Optimization Parameter Variation: Improving Biobjective Optimization of Temporary Facility Planning

Abstract

Recent research interest in nondominated sorting genetic optimization has increased remarkably due to its effectiveness in handling multiple objectives without scalarization. Algorithms for these evolutionary optimization methods have been used to solve a wide variety of civil engineering optimization problems. Despite the effectiveness of these algorithms, their optimization parameter tuning is a time-consuming process that, if improperly executed, can hinder performance and lead to premature convergence toward incomplete, suboptimal Pareto fronts. Existing tuning methods either require a substantial amount of time or rely on inadequate methods. The main objective of this study is to investigate the performance of these algorithms under the effect of optimization parameter variation. Using a biobjective benchmark problem on optimizing temporary facility planning, the study specifically investigates the effects of varying the population size, number of generations, crossover type, probability of crossover, and mutation rate on the algorithm behavior. The results are used to recommend strategies for parameter tuning in order to accelerate convergence toward optimal Pareto fronts and improve solution dispersion ranges for biobjective optimization problems. The findings of this study should prove most useful to scholars and construction planners, especially those who are involved in temporary facility planning.