An evolutionary-state-based selection strategy for enhancing differential evolution algorithm

The survivor selection operation is one of the factors affecting the performance of differential evolution (DE) algorithms. However, in single-objective DE algorithms, researchers seldom address the improvement of survivor selection operations. Most DE algorithms use greedy selection operations. In this research, we propose an evolutionary-state-based selection (ESS) strategy for single-objective DE, which constructs a probability model based on evolutionary states to solve real-parameter optimization problems. In the survivor selection process, it is no longer true that only the winner can survive to the next generation; the loser also has a certain probability of surviving to the next generation, and this probability is related to the evolutionary states. Based on ESS, a new DE variant called evolutionary-state-based differential evolution (ESDE) is proposed. To verify the ESDE and ESS performance, ESDE was compared with eight state-of-the-art DE variants, and ablation experiments were performed on ESDE to analyze the impact of ESS on the ESDE performance. The experimental results show that ESDE is significantly better than the other eight DE variants, and ESS could significantly improve the DE performance.