Mutual learning differential particle swarm optimization

This study proposes a mutual learning strategy to develop a high performance hybrid algorithm based on particle swarm optimization and differential evolution. In the mutual learning strategy, the position information in PSO subswarm is employed for DE mutation, and the DE individuals are used to construct learning exemplar for PSO subswarm together with particles' historical best position. A novel elite DE mutation is proposed to speed up the convergence rate of DE subswarm. Based on mutual learning technique, the mutual learning differential evolution particle swarm optimization (MLDE-PSO) is proposed. To evaluate the performance of MLDE-PSO, three groups of test functions are employed, namely thirteen basic functions, thirteen rotated basic functions and thirty CEC2017 functions. The test results are compared with three state-of-the-art PSO algorithms, three recently PSO algorithms and DE/rand/1. The test results indicate that the proposed MLDE-PSO performs better than the other seven comparison algorithms, especially on rotated functions and CEC2017 functions. The rotation test shows that MLDE-PSO is not very sensitive to rotation transformation.