Differential evolution with objective and dimension knowledge utilization

Abstract

Performance of differential evolution, which is one of the most competitive evolutionary algorithms, heavily depends on the utilization of feedback information. The feedback information can be from objective, solution and dimension spaces. To facilitate the better utilization of feedbacks for performance enhancements, this paper proposes an objective-dimension feedback (ODF) method with two novel mechanisms to, respectively take advantages of dimension and objective space knowledge. The first mechanism, named “Small diversity dimensions exploit, Large diversity dimensions explore” classifies dimensions into exploitation and exploration dimensions according to their diversity rankings and assigns them with collective and single dimensional learning strategies, respectively. The second mechanism, named “number of dimensions automatic configuring” automatically configures the number of dimensions performing exploitation and exploration in each solution according to its fitness ranking. Experiments on 29 benchmark functions confirm the effectiveness of ODF by performance comparisons with single utilization of objective and dimension space knowledge, single utilization of dimensional learning strategies and several objective, solution and dimension space knowledge-based methods from literatures.