Abstract: This paper proposes a Random Crosser Global Harmony Search（RCGHS） algorithm for the problem of premature convergence in harmony search algorithm. In the improvisation stage, the new harmony vector is generated dynamically by means of random crossover for the global optimization problems, i.e., the worst harmony learning from the best harmony and the random selected other harmony learning from the best harmony with different strategies. Mutation strategy is employed to improve the diversity of harmony memory. In the updating stage, the worst harmony vector is updated by the optimal individual of the improvising harmony and the random opposition-based learning harmony vector. Finally, the simulation is carried out using the optimization algorithm of AGHS, ABC, PSO, GWO and other HS variants that are recently proposed. The simulation results demonstrate the RCGHS algorithm has higher convergence precision and convergence rate.

Key words: harmony search algorithm, random crosser, opposition-based learning, local optimum

摘要： 针对和声搜索算法易陷入局部最优的不足，提出了一种随机交叉全局和声搜索（RCGHS）算法。通过最差和声向最优和声学习提高算法的全局搜索性能，引入其他和声向最优和声学习的交互策略提高算法的局部搜索性能。将两种学习策略随机交叉动态产生新和声，平衡算法的全局搜索和局部搜索性能。在和声记忆库更新阶段，利用即兴创作产生的和声向量与随机反向学习产生的和声向量中较优的个体更新和声记忆库。将RCGHS算法与目前文献中较优的几种改进HS算法、ABC算法、PSO算法和GWO算法进行性能测试，测试结果表明RCGHS算法具有较高的寻优精度和较快的收敛速度。

关键词: 和声搜索算法, 随机交叉, 反向学习, 局部最优