基于社会力模型算法的SVM参数优化  

SVM PARAMETERS OPTIMISATION BASED ON SOCIAL FORCE MODEL ALGORITHM

作　　者：张俊杰[1] 阎高伟[1] 李浩志[1] 

机构地区：[1]太原理工大学信息工程学院,山西太原030024

出　　处：《计算机应用与软件》2015年第9期235-239,共5页Computer Applications and Software

基　　金：国家自然科学基金项目(60975032);山西省自然科学基金项目(2011011012-2)

摘　　要：为了解决支持向量机(SVM)参数优化的问题,提出一种改进的基于社会力模型群智能优化算法SFSO(Swarm Optimization algorithm based on Social Force Model)的SVM参数优化方法。SFSO通过期望力和排斥力使算法在全局搜索和局部搜索中能够较好的平衡,利用SFSO特有的搜索机制对SVM的惩罚因子和径向基函数进行优化,提高SVM的分类性能。通过对几个benchmark函数和常用的UCI数据集进行测试表明:改进后的SFSO算法不仅对于求解函数优化问题具有较强的鲁棒性和较高的求解精度,而且经改进SFSO算法优化后的SVM具有更快的收敛速度和更高的分类准确率。In order to solve the problem of SVM parameters optimisation, we proposed an improved SVM parameters optimisation method, which is based on swarm intelligence optimisation algorithm of social force model （SFSO）. SFSO enables the algorithm to be able in well balance in both global and local searches through expectation force and exclusion force. We use special search mechanism of SFSO to optimise the penalty factor of SVM and the radial basis function, thus improve the classification performance of SVM. It is demonstrated through tests on several benchmark functions and common UCI datasets that the improved SFSO algorithm has stronger robustness and higher precision in solving function optimisation problems. Moreover, the SVM optimised by the improved SFSO has faster convergence speed and higher classification accuracy.

关 键 词：支持向量机 参数优化 SFSO算法 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程]