Population initialization based on search space partition and Canopy K-means clustering

In order to improve the ability of exploration and exploitation and improve the convergence and evolutionary efficiency for differential evolution algorithms, a population initialization method based on uniform partition of search space, local search and clustering is proposed. Firstly, the decision variable space is partitioned uniformly, and an individual is randomly selected from each subspace, and the selected individuals can cover the whole search space.Then Hooke-Jeeves algorithm is used to search each subspace locally, and the local optimal individuals are got.Combined with the improved Canopy algorithm and K-means clustering algorithm, the promising region in the search space is identified.Based on this, the local optimal individuals generated by local search are screened, and the individuals for the initial population are finally generated.Compared with other population initialization methods for five CEC2017 test functions, the running time of the proposed method can be reduced to 0.75 times, and the fitness function can be reduced to 0.03 times that of the existing methods.And the proposed method has the minimum standard deviation and the optimal convergence characteristics. © 2020, Editorial Office of Control and Decision. All right reserved.