一种ε可调的在线支持向量回归及其训练算法

Epsilon-adjustable on-line support vector regression and its training algorithm

标准支持向量回归问题中,噪声较大的时段将包含较多的支持向量。提出一种时间窗内!可调的支持向量回归方法,根据各时间窗的支持向量的比例动态调整!,能够处理噪声时变的回归问题。并给出一种!调整时的在线训练算法,避免重复求解凸规划问题。实例表明该方法的泛化能力和拟合精度较标准支持向量回归为优。

The distribution of support vector is analyzed.An epsilon-adjustable support vector regression is proposed.According to the ratio of the number of support vectors to the number of samples in each time window,epsilon is adjusted to handle timevarying noise of system.An on-line training algorithm is presented to avoid repetitious solving the convex programming problem. Simulation shows that the generalization performance and the accuracy of the proposed method are better than the standard support vector regression.