A Two-Stage Ensemble of Differential Evolution Variants for Numerical Optimization

Differential evolution (DE) is a popular paradigm of evolutionary algorithms, which has been widely applied to solve diverse optimization problems and has gained much success in a series of academic benchmark competitions. Recently, ensemble methods received an increasing attention in designing high-quality DE algorithm. Motivated by this consideration, a novel two-stage ensemble of DE variants, called TSEDE, has been proposed in this paper. In TSEDE, based on the number of fitness evaluations, the whole evolutionary process is divided into two stages. In the former stage, TSEDE using a multi-population-based framework focuses on improving the searchability, which employs three popular and efficient DE variants, namely SHADE, JADE, and “DE/current-to-rand/1.” In the latter stage, LSHADE is used to emphasize the convergence. Moreover, an elite strategy is used to ensure that the current best solution is assigned to each constituent variant at each generation. TSEDE is tested on the CEC2005 benchmark suit and compared with nine typical algorithms. The results confirm that the proposed method is very competitive.