Adaptive Differential Evolution Algorithm with Multiple Gaussian Learning Models

Abstract

AbstractThe search efficiency of the differential evolution (DE) algorithm significantly depends on its mutation strategies and control parameters. The multiple Gaussian learning model-based parameter adaptation mechanism is useful in choosing suitable parameters. Naturally, it is worth studying whether this mechanism is also beneficial for the automatic selection of the appropriate mutation strategies. To this end, this paper proposes an adaptive Differential Evolution algorithm with Multiple Gaussian Learning Models, known as MGLMDE, which includes two adaptation mechanisms, i.e., the multiple Gaussian learning model-based parameter adaptation mechanism (MGLMP) and multiple Gaussian learning model-based strategy adaptation mechanism (MGLMS). MGLMP and MGLMS determine the future mutation strategies and control parameters, respectively, using multiple Gaussian models to learn from successful historical memories. The linear population size reduction (LPSR) mechanism is used to control the population size. The proposed algorithm is evaluated via comparisons with some powerful DE methods on lots of test problems. The experimental results demonstrate that the proposed method is better than or at least highly competitive with the state-of-the-art DE algorithms.KeywordsDifferential evolutionMultiple Gaussian learning modelsNumerical optimizationParameter adaptationStrategy adaptation