Differential evolution with dual-space-based population size adaptation and dynamic classification-based mutation strategy

The performance of the differential evolution (DE) algorithm is highly dependent on the setting of parameters and the design of mutation strategy. This paper proposes a DE algorithm based on a dual-space-based population size adaptive (DSPSA) method and a dynamic classification strategy (DSDCDE) to adaptively adjust population diversity during the evolutionary process and balance the algorithm’s exploration and exploitation capabilities. DSPSA is the first method to use the fitness space information and search space information to adapt to the population size. Unlike other DE variant population size adjustment strategies, DSPSA adaptively adjusts population size using dual-space information. This helps to delete individuals who lack prospects while retaining those with more significant potential, thus enhancing the individual’s ability to escape local optima. We propose an enhanced mutation operator based on dynamic classification that uses elite individuals to guide population evolution and introduces information about inferior individuals to increase population diversity. In addition, dynamic classification-based improved update methods are employed for scale factor and crossover rate, introducing individual information to provide tailored parameters for individuals at different optimization stages. This ensures a proper balance between global exploration and local exploitation capabilities. The Wilxoncon signed-rank test results compared with LSHADE-cnEpSin, jSO, IMODE, AL-SHADE, LADE, IDE-EDA, APSM-jSO in the CEC 2017 benchmark suite were 96/68/39 at 30D, 111/42/50 at 50D and 106/43/54 at 100D. The numerical results indicate that DSDCDE exhibits competitiveness regarding convergence speed and final accuracy.