On the Disruption-level of Polynomial Mutation for Evolutionary Multi-objective Optimisation Algorithms

Abstract

This paper looks at two variants of polynomial mutation used in various evolutionary optimisation algorithms for mutliobjective problems. The first is a non-highly disruptive and the second is a highly disruptive mutation. Both are used for problems with box constraints. A new hybrid polynomial mutation that combines the benefits of both is proposed and implemented. The experiments with three evolutionary multi-objective algorithms on well-known multi-objective optimisation problems show the difference in terms of generational distance, hypervolume, convergence speed and hit rate metrics. The hybrid polynomial mutation in general retains the advantages of both versions in the same algorithm.