支持向量机(support vector machine,SVM)的学习性能和泛化能力在很大程度上取决于参数的合理设置.将支持向量机的参数选择问题转化为优化问题,以模型预测均方根误差为评价函数,提出一种引入混沌变异操作的改进分布估计算法(estimation of distribution algorithm,EDA),并将其用于优化求解ε-支持向量机的参数:惩罚因子、不敏感损失系数以及高斯径向基核函数的宽度.由于改进EDA利用混沌运动的随机性和遍历性等特点在解空间内进行优化搜索,能够较好解决传统EDA易于陷入局部极小的缺陷.Chebyshev混沌时间序列预测仿真结果表明:改进EDA是选取SVM参数的有效方法. The learning performance and generalization ability of support vector machine (SVM) depend largely on the reasonable setting of parameters. The problem of parameter selection of SVM is transformed into optimization problem, and the root mean square error An estimation of distribution algorithm (EDA) which introduces chaotic mutation operation is proposed and used to optimize the parameters of ε-SVM: penalty factor, insensitive loss coefficient and Gaussian radial The width of kernel function.Because the improved EDA makes use of the randomness and ergodicity of chaotic motion and so on to search optimally in the solution space, it can solve the defect that the traditional EDA is easy to fall into local minima. The simulation results of Chebyshev chaotic time series prediction : Improving EDA is an effective way to choose SVM parameters.