摘要： 针对标准支持向量机对噪声和异常值比较敏感的问题，通过限定噪声和异常值的损失上界，提出一种基于不对称Ramp损失函数的鲁棒支持向量回归机模型，应用凹凸过程将其由非凸优化问题转化为凸优化问题并利用牛顿法进行求解。对上证指数和香港恒生指数收盘价的预测结果表明，该模型能在一定程度上抑制噪声和异常值的影响，从而提高预测精度及减少下跌风险，达到规避风险的目的。

关键词: 支持向量机, 时间序列, 鲁棒性, 不对称损失函数, 牛顿法

Abstract: Aiming at the problem that standard Support Vector Machine(SVM) is sensitive to noise and outliers, by setting the upper bound of loss caused by noise and outliers, this paper presents a robust Support Vector Regression(SVR) based on asymmetric ramp loss function. The concave-convex procedure is employed to transform the associated non-convex optimization problem into a convex one. A Newton method is introduced to solve the robust model. Numerical experiments on the closing price of Hong Kong’s Hang Seng index and Shanghai Stock index show that the model can reduce the noise and the influence of the abnormal values to a certain extent, increase the prediction accuracy and reduce risk of falling to avoid risk.

Key words: Support Vector Machine(SVM), time sequence, robustness, asymmetric loss function, Newton method

中图分类号: 

• TP39