Adaptive Mutation based on Population Distribution in DE with an individual-dependent mechanism

Differential evolution (DE) is a method of evolutionary computation(EC) with a superior search capability for real-value optimization in nonlinear problems, nondifferential problems and multimodal problems, among others. While various DE methods have been proposed, DE with an individual-dependent mechanism (IDE) features a pair of control strategies and its superior search performance has been confirmed in various benchmark problems. This study aims to further improve the individual-dependent mutation (IDM) strategy in IDE. The IDM strategy divides individuals into superior and inferior populations over the search process and applies mutation in each population. Further, the ps criterion is used as a reference in dividing the population and applying mutation. This criterion changes by the number of generations, but it is not based on the individuals 'information. From our preliminary experiments, premature convergence occurs in some problems. This study focuses on the ps criteria and proposes a method of adaptively controlling the value based on the variance of individuals. Performance evaluation was performed using a 30, 50 and 100-dimensional benchmark problem generally used in EC. In the performance evaluation, we showed that the proposed method significantly outperformed conventional DE and IDE, and improved the search performance by preventing convergence to the local minima in problems prone to such an issue by controlling the variance of the individuals.