一种支持向量机参数优化的GA-Powell算法

A GA-Powell Algorithm for Parameter Optimization of Support Vector Machine

支持向量机的核心是核函数,选择合适的核函数参数是支持向量机理论研究的重点。文中将遗传算法与Powell算法相结合,提出了GA-Powell算法来优化核函数的参数。首先利用遗传算法找到一个初始最优解,再利用Powell算法在所得解附近进行寻优,反复迭代产生最优解。该算法在保留遗传算法较强的全局搜索能力的同时具有Powell算法的较强的局部搜索能力,使得混合算法具有更加精确和快速的收敛性。将该算法应用到银行基金项目的分类实验中取得了良好的结果。

Support vector machine （SVM） is the core of the kernel function, selecting the proper parameters of kernel function are the fo- cus of support vector machine theory research. Combined GA and Powell algorithm, the GA-Powell algorithm was proposed to search op- timal parameters. First, use the genetic algorithm to find a initial solution, and then use the Powell algorithm, eventually produce the opti- mal solution. This method retained the global search capability of GA algorithm and the good local convergence of Powell, with more ac- curate and faster convergence. The algorithm is applied to the practice of bank found project classification and has good result.