) + Ec ( k ) Ed ( k ) The constraints for charge/discharge operation of the battery are given below. (9) E min E (k ) E max for all k (10) w . Emin E (k ) Emax at the end of dis ch arg e (11) Emin E (k ) . Emax at the end of ch arg e The charge operation ends when E (k ) increase to a limit up to .E , whereas discharge operation ends when E (k ) decrease to a limit up to w.E max . In order to ensure that sufficient reserve power is available to prevent unexpected outages, the minimum charge stored in the battery is considered as w.E max . c. Inverter capacity limit (12) 0 Pinv ( k ) Pinv max
max

3. Implementation of Binary PSO for UCP Particle swarm optimization, first introduced by Kennedy and Eberhart, is one of the modem heuristic optimization algorithms [5]. In UCP, binary numbers 1 and 0 are used to indicate the unit status ON/OFF. The PSO algorithm used in [5] is essentially a real-coded algorithm, thus, some modifications are needed to deal with binary-coded optimization problem. In the proposed binary PSO, the relevant variables are interpreted in terms of changes of probabilities. In binary particle swarm, xpf and gbest can take on values of 0 or 1 only. The velocity xpf will determine a probability threshold. If xpf is higher, the individual is more likely to choose 1, and lower values favour the 0 choice. Such a threshold needs to stay in the range [0.0, 1.0]. One straightforward function for accomplishing this is common in neural networks. The function is called the sigmoid function [6] and is defined as follows: 1 f ( x pq ) = (13) 1 + exp( x )
pf

However, to improve the performance of binary PSO, another function, tanh is used as given in (14).
f ( x pf ) = tanh( V pf ) = exp(2 * x pf ) 1 exp(2 * x pf ) + 1

(14)

A random number is generated between 0 and 1, to decide the unit status as 0 or 1. If f (xpf) is greater than rand (0, 1) then, the unit status is 1 otherwise 0. The chances of bit flipping are found to be more in the case of the tanh function than in the sigmoid function and gives better quality of solution. 3.1 Repair Strategy in Binary Coded PSO Whenever the commitment status for each time interval is generated randomly or if the particle position is modified, violation of minimum up/ down time constraints (sec. 2.1.4) and spinning reserve constraint (2) has to be checked as follows. Step 1: If the spinning reserve (2) is met, then go to step 3. Otherwise, go to next step. Step 2: The less expensive units which are in the OFF state are identified and turned ON. Then go to step 1. Step 3: If the spinning reserve constraint is satisfied, then the minimum up and down time constraints (sec. 2.1.2) are checked for each unit. If there is any violation in the minimum up or down time constraint then a repair scheme is performed to overcome the violation. For instance, let us assume that the Ton and Toff for a hypothetical unit is 4 and 5. For a scheduling interval of 12 hours, if the actual off time for that unit is 3 hours (5th -7th hour), then it violates the Toff constraint. In this case, the unit status before 5th hour or after 7th hour should be made OFF. By doing this change, if it violates the Ton constraint, then the status of the units are made ON.
389

Step 4: The repair scheme in step 3 may affect the spinning reserve constraint of the system. If the reliability level is met, then accept the feasible solution. Otherwise go to step 2. A minimum number of trials should be set for the repair mechanism. These steps are carried out for the entire hourly load. 3.2 Repair Strategy in Real Coded PSO Whenever the particle position is modified by real coded PSO algorithm, violation of capacity limits of generating unit (sec. 2.1.3) has to be checked as follows: Step 1: If the generated power (Pi ) capacity limit (sec. 2.1.3) is met, then go to step 3. Otherwise, go to next step. Step 2: If Pi >Pmax then Pi=Pmax. If Pi<Pmin then Pi=Pmin. Step 3: End 3.3 Evaluation of Fitness of the Population Solve EDP to calculate the total operating cost by satisfying all system constraints. Evaluate the fitness value of each particle corresponding to the velocity. In order to avoid premature convergence a fitness function as in (15) is used. FIT p = 1/Cost p , if Cost p > 0 (15) = 1 + abs ( Cost ), if Cost < 0
p p

Costp is the cost of generation of pth particle position. Cost is calculated using the objective function (1). The step-by-step procedure for the proposed method is given as a flowchart in Fig.1. 4. Results and Discussions All the programs are developed using MATLAB 7.01.The system configuration is Pentium IV processor with 3.2GHz speed and 4 GB RAM. Here for all the test system, proposed binary coded PSO is used to solve the UCP and the real-coded PSO is used to solve the EDP. 4.1 Test system 1- 10 unit system The generator cost coefficient, load profile and power generation limits are adapted from ref. [7] and the system spinning reserve is considered as 10 % of load at each hour.

390

Economic dispat ch using real coded P SO algorithm Start Generate random initial real coded particle for t he status of gth particles

Start Input system data, set binary coded PSO parameters Generate random initial binary population of m particles

Set iter=1

Update the velocity of each particles using v pf t+1= w.v pf t+c1.rand(0,1).(x gbest-x pf t)+c2.rand(0,1).(x pbest-x pf t) here, xindicated real values Modify the particles position x pf t+1=x pf t+v pf t+1

Set iter=1

Calling real coded PSO program

Update t he velocit y of each part icles using v pf t+1= w.v pf t+c1.rand(0,1).(x gbest-x pf t)+c2.rand(0,1).(x pbest-x pf t) here, X indicates binary value Modify the particles position using x pf t+1=x pf t+v pf t+1 Do repair strategy for constraint management as given in section 3.1 Collect Fitness ( FIT ) for each particle position with corresponding st at us

Do repair strategy for constraint management as given in section 3.2 Calculate Fitness ( FIT) of the modified particle position using (15) Memorize the best solution at each interval and corresponding fitness

Is Iter<= it er m ax No

Yes

it er =iter+1

Memorize the best solution at each interval iter =iter+1

Store best fitness with corresponding generator values. T hen do g=g+1 Yes Is Iter<=itermax No Is g<=gr m ax No Output the best solution of generat ed UC status and FIT

No End Output the best solution

Yes

Fig. 1 Flowchart for the implementation of UCPT Table 1 UC status and power dispatch for 10 unit system Operating Unit no. Hour Unit no. Cost ($) 1100000000 13683 13 1111111100 1100000000 14554 14 1111111000 1100100000 17709 15 1111100000 1100100000 18598 16 1111100000 1101100000 20580 17 1111100000 1111100000 23487 18 1111100000 1111100000 23262 19 1111100000 1111100000 24150 20 1111111100 1111111000 28111 21 1111111000 1111111100 30118 22 1100111000 1111111110 31976 23 1100010000 1111111111 33953 24 1100000000 Total ($) Table 2 Comparison of results-10 unit system
Solution Technique Minimum operating cost $ Mean operating cost $ Maximum operating cost $

Hour 1 2 3 4 5 6 7 8 9 10 11 12

Operating Cost ($) 30058 27251 24150 21514 20642 22387 24150 30548 27251 22736 17645 15427 563940.0

PSO[7] MIPSO[6] APSO[8] MHPSO[9] IBPSO[10] QBPSO[11] HPSO[12] BRCSA

574285.1 572075.0 565450.0 564419.0 563977.0 563977.0 563942.3 563940.0


391

564423.9 564155.0 564772.3 563945.0

565312 565785.3 564316.0

In [6-12], the binary PSO solution is obtained using the sigmoid function which is given in equation (13). However in the proposed method, to improve the performance of the binary PSO, tanh function is used and the results are compared with ref. [6-12]. The total production cost obtained for the 10 unit system for a 24-hour-time interval is $ 563940. The detail status of each unit is given in Table 1. Table 2 provides the comparison of the total cost obtained from binary PSO using tanh with that of PSO using sigmoid function. It is clearly seen that the proposed tanh function in binary coded PSO yields better results than other techniques available in the literature. The minimum cost so far reported in literature is $ 563942 [12] which is $ 2 higher than the optimum solution obtained from BRPSO using tanh function. It is inferred from Fig. 1, that the characteristics of binary PSO using tanh function steadily reaches the minimum value after few iterations when compared with PSO using sigmoid function. It can be concluded that convergence characteristic of binary PSO using tanh function is fast and yield better solution than the binary PSO that uses sigmoid function. 4.2 Test system 2:26 unit system The fuel cost coefficient and load profile is adapted from the ref [13]. Here, two test cases are carried out to validate the proposed algorithm. In case 1, the system spinning reserve is set to the maximum capacity of the largest committed unit as in the ref [13] and results are compared with [13-15]. In case 2, the system spinning reserve is set by the reliability index loss of load probability (LOLP) as in the ref. [4]. Hence in case 2 for the complete time horizon, the calculated reliability index (LOLP) value should be less than or less than equal to the specified LOLP value (LOLPspec). For the various value of LOLPspec (0.5 5, 1% and 1.5 %) are considered and result are compared with ref. [4]. Out of 30 trials, the best total production cost obtained for case 1 and case 2 is given in Table 3. Table 3 provides the comparison of the total operating cost obtained from BRPSO algorithm for case 1 and 2 with that of other techniques existing in the literature. In both cases, the minimum cost produced by BRPSO is less than results reported in literature. 4.3 Solution Quality and Convergence efficiency The convergence characteristics GA and PSO for case 1 and case 2 (26 unit system) discussed here is shown in Figs. 2 and 3. It is inferred from Fig.2 and Fig. 3 that the characteristic of PSO algorithm steadily reaches the minimum value after few iterations and produce better quality of solution
Table 3 UCP solution-26 unit system- case 1
Case 1 Solution techniques ANN-DP[13] ILR[14] IPL-ALH[15] GA BRPSO Minimum cost ($) 729326.50 725996.90 721352.90 721310.00 721208.00 Solution techniques SA[4] GA BRPSO Case 2 0.5 % 722401 722466 722358 Minimum cost ($) 1% 717938 717910 717867 1..5 % 710696 710652 710557

Table 4 UCP solution-26 unit system- hybrid systemBRPSO Thermal system (case 2, 0.5 %) Hybrid system Minimum cost $ 722358.00 671681.22

4.4 Test System 3Hybrid power system In hybrid generation system the reliability of the system is a major task. Hence when integrating the wind power with thermal power system, here the reliability function is considered as constraint in UC problem. The reliability data for the conventional generating unit is taken from [4]. Forecast wind power is given in Appendix A. Here it is assumed that the wind farm with 100 identical 1.75 MW WTG units with FOR of 0.04, and almost 10 % of the installed system capacity is integrated with the
392

26 unit system. The wind penetration in this case is 5 % of the system load at each hour of the dispatch period [4]. Here, inverter capacity of 120 MW is used to improve the system reliability. Emin and Emax of the energy storage battery are 30 MWh and 300 MWh respectively. The battery co-efficient e , e and w are 0.9, 1 and 0.1 respectively. The total number of dispatching unit in the system is twenty seven (26+1) including one energy storage unit. From Table 4, it is noted that, a cost saving of 50676.78 $ is obtained per day for the proposed hybrid generation system compared with case 2 in

Fig.2 Convergence graph-10 unit system;

Fig.3 Convergence graph-26 unit system-case 1

Fig.4 Convergence graph-26 unit system-case2;

Fig.5 Battery operation

previous section and therefore a significant amount of saving will be reflected per annum. The charge and discharge power of battery at each hour is shown in Fig. 5. 5. Conclusion Application of BRPSO is a new approach in solving the UCP for hybrid power system. A detailed study has been carried out to bring out the significance of integrating the wind energy with conventional units in every power generation system. Due to the intermittency and unpredictability of the wind power generation, an energy storage facility is incorporated in the proposed model. This helps in smoothing out the fluctuating nature of wind power (reliability impact of climate change) and improves the continuity of power supply from a wind generation system. From the result it is clearly understood that the proposed hybrid model and solution methodology leads to energy (cost) savings and provides eco-friendly society. Also results demonstrate that BRPSO is a very competent method to solve the UCP. BRPSO generates better solutions than the other methods, mainly because of its intrinsic nature of updates of positions and velocities. The second reason is the introduction of tanh function in UCP increase the probability of flipping the status of the binary variable thereby improving the performance of the binary PSO and determines the optimal status. The comparison of the results with other methods reported in the literature shows the superiority of the proposed methodology and its potential for solving UCP in a power system. 6. Reference [1] AJ. Wood and BF Wollenberg: Power generation, operation and control. (John Wiley and sons, New York, 1996).
393

[2] Padhy N.P: Unit Commitment-a Bibliographical survey IEEE Trans. Power Syst. Vol. 19(2004), p. 11961205. [3] Burns R. M and Gibson C. A. Optimization of Priority Lists for a Unit Commitment Program. In: Proceedings of IEEE PES (summer meeting), Vol. 75(1975), 453-461. [4] Simopoulos, D. N., Kavatza, S. D. and Vournas, C. D: Reliability constrained unit commitment using simulated annealing, IEEE Trans. Power Syst., Vol. 21(2006), p. 16991706. [5] J. Kennady and RC. Eberhart: Particle swarm optimization, Proc. of IEEE Int. Conf. on Neural Networks, Piscataway, NJ, IV1995; 1942-48. [6] Lingfeng Wang and Chanan Singh: Unit commitment considering generator outages through a mixed-integer particle swarm optimization algorithm. Applied soft computing, Vol. 9(2009). p. 947-953. [7] T. O. Ting, M.V.C. Rao, C.K. Loo and S.S. Ngu: Solving Unit Commitment Problem Using Hybrid Particle Swarm Optimization, Journal of Heuristics, Vol. 9 (2003), p. 507520. [8] V.S. Pappala and I. Erlich: A New Approach for Solving the Unit Commitment Problem by Adaptive Particle Swarm Optimization, Power and Energy Society General Meeting Conversion and Delivery of Electrical Energy in the 21st Century, 2008 IEEE , p. 1-6 [9] Le Thanh Xuan Yen, Deepak Sharma, Dipti Srinivasan and Pindoriya Naran Manji. A Modified Hybrid Particle Swarm Optimization Approach for Unit Commitment, Evolutionary Computation (CEC) 2011, p. 1738 - 1745 [10] Xiaohui Yuan, Hao Nie, Anjun Su, Liang Wang and Yanbin Yuan.: An improved binary particle swarm optimization for unit commitment problem. Expert Systems with Applications Vol. 36(2009), p. 80498055. [11] Yun-Won Jeong, Jong-Bae Park, Se-Hwan Jang and Kwang Y. Lee, Fe: A New Quantum-Inspired Binary PSO for Thermal Unit Commitment Problems, Intelligent System Applications to Power Systems, 2009; ISAP '09: p. 1-6 [12] T. O. Ting, M. V. C. Rao and C. K. Loo: A Novel Approach for Unit Commitment Problem via an Effective Hybrid Particle Swarm Optimization, IEEE Trans. Power Syst. 21(1) 2006, p. 411-418. [13] C. Wang and S.M. Shahidehpour: Effects of ramp rate limits on unit commitment and economic dispatch, IEEE Trans. Power Syst., Vol. 8 (1993), p.13411350. . [14] W. Ongsakul and N. Petcharaks: Ramp rate constrained unit commitment by improved adaptive Lagrangian relaxation. In: Proceedings of the Electricity Supply Industry in Transition: Issues and Prospect for Asia, Asian Institute of Technology, 2004. [15] Vo Ngoc Dieu and Weerakorn Ongsakul: Ramp rate constrained unit commitment by improved priority list and augmented Lagrange Hopfield network. Electric Power Systems Research, Vol. 78 (2008), p. 291301. Appendix A
Hou r 1 2 3 4 5 6 Wind power 60 65 85 85 88 87 Hou r 7 8 9 10 11 12 Wind power 95 122 127 100 134 100 Hou r 13 14 15 16 17 18 Wind power 98 100 120 130 70 80 Hou r 19 20 21 22 23 24 Wind power 70 70 130 124 110 70

394

Вам также может понравиться