Using adaptive novelty search in differential evolution

Abstract

Novelty search ensures evaluation of solutions in stochastic population-based nature-inspired algorithms according to additional measure, where each solution is evaluated by a distance to its neighborhood beside the fitness function. Thus, the population diversity is preserved that is a prerequisite for the open-ended evolution in evolutionary robotics. Recently, the Novelty search was applied for solving the global optimization into differential evolution, where all Novelty search parameters remain unchanged during the run. The novelty area width parameter, that determines the diameter specifying the minimum change in each direction needed the solution for treating as the novelty, has a crucial influence on the optimization results. In this study, this parameter was adapted during the evolutionary process. The proposed self-adaptive differential evolution using the adaptive Novelty search were applied for solving the CEC 2014 Benchmark function suite, and the obtained results confirmed the usefulness of the adaptation.