基于加权系数寻优的回归型加权支持向量机

在加权回归型支持向量机中，由于考虑到不同数据对预测函数贡献程度的差异性，其预测效果往往优于标准的回归型支持向量机，该文针对现有回归型加权支持向量机使用中直接选择加权系数法存在的不足，提出了一种对加权系数进行优化的新方法。该方法通过选取曲率变化大、形式简单的幂函数作为候选加权函数，并采用格子搜索法寻找最优参数，从而可以确定出最优加权系数。仿真实验表明：在利用加权支持向量机训练时间序列数据集时，采用该方法确定最优加权系数，比目前常用选择加权系数的方法效果好。

Weighted Support Vector Regression Based on Weighting Factor Optimization

The weighted support vector regression outperforms the standard support vector regression when applied in time sequence data set , for the merits that different input points can make different contributions to the learning of predictive function. In this paper, the selection problem of weighting factors in weighted support vector regression is considered. Based on the optimization technique, a new selecting approach is proposed to overcome the shortcoming of conventional methods. In the present method, a proper weighting factor is chosen as candidate weighting function being suited for suit certain time sequence data set. And then, a grid search approach is adopted to adaptively determine the suitable weighting factors. The experiment results show that the present method has a better performance than usual weighting factor selection method.