中文摘要：

针对目前金融时间序列预测方法的不足，在利用训练样本与测试样本间马氏距离对惩罚因子进行加权的基础上，改进传统的支持向量回归机（SVR）。通过以上海证券综合指数趋势的预测为例子，与标准BP人工神经网络（BPANN）和SVR方法进行了对比，发现该方法能获得更准确的预测结果。结果表明，该方法能充分反映股票价格时间序列趋势规律，是研究金融时间序列预测问题的有效方法。

英文摘要：

According to the disadvantages of t financial time serial forecast, the improved support vector regression （SVR） is developed by using Mahalanobis distance between training and testing samples to get weighted penalty coefficients. Taking Shanghai Composite Index trend forecast as an example, more accurate results can be acquired by this method compared with normal BP artificial neural network. The results indicate that the method could sufficiently reflect the trend of stock price time series, and it is an effective approach for financial time series forecast.