Differential Evolution Using Historical Knowledge

Differential evolution (DE) is a simple but efficient algorithm for the global optimization over continuous spaces. However, the problem of premature convergence still exists. When trapped in evolution stagnation, DE usually requires much time to jump over. In this paper, the algorithm of DE/rand/1/bin is improved by making use of the historical knowledge. An auxiliary population (AP) is used as a warehouse for storing the information of candidate solutions. This scheme enables AP as a resource, which can maintain the population diversity without computation consumed. A new operator with the extended search direction (ESD) is presented to prevent the premature convergence by use of the historical knowledge of candidate solutions. The proposed strategy attempts to balance the exploration and exploitation abilities Of DE. The comparison shows that the improved DE algorithm performs better than DE/rand/1/bin and PSO.