| Abstract | In this study, we address the long-term power generation expansion planning (PGEP) problem which is a highly constrained nonlinear discrete optimization problem. The problem is formulated into a mixed integer nonlinear programming (MINLP) model that determines the most economical investment plan for additional thermal power generating units. Specifically, with the incorporated environmental costs, the optimal generation level of each plant over a planning horizon is determined, subject to the integrated requirements of power demands, power capacities, loss of load probability (LOLP) levels, locations, and environmental limitations for emission controls. To handle such problem complexity, a decomposition scheme is applied by dividing the problem into two subproblems. Firstly, the investment subproblem is an integer program (IP) related to the investment plan to select the new power plants for construction. Secondly, the operation subproblem is a linear program (LP) deals with the operational plan of those selected power plants. Based on the Genetic Algorithm (GA) and Benders' Decomposition (BD), we propose three alternative solution methods for solving the formulated PGEP model namely a Genetic Algorithm (GA), GA-Benders' Decomposition (GA-BD) and GA-BD-LP Decomposition (GA-BD-LPD). For the proposed GA-based procedure, firstly it solves for a generation mix the combinatorial or IP part of solutions, and secondly given each combinatorial solution, it solves the LP subproblem for other continuous decision variables. The proposed GA is subsequently evaluated for its effectiveness and efficiency on seven small- to medium¬sized case studies of Thailand. The computational experiments show that the proposed GA-heuristic-based method can solve the PGEP problem effectively with significant saving in runtime when compared to the results obtained from a commercial optimization package LINGO. The proposed GA-BD heuristic, a combination of GA and Benders' Decomposition methods, applies the decomposition structure of Bender to the PGEP problem. Similar to IP, the master problem (MP) deals with the integer variables representing the selection of power generating plants, but also includes the Benders' cuts as its additional constraints. Given each investment plan prescribed by MP, an LP subproblem is solved for an optimal operational plan. Each combined (MP and LP) solution is used to generate a Benders' cut for MP in the next iteration. Instead of solving MP to optimality, we propose using GA in solving MP. For each of the q best MP chromosomes, the corresponding LP is solved to optimality and a Benders' cut is consequently constructed and appended to MP in the next generation. As such, both MP and LP are iteratively and alternately solved, until a stopping criterion is met, either by reaching the maximum number of generations or no improvement for a certain number of generations. To demonstrate the effectiveness and efficiency of the proposed GA-BD algorithm, we conduct a computational experiment on the previous case studies and compare the results with those of the previously proposed GA method. Moreover, we apply the GA-BD to two larger-scale case studies of Thailand using two problems of large size and extra large size. They will be referred to as the real cases. The proposed GA-BD can successfully solve the large problem; nevertheless it cannot solve the extra large problem. To handle the extra large size problems, the GA-BD-LPD, is proposed by extending the proposed GA-BD. The method decomposes the LP problem into many small LP subproblems, solving each LP subproblem, and then re-assembling all of those small LP solutions. The results of the proposed GA-BD-LPD approach show better performance in terms of solution quality and computational runtimes when compared to the proposed GA-BD that cannot even solve the real extra large problem. To show the performance of the proposed GA-BD-LPD, the computational experiments are conducted on all the problems: the large, medium, and small problems; and finally, we compare the results obtained by the proposed GA-BD-LPD to those obtained by the proposed GA-BD. For the comparison of the capability and suitability of each proposed solution approach, it can be concluded that the most suitable approach for solving the large and extra large (the real case) problems is the GA-BD-LPD, as well as for the medium problem. But for the small problem, the GA-BD method is more suitable than the other proposed methods. The study, it showed that the proposed GA-heuristic-based approaches yield high¬quality solutions in reasonable computational runtimes, for solving the PGEP problem. The proposed GA-BD-LPD is the most efficient method for solving the PGEP problem, followed by the proposed GA-BD and the proposed GA respectively |