基于改进的SVMR的混沌时间序列预测

针对标准支持向量机处理大规模数据集会出现训练速度慢、计算量大的缺点,提出了一种基于二叉树模型的支持向量机回归方法。通过二叉树模型将大样本数据集自适应分解成若干个子集,利用支持向量机分段提出支持向量,再把这些支持向量汇合成一个训练样本集进行训练产生决策函数,并将其应用到混沌时间序列的预测。与标准算法相比,该方法在保证泛化精度一致的前提下,极大地加快了训练速度。

Chaotic Time Series Forecasting Based on Improved SVMR

Im accordance with the problem that the large-scale samples training process of standard support vector machine is slow and with large computation,a support vector machine regression(SVMR) based on binary tree model(BTM) is proposed.A large-scale sample is adaptively decomposed into several subsets by binary-tree model,and then the support vectors in different subsets are independently extracted and recombined into a new training sample. So the decision function can be obtained by training the new sample,and applied to chaotic time series forecasting.Compared with the standard support vector machine,the proposed method can greatly speed up training process with almost the same precision.