An efficient differential evolution algorithm based on orthogonal learning and elites local search mechanisms for numerical optimization

Differential evolution (DE) is an efficient stochastic algorithm for solving global numerical optimization problems. To effectively relieve the stagnation and premature convergence problems in DE, this paper presents an efficient DE variant, abbreviated as OLELS-DE, by designing orthogonal learning and elites local search mechanisms. More specifically, the stagnation or premature convergence phenomenon will be detected by monitoring the best individual's update condition during the evolution, then a population diversity estimation technique is utilized to distinguish between these two conditions empirically. To recover the population's evolution vitality according to the classification results, the enhanced orthogonal learning scheme is employed by selecting two different groups of individuals for constructing the orthogonal experimental design procedure. Moreover, the elites local search method is developed by selecting several well-performing elite individuals based on the Gaussian distribution model to further assist the former orthogonal learning mechanism. This scheme is designed to enhance the exploitation ability by searching the regions around elite individuals. The parameters and strategies in above two mechanisms are designed on the expectation of balancing the local exploitation and global exploration capabilities. The optimization performance of proposed OLELS-DE algorithm is evaluated based on 30 benchmark functions from CEC2014 test suite and is compared with eight state-of-the-art DE variants. As it was anticipated, the incorporation of orthogonal learning and elites local search mechanisms helps OLELS-DE have significantly better or at least comparable performance to the adopted DE competitors.