针对现有和声搜索算法收敛速度慢、参数选取困难的不足,提出了一种求解数值优化问题的轮盘赌自适应和声搜索算法.该算法在和声库学习环节用轮盘赌选择取代HS算法的完全随机选择和GHS算法的贪婪选择,在提高收敛速度的同时克服了GHS算法由于贪婪选择造成的早熟;在参数选取中利用群体适应度方差生成概率PAR以自适应微调,然后根据和声库的信息、变量的取值范围和迭代次数进行自适应调整微调步长.仿真时设计了一个特殊函数用于例证轮盘赌选择机制的有效性和GHS算法的早熟问题,通过四个经典函数证明了该算法在收敛速度和收敛精度方面优于HS和IHS算法.
This paper proposed a roulette selection self adaptive harmony search algorithm for solving numerical optimization problems.It replaced the completely random choice of harmony search (HS) algorithm and the greedy choice of global best harmony search (GHS) algorithm with roulette selection in order to speed up convergence of HS and also to avoid premature of GHS.As for parameters setting,it involved population fitness variance to generate the parameter PAR,and then designed an automatic step adjustment according to harmony information,scope of variable and current iteration number.It designed a special function to prove effectiveness of roulette selection mechanism and the premature of GHS,further simulations of four functions show that the proposed algorithm is superior in convergence speed and precision to the HS and IHS algorithm.