KPCA-LSSVM方法在视频时间序列预测中应用

为提高时间序列预测精度及降低预测过程中的计算复杂度,提出一种基于核主成分分析(KPCA)与最小二乘支持向量机(LSSVM)相结合的预测方法.首先,将输入数据通过核方法映射至高维特征空间;然后,在特征空间上提取有效非线性主元;最终,通过LSSVM建立时间序列模型.为验证KPCA-LSSVM方法的有效性,将其应用于交通流及视频流预测中,在同等条件下,与单一的LSSVM及神经网络等预测方法进行比较.实验结果表明:基于KPCA-LSSVM建立的模型具有较好的推广性及较高的辨识精度.

Application of KPCA-LSSVM in Video Trace and Time Series Prediction

A prediction method based on kernel principal component analysis(KPCA)and least squares support vector machine(LSSVM)is proposed for the prediction of time series that increasing prediction precision and decreasing the computing complexity. Firstly, the input data will be mapped to high-dimensional feature space through kernel method, then the effective nonlinear principal element can be extracted in the feature space, and finally the time series model is established by LSSVM. In order to verify the validity of KPCA-LSSVM method, it is used in traffic flow and video flow prediction, and compared with single LSSVM and neural network in the same condition. The experimental results show that the model based on KPCA-LSSVM has good generalization and high identification accuracy compared with other methods.