Vector‐angle penalised NSGA‐III to solve many‐objective optimisation problems

Abstract

One of the major challenges in evolutionary many-objective optimisation is to maintain convergence and diversity among Pareto-optimal solutions. Taking both into consideration, this Letter presents a -NSGA-III algorithm which incorporates minimum-vector-angle principle in association operation of original non-dominated sorting genetic algorithm III (NSGA-III) scheme to solve unconstrained many-objective optimisation problems. Each non-dominated population member close to a reference point is emphasised in optimal solution set using minimum vector-angle penalty parameter with perpendicular distance in association operation. Performance evaluation of -NSGA-III algorithm is done over unconstrained DTLZ test suite by computing delta () and inverted generational distance as quality metrics. The improved performance of the suggested algorithm over NSGA-III, MOEA/D and VaEA could be considered as an alternative tool to handle optimisation problems with more than three conflicting objectives.