IEEE Transactions on Power Systems, Vol. 14, No. 1, February 1999 299 zy Generation Expansion Planning Based on an Advanced Evolutionary Programming Young-Moon Park** Jong-Ryul Won* Jong-Bae Park*** Dong-Gee Kim** Fellow Student Member zyxw * Korea Electric Power Research Institute, Taejon, 305-380, Korea School of Electrical Engineering, Seoul National University, Seoul, 151-742, Korea Electrical Engineering Department, Anyang University, Anyang, 708- 113, Korea ** *** Abstract : This paper proposes an efficient evolutionary programming algorithm for solvmg a generation expansion planning (GEP) problem known as a highly-nonlinear dynamic problem. Evolutionary programming (EP) is an optimization algorithm based on the simulated evolution (mutation, competition and selection). In this paper, some improvements are presented to enhance the efficiency of the EP algorithm fox solving the GEP problem. First, by a dom,iinmapping procedumkearly cumulative capacity vectors are transformed into one dummy vector, whose change can yield a kind of trend in the cost value. Next quadratic approximation technique and tournament selection are utilized. To validate the proposed approach, these algorithms are tested on two cases of expansion planning problems. Sirnulation red@ show that the proposed algorithm can provide successful results within a reasonable coniputational time compared with conventionh EP and dynamic programming. Keywords: Efficient evolutionary programming, Generation expansion planning, Domain mapping procedure, Quadratic approximation technique, Toumanient selection I. ‘INTRODUCTION Generation expansion planning (GEP) is one of the most important planning activities in the electric utilities. Optimal long-term generation expansion planning is to determine the least-cost capacity addition schedule zyxwvuts (i.e., the type and number of each candidate plant) that satisfies forecasted load demands within the given reliability criteria over a planning horizon (typically 10 to 20 years). Thus, the objective function of the least-cost GEP problem has been the expected sum of yearly discounted costs which incorporate construction costs, operation costs, salvage value and so on. A long-term GEP prolblem is a highly-constrained nonlinear zyxwvu PE-304-PWRS-0-2-1998 A paper recommended and approved by the IEEE Power System Operations Committee of the IEEE Power Engineering Society for publication in the IEEE Transactions on Power Systems. Manuscript submitted July 29, 1997; made available for printing January 20, 1998. discrete dynamic optimization problem which can only be fully solved by complete enumeration in its nature [l-51. Therefore, getting an optimal plan is expected to investigate every possible combination of candidate options over the planning horizon. However, it requires enormous calculation. The high nonlinearities of this problem are basically originated from the probabilistic production costing simulation [1,2,6], and a set of physical nonlinear constraints. To solve this complicated problem, a number of salient optimization methods have been successfully applied during the past decades. Park et. al. applied the Pontryagin’s maximum principle for dynamic Optimization whose solution is in a continuous space [7]. Bloom applied the mathematical programming using a decomposition method, solving it in a continuous space [3]. Dynamic programming (DP) based framework is one of the most popular algorithms in a generation expansion planning problem [ 1,2,4,5,8,9,10]. However, the so-called ‘curse of the dimensionality’ has caused important problem in the direct application of conventional DP to a practical GEP problem [1,2,4,5]. David et. al. developed a heuristic-based DP and applied the fuzzy set theory to reduce the number of configurations [4,5]. Recently, Fukuyama et. al. [ l l ] and Park et. al. [12] applied genetic algorithm to solve a simple GEP problem, and showed promising good performances. Evolutionary computations (ECs) are emerging as efficient approaches for various search, classification and optimization problems. ECs emphasize adaptation and operation in a manner analogous to biological evolution. The most popular evolutioiiary computational models are evolutionary programming (EP), evolutionary strategies (E%) and genetic algorithms (GAS) [13-181. All these are optimizalion algorithms which simulate the evolution of individual to find the global optimal solution. The main advantages of these search algorithms lie in their global convergence, inherent parallel processing nature, problem independence and great robustness. Among these ECs techniques, EP emphasizes the evolution of solutions through mutation by Gaussian distribuQon and competitive selection [ 161. Recently, global optimization techniques 0885-8950/99/$10.00 zyxwvu 0 1998 IEEE