混合波动差分进化算法  

Differential evolution algorithm with hybrid wave

作　　者：刘荣辉[1,2] 李蓓[1] 

机构地区：[1]河南城建学院计算机科学与工程系,河南平顶山467036 [2]东华大学管理学院,上海200051

出　　处：《计算机应用研究》2011年第11期4053-4056,4063,共5页Application Research of Computers

基　　金：国家自然科学基金资助项目(70971020)

摘　　要：针对传统差分进化算法在处理复杂优化问题时存在易陷入局部最优解和求解精度低的缺陷,提出了混合波动差分进化算法。该算法预先设置两组配置参数,并依据混合系数的区分,依次选择每组中的交叉因子和相关参数构成波动算子生成变异率用于新算法中;同时为了加快收敛速度,采用选中随机向量中适应度值最优的向量作为基向量,避免算法寻优的盲目性。对一系列经典Benchmark函数的测试,并将实验结果与其他算法进行比较,证明了本算法的收敛速度与优化质量均显著改善。Traditional differential evolution algorithm in dealing with complicated optimization problems has shortcomings,such as being trapped into local optimum easily and low solution precision.This paper proposed a new hybrid differential evolution algorithm with dynamic wave.Two predetermined parameters sets,which were distinguished by hybrid factors to generate fluctuation mutation and crossover rate,were selected each in turn.At the same time,to enhance the convergent rate,randomly selected vectors with the optimal fitness values we introduced to guide searching direction.Used Benchmark problems to verify this algorithm and the result of simulation,which was compared to other well-known algorithms.It indicates that this algorithm is better than several other algorithms both in convergence rate and quality of optimization.

关 键 词：差分进化 全局最优 混合波动 参数控制 函数优化 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程] TP301.6[自动化与计算机技术—控制科学与工程]