Genetic Algorithm with Efficient Selection Using Pareto Front Modeling and Data Envelopment Analysis
Mamoru Doi, Kenya Sugihara, Masao Arakawa · 2024
There exist various multi-objective optimization problems(MOPs) in the real world that require acquiring diverse solutions efficiently. Various multi-objective evolutionary algorithms (MOEAs) have been proposed to solve MOPs. Generally, in MOEAs, the same score criterion is conducted in Survival and Tournament Selection. In Survival, parent selection score should be conducted to ensure convergence and diversity, and in Tournament Selection, the selection score of individuals for crossover should be conducted to select individuals with high convergence compared to other individuals. Therefore, evolutionary computation algorithms are expected to perform efficient exploration while maintaining diversity. In this work, we propose an efficient selection algorithm using data envelopment analysis and pareto front modeling. In Survival, we perform an evaluation using existing methods to ensure high convergence and diversity, and in Tournament Selection, we use the data envelopment analysis method to select individuals with high convergence compared to other individuals. This allows us to balance both high convergence and diversity. Moreover, by implementing Pareto front modeling, we were also successful in addressing the issue of assuming convexity in the data envelopment analysis method. In comparison using the HV METRIC on the WFG and DTLZ benchmark functions, the proposed method showed superior results compared to the existing methods of NSGA-II, AGE-MOEA-II, and DEA-GA with the exception of 9-objective functions.