Principal coordinate strategy
Yifeng Gao, Dan Lv · 2012
Differential evolution (DE) algorithm has a wide use in optimization problems, whose performance is closely related to the separability of the fitness function. In this paper, we propose Principle Coordinate (PC) strategy, a new adaptive control strategy to improve DE's performance. PC attempts to maximize the fitness function's separability and make crossover operator more robust through coordinate rotation. In PC, Principal Component Analysis (PCA) is adopted to draw the ideal coordinate system from the difference vectors distribution. In the numerical experiments, PC is combined with two versions of classical DE algorithms to test its ability. The first experiment measures the accuracy of the coordinate system obtained by PC. In the second experiment, four benchmark functions and an engineering project are used to evaluate PC's efficiency. The results show that PC improves DE's efficiency, robustness and stability.