An improved differential evolution by hybridizing with estimation-of-distribution algorithm

Abstract

To fully exploit the strong exploitation of differential evolution (DE) and the strong exploration of the estimation-of-distribution algorithm (EDA), an improved differential evolution by hybridizing the estimation-of-distribution algorithm named IDE-EDA is proposed in the study. Firstly, a novel cooperative evolutionary framework is proposed to hybridize LSHADE-RSP, a state-of-the-art DE variant incorporating DE-based effective improvement strategies, with EDA. Secondly, the dominant individuals generated by LSHADE-RSP are used to establish the probability distribution model for EDA to enhance its exploitation in each generation, and a new control parameter is introduced to balance exploitation and exploration. Then, the use of greed strategy works via EDA to fully retain high-quality solutions to the next generation to improve the convergence speed. Finally, the greedy strategy is used to shrink the external archive when its size decreases due to the reduction of the population size. A comparison of IDE-EDA with cutting-edge DE-based and EDA-based variants, including AAVS-EDA, EB-LSHADE, ELSHADE-SPACMA, jSO, LSHADE-RSP, RWGEDA, HSES, and APGSK-IMODE, was implemented to verify its efficiency. The statistical test results on the IEEE CEC 2018 and IEEE CEC 2021 test suites demonstrate that IDE-EDA is an excellent hybrid algorithm. The MATLAB source code of IDE-EDA can be downloaded from https://github.com/Yintong-Li/IDE-EDA.