Abstract: Because of the problems of Differential Evolution（DE） algorithm such as premature convergence, low accuracy and tedious parameter setting for hard high-dimensional optimization problems, a dynamic differential evolution algorithm with weighted mutation strategy, called WMDDE, is presented. Firstly, two new weighted mutation operators of random disturbance are designed by weighting combination of DE/rand/1 and DE/best/1, which is utilized to balance the global and local search dynamically, and avoid premature convergence. Secondly, a self-adaptive parameter setting strategy of adjust scaling factor and crossover factor is designed, avoiding tedious parameter setting. Finally, experimental results on 11 benchmark functions show that the new algorithm can effectively avoid premature convergence and has the global convergence ability strongly, and its optimization rate solution accuracy, stability are better than the other four kinds of differential evolutions.

Key words: differential evolution algorithm, dimensional mutation, disturbance, premature convergence, parameter setting

摘要： 针对差分进化算法在解决高维优化问题时易早熟收敛、求解精度低和参数设置麻烦等问题，提出一种加权变异策略动态差分进化算法（WMDDE）。为了动态平衡全局搜索与局部搜索能力，跳出局部最优，将标准差分进化算法的变异策略DE/rand/1和DE/best/1进行加权组合，提出两种新的随机扰动加权变异算子。提出一种动态自适应调整缩放因子和交叉概率因子的策略，避免参数设置的麻烦，提高算法的稳定性。在11个Benchmark函数上的测试结果表明，新算法能有效避免早熟收敛，全局寻优能力强，且在高维时寻优速度、求解精度和稳定性均优于4种DE进化算法。

关键词: 差分进化算法, 维变异, 扰动, 早熟收敛, 参数调整