改进的混沌粒子群算法对WLS-SVM性能参数的优化

Improved Chaos Particle Swarm Optimization Algorithm for Optimizing the Parameters of WLS-SVM

加权最小二乘支持向量机(WLS-SVM)的学习性能和泛化能力取决于其正则化因子C和核函数参数σ的取值。对此,针对WLS-SVM建立C和σ的组合优化目标函数,采用基于Lozi映射的粒子群(PSO)算法来搜索最优目标函数值。迭代过程中,通过分别映射PSO个体最优位置,把产生的混沌序列中的最优解分别逆运算取代当前个体最优位置,引入混沌机制,以混沌变量的遍历性改进粒子群算法,提高全局搜索能力,避免过早陷入局部最优。将其应用于某玩具企业原料月消耗量预测,结果表明了文中所提方法的有效性。

The learning and generalization ability of the weighted least squares support vector machine （WLS-SVM） depends on the values of its regularization factor C and kernel function parameter σ.Thus,the optimization objective function based on C and σ is established,then an improved particle swarm optimization （PSO） algorithm based on the Lozi mapping is utilized to search the optimal objective function value.The chaos mechanism is introduced by mapping the PSO individual optimal value separately during iterations,then the current PSO individual optimal position is replaced with the optimal values through the inverse operation of the chaos sequence,in this way,the global search ability is improved and the premature local optimum is avoided using the ergodicity of the chaotic variables in the inproved particle swarm optimization.An experiment is applied to the prediction of a toy company＇s raw material monthly consumption,the results show the effectiveness of the proposed method.