基于布谷鸟搜索算法的SVR参数选择  

Parameter Selection of Support Vector Regression Based on Cuckoo Search Algorithm

作　　者：何思露 韩坚华[1] 

机构地区：[1]广东工业大学计算机学院,广州510006

出　　处：《华南师范大学学报（自然科学版）》2014年第6期33-39,共7页Journal of South China Normal University(Natural Science Edition)

基　　金：国家自然科学基金项目(61142012)

摘　　要：SVR(支持向量回归机)在解决非线性回归问题时有极大的优势,在其预测过程中,最重要的是参数的选择,不同的参数会造成预测结果的巨大差异.目前较为普遍的方法是利用遗传算法和粒子群算法进行参数选择,而这2种算法在解决多峰问题时的局限性,容易导致算法的效率低且准确度不高.鉴于布谷鸟搜索算法引入了Lévy飞行机制,能有效地跳出局部最优解,使算法收敛速度快,且结果具有对算法本身的参数变化不敏感的优点,该文将布谷鸟搜索算法应用于SVR参数寻优过程中.网络流量和白葡萄酒质量的预测实验结果表明,布谷鸟搜索算法相对于遗传算法、粒子群算法等其他启发式智能算法而言,收敛速度更快,寻参结果的精度更高.SVR （ support vector regression） has a great advantage in solving nonlinear regression problems.In the process of SVR＇s predictions, the most important step is the choice of parameters.The result will be very different because of the change of parameters.The common method is to use GA （ genetic algorithm ） and PSO （ particle swarm algorithm） for parameter selection, however, the limitations of these two algorithms in solving the problem of multi-modal can easily lead to low efficiency and the accuracy is not high.Cuckoo search algorithm introduces a Lévy flight mechanism that can effectively escape from local optimal solution.The algorithm converges fast, and the result is not sensitive to the parameters of the algorithm itself.The cuckoo search algorithm is applied to the param-eter selection of SVR.The experimental results of network traffic prediction and the wine quality prediction show that the cuckoo search algorithm is faster and better compared with GA, PSO algorithms.

关 键 词：布谷鸟搜索算法 支持向量回归机 参数选择 网络流量预测 

分 类 号：TP301.6[自动化与计算机技术—计算机系统结构]