Improving the Performance of NSGA III using Latin Hypercube Sampling
Shilpi Jain, Kamlesh Kumar Dubey · Research Square · 2022
Abstract Performance of multi-objective optimization techniques mainly depends on the convergence and the diversity of obtained Pareto-optimal solutions. A well-known non-dominated sorting genetic algorithm-III (NSGA-III) is a meta-heuristic multi-objective optimization technique which generates the Pareto-optimal solutions through various steps such as population initialization and genetic operations including the crossover, mutation, non-dominated sorting and selection on the basis of generated reference points. The core objective of presented paper is to integrate the Latin hypercube sampling (LHS) in NSGA-III as population initialization method for improving the performance of NSGA-III. LHS has the ability to produce a clear depiction of each input distribution, devoid of sampling artifacts. To reveal the applicability of LHS-NSGA-III algorithm, a construction case study is solved for time-cost-safety-quality trade-off optimization. 13 performance metrics were selected and measured to compare the convergence and diversity of Pareto-optimal solutions generated by original NSGA-III and LHS-NSGA-III. Results of the paper indicate that the LHS based NSGA-III provides better Pareto-optimal solutions than original NSGA-III.