Self-adaptive differential evolution algorithm based on population state information

The local optimum and stagnation state information of the population seriously affects the performance of differential evolution (DE) algorithm.An advanced DE algorithm with population state processing measures was proposed to address the above two issues.When the population falled into the local optimum, the individuals in the population were learned randomly by LBFGS method to improve the global quality of the solution, and Gaussian mutation was employed to trigger new individuals to jump out of local optimum.As for the stagnation state, the covariance matrix of the population was applied to reorganize the target individuals based on the rotation of the spatial coordinates to suppress the stagnation state of the population and enhance the global search ability of the algorithm.In addition, a new selection strategy was designed, which built an external archive to store abandoned individuals after greedy selection.When the trial individual was inferior to the target individual, the algorithm no longer generated the next generation with greedy selection strategy, but made reasonable intelligent selection around the external archive to ensure that the algorithm converges to the global optimum.Compared with eight state-of-the-art DE algorithms on 29 benchmark functions, the experimental results show that the proposed algorithm has better performance in terms of the solution accuracy and convergence speed.