Electrical Power and Energy Systems 29 (2007) 540550 www.elsevier.

com/locate/ijepes

An evolutionary programming based simulated annealing method for solving the unit commitment problem
C. Christober Asir Rajan
a

a,*

, M.R. Mohan

b,1

Department of EEE, Pondicherry Engineering College, Pondicherry 605014, India b Department of EEE, Anna University, Chennai 600 025, India

Received 16 February 2004; received in revised form 4 August 2006; accepted 22 December 2006

Abstract This paper presents a new approach to solve the short-term unit commitment problem using an evolutionary programming based simulated annealing method. The objective of this paper is to nd the generation scheduling such that the total operating cost can be minimized, when subjected to a variety of constraints. This also means that it is desirable to nd the optimal generating unit commitment in the power system for the next H hours. Evolutionary programming, which happens to be a global optimisation technique for solving unit commitment Problem, operates on a system, which is designed to encode each units operating schedule with regard to its minimum up/down time. In this, the unit commitment schedule is coded as a string of symbols. An initial population of parent solutions is generated at random. Here, each schedule is formed by committing all the units according to their initial status (at start). Here the parents are obtained from a pre-dened set of solutions, i.e. each and every solution is adjusted to meet the requirements. Then, a random recommitment is carried out with respect to the units minimum down times. And SA improves the status. The best population is selected by evolutionary strategy. The Neyveli Thermal Power Station (NTPS) Unit-II in India demonstrates the eectiveness of the proposed approach; extensive studies have also been performed for dierent power systems consists of 10, 26, 34 generating units. Numerical results are shown comparing the cost solutions and computation time obtained by using the Evolutionary Programming method and other conventional methods like Dynamic Programming, Lagrangian Relaxation and Simulated Annealing and Tabu Search in reaching proper unit commitment. 2007 Elsevier Ltd. All rights reserved.
Keywords: Unit commitment; Simulated annealing; Dynamic programming; Tabu search; Evolutionary programming

1. Introduction Electricity generating companies and power systems has the problem of deciding how best to meet the varying demand for electricity, which has a daily and weekly cycle. The short-term optimisation problem is how to schedule generation to minimize the total fuel cost or to maximize the total prot over a study period of typically a day, subject to a large number of constraints that must be satised.
Corresponding author. Tel.: +91 0413 2655281; fax: +91 0413 2655101. E-mail addresses: asir_70@hotmail.com (C.C.A. Rajan), mohan@ annauniv.edu (M.R. Mohan). 1 Tel.: +91 044 22351445; fax: +91 044 22350397. 0142-0615/$ - see front matter 2007 Elsevier Ltd. All rights reserved. doi:10.1016/j.ijepes.2006.12.001
*

The daily load pattern for a given system may exhibit large dierences between minimum and maximum demand. Therefore enough reliable power generation to meet the peak load demand must therefore be synchronized prior to the actual occurrence of the load. Thus it is clear that it is not proper and economical to run all the units available all the time. Since the load varies continuously with time, the optimum condition of units may alter during any period. Therefore the problem of determining the units of a plant that should operate for a given load is the problem of unit commitment. For total number of units of higher order, the problems associated with unit commitment have generally been dicult to solve because of uncertainty of particular aspects of the problem. For instance the availability of fuel in precise, load forecast

C.C.A. Rajan, M.R. Mohan / Electrical Power and Energy Systems 29 (2007) 540550

541

variable costs aected by the loading of generator units and the losses caused by reactive ows are some of the unpredictable issues. There are other problems of inconsistency that aect the overall economic operation of the electric power station. In order to reach a feasible solution for unit commitment problem (UCP), dierent considerations must be considered. Simulated annealing (SA) is the powerful general-purpose stochastic search technique to solve hard-constrained optimization problems. Though it takes long time, it has many strong features such as, it is easy to implement, requires little expert knowledge and is not memory intensive. Further, it can start with any initial solution and improves on it to nd optimal solution with a high probability. The SA optimization technique, begin with a randomly generated solution and then making successive random modications, until a stopping criterion is satised. Research endeavours, therefore, have been focused on; ecient, near-optimal unit commitment (UC) algorithms, which can be applied to large-scale, power systems and have reasonable storage and computation time requirements. A survey of existing literature [128] on the problem reveals that various numerical optimisation techniques have been employed to approach the complicated unit commitment problem. More specically, these are the Dynamic Programming method (DP), the Mixed Integer Programming method (MIP), the Lagrangian relaxation method (LR), the Branch and Bound method (BB), the Expert system (ES), the Fuzzy Theorem method (FT), the Hop Field method (H), the Simulated Annealing method (SA), the Tabu Search (TS), the Genetic Algorithm (GA), the Articial Neural Network (ANN), the Evolutionary Programming (EP) and so on. The major limitations of the numerical techniques are the problem dimensions, large computational time and complexity in programming. The DP method [1,2,15] is exible but the disadvantage is the curse of dimensionality, which results it may leads to more mathematical complexity and increase in computation time if the constraints are taken in to consideration. The MIP method [3] is capable of including in the solution all of the dierent constraints on the schedule. And it include its computational practicability for realistic systems and proper representation of reserves associated with different risk levels. The MIP method [4,5] for solving the unit commitment problems fail when the number of units increases because they require a large memory and suer from great computational delay. The LR approach [69] to solve the short-term UC Problems was found that it provides faster solution but it will fail to obtain solution feasibility and solution quality problems and becomes complex if the number of units increased. The BB method [10] is capable of including in the solution all of the dierent and complex constraints. Hence the solution obtained is always practically implementable. It is also possible to use the method with a number of objective functions, whichever may be appropriate to a particular system. And the times taken to solve the maintenance scheduling

problem are quite reasonable in view of the anticipated frequency of solution. The BB method [11] employs a linear function to represent fuel cost and start-up cost and obtains a lower and upper bounds. The diculty of this method is the exponential growth in the execution time for systems of a practical size. An ES algorithm [12,15] recties the complexity in calculations and saving in computation time. But it will face the problem if the new schedule is diering from schedule in database. In the FT method [13,15,26] using fuzzy set solves the forecasted load schedules error but it will also suer from complexity. The H neural network technique [14] considers more constraints but it may suer from numerical convergence due to its training process. SA [1619,25,26] is a powerful, generalpurpose stochastic optimisation technique, which can theoretically converge asymptotically to a global optimum solution with probability one. But it will take much time to reach the near-global minimum. The TS [2022,25] is an iterative improvement procedure that starts from some initial feasible solution and attempts to determine a better solution in the manner of a greatest-decent algorithm. However, TS is characterized by an ability to escape local optima by using a short-term memory of recent solutions. GA [15,2326] is a general-purpose stochastic and parallel search method based on the mechanics of natural selection and natural genetics. It is a search method to have potential of obtaining near-global minimum. And it has the capability to obtain the accurate results within short time and the constraints are included easily. The ANN [14,30] has the advantages of giving good solution quality and rapid convergence. And this method can accommodate more complicated unit-wise constraints and are claimed for numerical convergence and solution quality problems. The solution processing in each method is very unique. The EP [27,28] has the advantages of good convergent property and a signicant speedup over traditional GAs and can obtain high quality solutions. The Curse of dimensionality is surmounted, and the computational burden is almost linear with the problem scale. From the literature review, it has been observed that there exists a need for evolving simple and eective methods, for obtaining an optimal solution for the UCP. Hence, in this paper, an attempt has been made to couple EP with SA for meeting these requirements of the UCP, which eliminates the above-mentioned drawbacks. In case of SA, the temperature and demand are taken as control parameter. Hence the quality of solution is improved. The algorithm is based on the annealing neural network. Classical optimisation methods are a direct means for solving this problem. EP seems to be promising and is still evolving. EP has the great advantage of good convergent property and, hence, the computation time is considerably reduced. The EP combines good solution quality for SA with rapid convergence for EP. The EP Based SA (EPSA) is used to nd the short-term thermal unit commitment. By doing so, it can help to nd the optimum solution rapidly and eciently.

542

C.C.A. Rajan, M.R. Mohan / Electrical Power and Energy Systems 29 (2007) 540550

EP is capable of determining the global or near-global solution. It is based on the basic genetic operation of human chromosomes. It operates with the stochastic mechanics, which combine ospring creation based on the performance of current trail solutions, competition and selection based on the successive generations, form a considerably robust scheme for large-scale real-valued combinational optimisation. In this proposed work, the parents are obtained from a pre-dened set of solutions, i.e. each and every solution is adjusted to meet the requirements. And the selection process is done using Evolutionary Strategy [3133]. The application on the NTPS shows that we can nd the optimal solution eectively and these results are compared with the conventional methods. 2. Problem formulation The objective is to nd the generation scheduling such that the total operating cost can be minimized, when subjected to a variety of constraints [29]. In the UCP under consideration, an interesting solution would be minimizing the total operating cost of the generating units with several constraints being satised. The major component of the operating cost, for thermal and nuclear units, is the power production cost of the committed units and this is given in a quadratic form: F it P it Ai P 2 Rs=h 1 it Bi P it C i where Ai, Bi, Ci is the cost function parameters of unit i (Rs/MW2 h, Rs/MW h, Rs/h), Fit(Pit) the production cost of unit i at a time t (Rs/h) and Pit output power from unit i at time t (MW) nomenclature. The start-up cost depends upon the down time of the unit, which can vary from a maximum value, when the unit i is started from cold state, to a much smaller value, if the unit i has been turned o recently. The start-up cost calculation depends upon the treatment method for the thermal unit during down time periods. The start-up cost Sit, is a function of the down time of unit i as given in the following equation: S it Soi 1 Di expT off i =T downi Ei Rs 2 where Soi is the unit i cold start-up cost (Rs) and Di, Ei is the start-up cost coecients for unit i. The overall objective function of the UCP is FT
T X N X t 1 i 1

being the load balance constraints and the spinning reserve constraints. The other constraints include the thermal constraints, fuel constraints, security constraints etc. [29]. 2.1.1. Load balance constraints The real power generated must be sucient enough to meet the load demand and must satisfy the following factors:
N X i1

P it U it PDt

where PDt is the system peak demand at hour t (MW), N the number of available generating units, U(0, 1) the uniform distribution with parameters 0 and 1 and UD(a, b) is the discrete uniform distribution with parameters a and b. 2.1.2. Spinning reserve constraints The spinning reserve is the total amount of real power generation available from all synchronized units minus the present load plus the losses. The reserve is considered to be a pre-specied amount or a given percentage of the forecasted peak demand. It must be sucient enough to meet the loss of the most heavily loaded unit in the system. This has to satisfy the equation given below:
N X i1

P maxi U it P PDt Rt ;

16t6T

where P maxi is the maximum generation limit of unit i, Rt the spinning reserve at time t (MW) and T is the scheduled time horizon (24 h). 2.1.3. Thermal constraints The temperature and pressure of the thermal units vary very gradually and the units must be synchronized before they are brought online. A time period of even 1 h is considered as the minimum down time of the units. There are certain factors, which govern the thermal constraints, like minimum up time, minimum down time and crew constraints. 2.1.3.1. Minimum up time. If the units have already been shut down, there will be a minimum time before they can be restarted and the constraint is given in below: T oni P T upi 6

F it P it U it S it V it Rs=h

where Uit is the unit i status at hour t = 1 (if unit is ON) = 0 (if unit is OFF), Vit the unit i start up/shut down status at hour t = 1, if the unit is started at hour t and 0 otherwise, FT the total operating cost over the schedule horizon (Rs/ h) and Sit is the start-up cost of unit i at hour t (Rs). 2.1. Constraints Depending on the nature of the power system under study, the UCP is subject to many constraints, the main

where T oni is the duration for which unit i is continuously ON (h) and T upi is the unit i minimum up time (h). 2.1.3.2. Minimum down time. If all the units are running already, they cannot be shut down simultaneously and the constraint is given below: T off i P T downi 7

where T downi is the unit i minimum down time (h) and T downi is the duration for which unit i is continuously OFF (h).

C.C.A. Rajan, M.R. Mohan / Electrical Power and Energy Systems 29 (2007) 540550

543

2.1.4. Must run units Generally in a power system, some of the units are given a must run status in order to provide voltage support for the network. 3. Simulated annealing Simulated annealing (SA) is the powerful general-purpose stochastic search technique to solve hard-constrained optimization problems. This SA procedure has been successfully applied to a range of combinatorial problems in electrical engineering including power systems. 3.1. Physical concepts of simulated annealing Annealing, in physical terms, refers to the process of heating up a solid to a high temperature followed by slow cooling achieved by decreasing the temperature of the environment in steps. At each step the temperature is maintained constant for a period of time sucient for the solid to reach thermal equilibrium [1619]. At equilibrium, the solid could have many congurations; each corresponding to dierent spins of the electrons and to specic energy levels. At equilibrium, the probability of a given conguration, Pcong, is given by Boltzmann distribution: P config K expEconfig =C p 8 where Econg is the energy of the given conguration and K is a constant. Metropolis et al. proposed a Monte Carlo method to simulate the process of reaching thermal equilibrium at a xed temperature Cp. In this method, a randomly generated perturbation of the current conguration of the solid is applied so that a trial conguration is obtained. Let Ec < Et, till a lower energy level has been reached and the trial solution has to be altered. If Ec > Et, then the trial conguration is accepted as the current conguration with probability given below: expEc Et =C p 9

another, reaching the minimum energy of the system. This analogy can be stated as: Solutions in the combinatorial optimisation problems are equivalent to states (congurations) of the physical system. The cost of a solution is equivalent to the energy of a state. Demand as a control parameter is introduced to play the role of temperature in the annealing process. In applying the SAA to solve the combinatorial optimisation problems, the basic idea is to choose a feasible solution at random and then get a neighbour to this solution. A move to this neighbour is performed if it has a better lower objective value or, in case the neighbour has a higher objective function value, if exp (DE/Cp) > = U(0, 1), where DE is the increase in objective value in the neighbour. The eect of decreasing Cp is that the probability of accepting an increase in the objective function value is decreased during the search. The most important part in the SAA is to have a good rule for nding a diversied and intensied neighbourhood so that a large amount of the solution space can be explored. Another important part is how to choose the initial value of Cp and how Cp, should decrease during the search. SA means a simulation of the annealing process of metal. If the temperature is lowered carefully from a high temperature in the annealing process, the method metal will produce the crystal at 0 K. Kirkpatrick [17] developed an algorithm that nds the near-optimal solution by substituting the random movement of solution for the uctuation of particle of system in the annealing process and by making the objective function value correspond to the energy of the system which decreases with the descent of temperature. The owchart of SA method is shown in Fig. 1.

3.3. Simulated annealing algorithm for UCP Step (0): Find the initial feasible solution by optimum allocation. Step (1): Demand and temperature are taken as the control parameter. Step (2): Generate the trial solution. Step (3): Check for the stopping criterion. (a) If satised, go to the next hour for checking the same. (b) Else, decrement the system peak demand for that instant and again generate the trial solution. Step (4): Get the optimal schedule and (a) Assuming the fuel cost to be constant per hour. Equate the total power demand to the total no of units switched ON. (b) From the total fuel cost subtract the constant cost function C. Equate the remaining cost value to the ON units equally.

where Cp is the control parameter of the cooling schedule. The process continues where a transition to a conguration of higher energy level is not necessarily rejected. Eventually thermal equilibrium is achieved after a large number of perturbations, where the probability of a conguration approaches Boltzmann Distribution. By gradually decreasing Cp and repeating Metropolis simulation, new lower energy levels become achievable. As Cp approaches zero, the least energy congurations will have a positive probability of occurring. 3.2. Applications of simulated annealing to combinational optimisation problems By making an analogy between the annealing process and the optimisation problem, a great class of combinatorial optimisation problems can be solved following the same procedure of transition from equilibrium state to

544

C.C.A. Rajan, M.R. Mohan / Electrical Power and Energy Systems 29 (2007) 540550

3.5. Generating an initial solution The SA algorithm requires a starting feasible schedule, which satises all the system and units constraints. This schedule is randomly generated. The algorithm given in [16] is used for nding this starting solution. 3.6. Operating cost calculation Once a trail solution is obtained, the corresponding total operating cost is determined. Since the production cost is a quadratic function, the EDP is solved using a quadratic programming routine. The start-up cost is then calculated for the given schedule. 3.7. Stopping criteria There may be several stopping criteria for the search. For this implementation, the search is stopped if the following conditions are satised: The load balance constraints are satised. The spinning reserve constraints are satised.

3.8. Implementation To solve for the UCP using SA, software in Turbo C package is developed. The software provides interactive approach in dealing with the various data input required for solving the UCP from the user. 4. Evolutionary programming (c) Assume the initial temperature of the turbine as 660 and a generation of 210 MW. (d) If the demand decreases/increases, then the temperature should decrease/increase by an amount C pk1 C pk =1 C pk ln1 d=3rC pk where Cp is the control temperature (), d is the distance parameter and r is the standard deviation of the cost value generated. Step (5): Calculate the total operating cost Ft as the summation of running cost and start-up shut down cost by Eq. (3). 4.1. Introduction EP is a mutation-based evolutionary algorithm applied to discrete search spaces. David Fogel (Fogel 1988) extended the initial work of his father Larry Fogel (Fogel, 1962) for applications involving real-parameter optimization problems. Real-parameter EP is similar in principle to evolution strategy (ES), in that normally distributed mutations are performed in both algorithms. Both algorithms encode mutation strength (or variance of the normal distribution) for each decision variable and a self-adapting rule is used to update the mutation strengths. Several variants of EP have been suggested (Fogel, 1992). 4.2. Evolutionary strategies For the case of Evolutionary strategies Fogel remarks evolution can be categorized by several levels of hierarchy: the gene, the chromosome, the individual, the species, and the ecosystem. Thus, while Genetic Algorithms stress models of genetic operators, Evolutionary Strategies emphasize mutational transformation that maintains behavioral linkage between each parent and its ospring

Fig. 1. Flow chart of simulated annealing algorithm.

3.4. Generating trial solution The neighbors should be randomly generated, feasible, and span as much as possible the problem solution space. Because of the constraints in the UCP this is not a simple matter. The most dicult constraints to satisfy are the minimum up/down times. The implementation of new rules to obtain randomly feasible solutions faster are done by the rules is described in [16].

C.C.A. Rajan, M.R. Mohan / Electrical Power and Energy Systems 29 (2007) 540550

545

at the level of the individual. Evolutionary Strategies are a joint development of Bienert, Rechenberg, and Schwefel. The rst applications were experimental and addressed some optimization problems in hydrodynamics. 4.3. EP general algorithm Evolutionary programming [27], [3133] is conducted as a sequence of operations and is given below. The schematic diagram of the EP algorithm is shown in Fig. 2.

(1) The initial population is determined by setting si = Si $ U(ak, bk)k, i = 1, . . . , m, where Si is a random vector, si is the outcome of the random vector, U(ak, bk)k denotes a uniform distribution ranging over [ak, bk] in each of k dimensions, and m is the number of parents. (2) Each si, i = 1, . . . , m, is assigned a tness score #(si) = G(F(si), vi), where F maps si ! R and denotes the true tness of si, vi, represents random alteration in the instantiation of si, random variation imposed

Fig. 2. Flowchart for EPSA for UCP.

546

C.C.A. Rajan, M.R. Mohan / Electrical Power and Energy Systems 29 (2007) 540550

on the evaluation of F(si), or satises another relation si, and G(F(si), vi) describes the tness score to be assigned. In general, the functions F and G can be as complex as required. For example, F may be a function not only of a particular si, but also of other members of the population, conditioned on a particular si. (3) Each si, i = 1, . . . , m, is altered and assigned to si+m such that si m si;j N 0; bj#si zj ; j 1; . . . ; k

(5) Create the ospring trail solution p0i using the following steps. (a) Calculate the standard deviation rj bF Tij = minF Ti P j max P j min (b) Add a Gaussian random variable N (0, r2 j to all the state variable of pi, to get p0i . Select the rst Np individuals from the total 2Np individuals of both pi and p0i using the following steps for next iteration. (a) Evaluate r = (2Np random (0, 1) + 1) (b) Evaluate each trail vector by Wpi = sum(Wx), where x = 1, 2, . . . , Np, i = 1, 2, . . . , 2Np such that Wx = 1 if FTij/(FTij + FTir) < random (0, 1), otherwise, Wx = 0. Sort the Wpi in descending order and the rst Np individuals will survive and are transcribed along with their elements to form the basis of the next generation. The above procedure is repeated from step (2) until a maximum number of generations Nm is reached. Selection process is done using Evolutionary strategy.

(6)

N (0, bj#(si) + zj) represents a Gaussian random variable with mean l and variance r2, bj is a constant of proportionality to scale #(si), and zj represents an oset to guarantee a minimum amount of variance, (4) Each si+m, i = 1, . . . , m, is assigned a tness score #sim GF sim ; vim (5) For each si, i = 1, . . . , 2m, a value wi is assigned according to wi w t
c X

(7)

(8) (9)

w t

t 1

1; if #si 6 #si ; 0; otherwise;

5. Evolutionary programming based simulated annealing for UCP 5.1. Simulated annealing (1) Take the parent as the initial feasible solution. (2) Take demand and temperature as control parameter and generate the trial solution. (3) Check for the stopping criterion. (4) If false, decrement system peak demand and go to step 2. (5) If true, generate the optimal solution and calculate the total operating cost. 5.2. EP based SA

where q = [2mu1 + 1], q 5 i, [x] denotes the greatest integer less than or equal to x, c is the number of competitions, and u1 $ U (0,1). (6) The solutions si, I = 1, . . . , 2m, are ranked in descending order of their corresponding value Wi [with preference to their actual scores #(si) if there are more than m solutions attaining a value of c]. The rst m solutions are transcribed along with their corresponding values #(si) to be the basis of the next generation. (7) The process proceeds to step 3, unless the available execution time is exhausted or an acceptable solution has been discovered.

4.4. Evolutionary programming for UCP (1) Initialize the parent vector p = [p1, p2, . . . , pn], i = 1, 2, . . . , Np such that each element in the vector is determined by pj $ random (pjmin, pjmax), j = 1, 2, . . . , N, with one generator as dependent generator. (2) Initialize the parent vector p = [p1, p2, . . . , pn], i = 1, 2, . . . , Np such that each element in the vector is determined by pj $ random (pjmin, pjmax), j = 1, 2, . . . , N, with one generator as dependent generator. (3) Calculate the overall objective function if the UCP is given in Eq. (3) using the trail vector pi and nd the minimum of FTi. (4) Initialize the parent vector p = [p1, p2, . . . , pn], i = 1,2, . . . , Np such that each element in the vector is determined by pj $ random (pjmin, pjmax), j = 1, 2, . . . , N, with one generator as dependent generator.

In the SA technique for solving UCP, Initial Operating Schedule status in terms of maximum real power generation of each unit is given as input. As we know that SA is used to improve any given status by avoiding entrapment in local minima, the ospring obtained from the EP algorithm is given as input to SA and the rened status is obtained. And Evolutionary Strategy selects the nal status. (1) Get the demand for 24 h and number of iterations to be carried out. (2) Generate population of parents (N) by adjusting the existing solution to the given demand to the form of state variables. (3) Unit down time makes a random recommitment. (4) Check for constraint in the new schedule. If the constraints are not met then repair the schedule as given below in Section 5.3.

C.C.A. Rajan, M.R. Mohan / Electrical Power and Energy Systems 29 (2007) 540550

547

(5) Perform ELD and calculate total production cost for each parent. (6) Add the Gaussian random variable to each state variable and hence create an ospring. This will further undergo for some repair operations as given below in Section 5.4. Following these, the new schedules are checked in order to verify that all constraints are met. (7) Improve the status of the evolved ospring and verify the constraints by SA. (8) Formulate the rank for the entire population. (9) Select the best N number of population for next iteration. (10) Has iteration count reached? If yes go to step 11 else go to step 2. (11) Select the best population(s) by Evolutionary strategy [3133].

(12) Print the optimum schedule. The ow chart for EPSA for UCP is shown in Fig. 2. 5.3. Repair mechanism A repair mechanism to restore the feasibility of the constraints is applied and described as follows: Pick at random one of the OFF units at one of the violated hours. Apply the rules in Section 3.3 to switch the selected unit from OFF to ON keeping the feasibility of the down time constraints. Check for the reserve constraints at this hour. Otherwise repeat the process at the same hour for another unit.

Table 1 Daily generation of seven units in MW Hour 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 Pmax 840 757 775 773 770 778 757 778 770 764 598 595 545 538 535 466 449 439 466 463 460 434 530 840 I 60 60 60 60 60 60 60 60 60 60 60 60 0 0 0 0 0 0 0 0 0 0 60 60 II 80 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 80 III 100 100 100 100 100 100 100 100 100 100 99 100 100 99 100 0 0 0 0 0 0 0 0 100 IV 101 100 115 113 110 118 100 118 110 104 97 96 99 97 96 116 101 97 116 113 110 95 120 101 V 149 147 150 150 150 150 147 150 150 150 142 139 146 142 139 150 148 142 150 150 150 139 150 149 VI 150 150 150 150 150 150 150 150 150 150 0 0 0 0 0 0 0 0 0 0 0 0 0 150 VII 200 200 200 200 200 200 200 200 200 200 200 200 200 200 200 200 200 200 200 200 200 200 200 200

Table 3 Comparison of costs and CPU time for NTPS, 10, 26, 34 unit systems System 7 Unit (NTPS) Methods SA TS EP EPSA DP LR SA TS EP EPSA DP LR SA TS EP EPSA DP LR SA TS EP EPSA DP LR Total cost (pu) 0.94780 0.95531 0.93461 0.92592 1.00000 0.97843 0.94422 0.95322 0.94071 0.92290 1.00000 0.97183 0.93906 0.94887 0.92814 0.91090 1.00000 0.96642 0.93500 0.94382 0.92400 0.90918 1.00000 0.96197 CPU time (s) 84 110 66 60 130 115 200 230 186 180 260 235 1810 1820 1785 1778 1860 1878 6800 6842 6792 6776 6865 6824

10 Unit

26 Unit

34 Unit

Table 2 Generation system operation data Unit Pmin (MW) 15 20 30 25 50 50 75 Pmax(MW) 60 80 100 120 150 150 200 Running cost Ci (Rs) 1 2 3 4 5 6 7 750 1250 2000 1600 1450 4950 4100 BI (Rs/MW h) 70 75 70 70 75 65 60 Ai (Rs/MW h ) 0.255 0.198 0.198 0.191 0.106 0.0675 0.074
2

Start-up cost Soi (Rs) 4250 5050 5700 4700 5650 14100 11350 Di (Rs) 29.5 29.5 28.5 32.5 32 37.5 32 Ei (Rs) 10 10 10 9 9 4.5 5.5

548

C.C.A. Rajan, M.R. Mohan / Electrical Power and Energy Systems 29 (2007) 540550

5.4. Making ospring feasible While solving the constrained optimization problem, there are various techniques to repair an infeasible solution [27,31]. In this paper we have chosen the technique, which evolve only the feasible solutions. That is the schedule, which satises the set of constraints as mentioned earlier. Here, in this paper, the selection routine is involved as culling force to eliminate the feasible schedules. Before the best solution is selected by evolutionary strategy, the trail is made to correct the unwanted mutations.

5.5. Implementation Software programs were developed using MATLAB software package and the test problem was simulated for 10 independent trials using Evolutionary Programming Algorithm. The training and identication part as implemented in the SA technique is employed here and considered as a process involving random recommitment, constraint verication and ospring creation. 6. Numerical results A NTPS in India with seven generating units, each with a capacity of 210 MW, has been considered as a case study. A time period of 24 h is considered; the unit commitment problem is solved for these seven units and also compared with 10, 26, and 34 generating unit power systems. The required inputs for solving the UCP are briefed here. The total number of generating units, the maximum real power generation of each unit and the cost function parameters of each unit are tabulated for a day, respectively, as shown in Tables 1 and 2 for NTPS. The status of unit i at time t and the start-up/shut-down status obtained are the necessary solution for SA, TS, EP, EPSA, DP, LR methods for NTPS. The comparison of the total costs and central processing unit (CPU) time is shown in Table 3 for NTPS, 10, 26, and 34 generating unit power systems.

Fig. 3. Total production cost for two iterations.

Fig. 6. Total production cost for 10 iterations. Fig. 4. Total production cost for three iterations.

Fig. 5. Total production cost for four iterations.

Fig. 7. Total production cost for 20 iterations.

C.C.A. Rajan, M.R. Mohan / Electrical Power and Energy Systems 29 (2007) 540550

549

Figs. 37 represents the total production cost obtained by each parent for 24, 10 and 20 iterations in EP method. Fig. 8 gives the plot of EP average performance from 100 runs. The Fig. 9 gives the plot of number of iteration versus the time taken to complete those iterations and the maximum production cost obtained under each iteration. From these results, the EPSA method had lesser total cost and took lesser CPU time in all the power systems considered including NTPS. As we indicated in the paper, the EP algorithm has also proved to be an ecient tool for solving the important economic dispatch problem for units with nonsmooth fuel cost functions as referred in [28]. Such functions may be included in the proposed EP search for practical problem solving. There is no obvious limitation on the size of the problem that must be addressed, for its data structure is such that the search space is reduced to a minimum; no relaxation of constraints is required; instead, populations of feasible solutions are produced at each generation and through out the evolution process. The main advantages of the proposed algorithm are speed. The proposed EPSA approach was compared to the related methods in the references indented to serve this purpose, such as the DP with a zoom feature, the SA, and the GA approaches. Further SA can start with any initial solution and improves on it to nd optimal solution with a high probability. By means of stochastically searching multiple points at one time and considering trail solutions of successive generations, the EPSA approach avoids entrapping in local optimum solutions. Also, disadvantages of huge memory size required by the SA method are eliminated. Moreover, intellectual schemes of encoding and decoding entailed by the GA approach are not needed in the proposed EPSA approach. The problem of power unbalance

Fig. 9. Number of iterations vs time taken and maximum production cost.

previously existing in the solution of GA is circumvented as well in this paper. In comparison with the results produced by the referenced techniques, the EPSA method obviously displays a satisfactory performance with respect to the quality of its evolved solutions and to its computational requirements. 7. Conclusion This paper presents an EP based SA method to the unit commitment problem. In this method, the essential processes simulated in the procedure are mutation, competition, and selection. The mutation rate is computed as a function of the ratio of the total cost by the schedule of interest to the cost of the best schedule in the current population. Competition and selection are applied to select from among the parents and the ospring, the best solutions to form the basis of the subsequent generation. In this proposed work, the parents are obtained from a pre-

2.6

x 10

2.55

Average Best Worst

Total cost (Rs)

2.5

2.45

2.4

2.35

2.3

100

200

300

400

500

600

700

800

900

1000

No.of generations
Fig. 8. EP average performance from 100 runs.

550

C.C.A. Rajan, M.R. Mohan / Electrical Power and Energy Systems 29 (2007) 540550 [11] Cohen AI, Yoshimura M. A branch and bound algorithm for unit commitment. IEEE Trans Power Apparatus 1983;102 (2):44451. [12] Ouyang Z, Shahidehpour SM. Short term unit commitment expert system. Int J Electr Power Syst Res 1990;20:113. [13] Su CC, Hsu YY. Fuzzy dynamic programming: an application to unit commitment. IEEE Trans Power Syst 1991;6 (3):12317. [14] Sasaki H, Watanabe M, Kubokawa J, Yorino N. A solution method of unit commitment by articial neural networks. IEEE Trans Power Syst 1992;7 (3):97481. [15] Padhy NP. Unit commitment using hybrid models: a comparative study for dynamic programming, expert systems, fuzzy system and genetic algorithms. Int J Electr Power Energy Syst 2000;23 (1):82736. [16] Zhuang F, Galiana FD. Unit commitment by simulated annealing. IEEE Trans Power Syst 1990;5 (1):3118. [17] Kirkpatrick S, Gelatt Jr CD, Vecehi MP. Optimisation by SA. Science 1983;220:4598. [18] Selim Shokri Z, Alsultan K. A simulated annealing algorithm for the clustering problem. Pattern Recognit 1991;24 (10):10038. [19] Mantawy AH, Abdel-Magid Youssef L, Selim Shokri Z. A simulated annealing algorithm for unit commitment. IEEE Trans Power Syst 1998;13 (1):197204. [20] Mantawy AH, Abdel-Magid Youssef L, Selim Shokri Z. A unit commitment by Tabu Search. IEE Proc Generat Transmission Distribut 1998;145 (1):5664. [21] Lin Whei-Min, Cheng Fu-Sheng, Tsay Ming-Tong. An improved Tabu search for economic dispatch with multiple minima. IEEE Trans Power Syst 2002;17 (1):10812. [22] Bai X, Shahidehpour M. Hydro-thermal scheduling by Tabu search and decomposition method. IEEE Trans Power Syst 1996;11 (2):96875. [23] Wu Yong-Gang, Ho Chun-Ying, Wang Ding-Yi. A diploid genetic approach to short-term scheduling of hydro-thermal system. IEEE Trans Power Syst 2000;15 (4):126874. [24] Hong Ying-Yi, Li Chih-Yuan. Genetic algorithm based economic dispatch for cogeneration units considering multiplant multibuyer wheeling. IEEE Trans Power Syst 2002;17 (1):13440. [25] Mantawy AH, Abdel-Magid Youssef L, Selim Shokri Z. Integrating genetic algorithm, Tabu search and simulated annealing for the unit commitment problem. IEEE Trans Power Syst 1999;14 (3):82936. [26] Wong Kit Po, Suzannah, Wong Yin Wa. Combined genetic algorithm/simulated annealing/fuzzy set approach to short-term generation scheduling with take-or-pay fuel contract. IEEE Trans Power Syst 1996;11 (1):12835. [27] Juste KA, Kita H, Tanaka E, Hasegawa J. An evolutionary programming solution to the unit commitment problem. IEEE Trans Power Syst 1999;14 (4):14529. [28] Yang HT, Yang PC, Huang CL. Evolutionary programming based economic dispatch for units with non-smooth fuel cost functions. IEEE Trans Power Syst 1996;11 (1):1127. [29] Wood AJ, Woolenberg BF. Power generation and control. 2nd ed. New York: John Wiley and Sons; 1996. [30] Arts E, Korst Jan. A stochastic approach to combinatorial optimization and neural computing. New York: John Wiley & Sons Ltd; 1989. [31] Fogel DB. Evolutionary computation. Toward a new philosophy of machine intelligence. New York: IEEE Press; 1995. [32] Back T. Evolutionary algorithms in theory and practice. New York: Oxford University Press; 1996. [33] Fogel LJ, Owens AJ, Walsh MJ. Articial intelligence through simulated evolution. New York: John Wiley & Sons Inc.; 1996.

dened set of solutions, i.e. each and every solution is obtained from the SA method. Then, a random recommitment is carried out with respect to the units minimum down times. And the selection process is done using Evolutionary Strategy. In comparison with the results produced by the referenced techniques (EP, DP, LR and SA and TS), the EPSA method obviously displays a satisfactory performance. There is no obvious limitation on the size of the problem that must be addressed, for its data structure is such that the search space is reduced to a minimum; No relaxation of constraints is required; instead, populations of feasible solutions are produced at each generation and throughout the evolution process; Multiple near optimal solutions to the problem involving multiple constraints and conicting objectives can be obtained in a reasonable time with the use of heuristics; It works only with feasible solutions generated based on heuristics, thus avoiding the computational burden entailed by the GA methods which rst generate all feasible solutions and then purge the infeasible ones. References
[1] Snyder WL, Powell HD, Rayburn C. Dynamic programming approach to unit commitment. IEEE Trans Power Syst 1987;3 (2):33950. [2] Hsu YY, Liang CC, Lin CJ, Huang CT. Dynamic security constrained multi-area unit commitment. IEEE Trans Power Syst 1991;6:104955. [3] Dillon TS, Edwin KW, Kochs HD, Taud RJ. Integer programming approach to the problem of optimal unit commitment with probabilistic reserve determination. IEEE Trans Power Apparatus Syst 1978;97 (6):215666. [4] Takriti Samer, Birge John R. Using integer programming to rene Lagrangian-based unit commitment solutions. IEEE Trans Power Syst 2000;15 (1):1516. [5] Mohan MR, Kuppusamy K, Abdullah Khan M. Optimal short-term hydro-thermal scheduling using decomposition approach and linear programming method. Int J Electr Power Energy Syst 1992;14 (1):3944. [6] Ngundam JM, Kenfack F, Tatietse TT. Optimal scheduling of largescale hydrothermal power systems using the Lagrangian relaxation technique. Int J Electr Power Energy Syst 2002;22 (1):23745. [7] Bakirtzis AG, Zoumas CE. Lambda of Lagrangian relaxation solution to unit commitment problem. IEE Proc Generat Transmission Distribut 2000;147 (2):1316. [8] Chang Chuan-ping, Liu Chih-wen, Liu Chun-chang. Unit commitment by Lagrangian relaxation and genetic algorithms. IEEE Trans Power Syst 2000;15 (2):70714. [9] Nieva R, Inda A, Guillen I. Lagrangian reduction of search range for large scale unit commitment. IEEE Trans Power Syst 1987;2:46573. [10] Egan Gerard T, Dillon Tharam S, Morsztyn Karol. An experimental method of determination of optimal maintenance schedules in power systems using the branch and bound technique. IEEE Trans Syst Man Cybernet 1976;6 (8):53847.