基于反馈策略的引力搜索算法及其在支持向量机中应用

Gravitational search algorithm based on feedback mechanism and its application in SVM

针对标准引力搜索算法(SGSA)在高维多峰函数寻优过程中容易出现早熟的问题,提出一种基于反馈策略的引力搜索算法(FGSA)。由于粒子在进化过程中群体多样性损失过快,采用粒子与最佳位置的距离和最邻近粒子的距离两个参数来均衡优化算法的勘探和开发能力,并将变异操作引入到FGSA中。通过对选取的四个基准函数测试,验证了FGSA和SGSA相比,在高维多峰函数寻优时,精确度和稳定性都有显著提高。同时,针对支持向量机(SVM)分类问题时,可有效地找出合适的特征子集及SVM参数,并取得较好的分类结果。

To overcome premature problem of high-dimensional muhimodal function in the optimization process by Standard Gravitational Search Algorithm （SGSA）, a new Gravitational Search Algorithm based on Feedback mechanism （FGSA） was proposed. Considering the large loss in population diversity during the evolution process, the distance to the optimum position and the distance to its nearest neighbor were introduced into the proposed algorithm to balance the trade-off between exploration and exploitation ability. Mutation operation was also embedded into FGSA. The test results of four benchmark functions demonstrate that FGSA has much better stability and accuracy than SGSA in finding global optimum. Furthermore, when FGSA is applied to Support Vector Machine （SVM） classification, suitable feature subsets and SVM parameters can be effectively found out, and better classification results will be obtained.