Adaptive Differential Evolution with Landscape Modality Detection for Global Optimization

摘要

In most differential evolution (DE) algorithms, little work for the design of the mutation operator is directly relevant to the information of fitness landscape of the problem being solved. As the previous studies show, different mutation strategies are suitable for different problems with different fitness landscapes, and the performance of the mutation strategy is tightly linked to the fitness landscape. Therefore, to enhance the performance of DE with the fitness landscape of the problem being solved, an adaptive DE algorithm with landscape modality detection (LMD) technique, named DE-LMD, is proposed in this study. With LMD, DE-LMD can automatically detect the modality of the optimized problem. After that, a mixed strategy with two DE mutation operators, DE/current-to-best/1 and neighborhood-guided mutation, is adopted for the problems with different landscape modalities. The experimental results have demonstrated the high performance of DE-LMD on different kinds of optimization problems.