Differential Evolution Algorithm with Fine Evaluation Strategy for Multi-dimensional Function Optimization Problems

Abstract

For multi-dimensional function optimization problems, classical differential evolution (DE) algorithm may deteriorate its intensification ability because different dimensions may interfere with each other. To deal with this intrinsic shortage, this paper presents a DE algorithm framework with fine evaluation strategy. In the process of search, solution is updated and evaluated dimension by dimension. In each dimension, the updated value will be accepted only if it can improve the solution. In case that there is no improvement found in any dimension, the new solution, which is calculated using classical mutation operator only, will be accepted in low probability. This strategy can improve diversification and keep DE algorithm from premature convergence. Simulation experiments were carried on typical benchmark functions, and the results show that fine evaluation strategy can improve the performance of DE algorithm remarkably.