基于小波消噪技术的投影寻踪自回归预报模型

Projection Pursuit Auto Regression Forecasting Model Based on Wavelet De-noising Technology

针对径流序列呈现出的随机性和非线性等,基于耦合小波分析理论的降噪功能以及投影寻踪自回归模型的非线性逼近功能,建立一种组合预报模型,即PPAR-WDT。模型运用Mallat算法将径流序列进行分解;然后利用阈值消噪技术对含有噪声的高频信号序列进行降噪处理,最后重构新序列;运用投影寻踪自回归模型进行预报。将新组合模型应用于某水文站中长期径流预报的结果表明,相比于单一投影寻踪自回归等3种模型,PPAR-WDT模型具有更高的预报精度和更好的稳定性。

Considering the randomness and nonlinearity of runoff series, the Projection Pursuit Auto Regression model based on Wavelet De-noising Technology （PPAR-WDT） is proposed which combines the de-noising function of wavelet analysis theory with the nonlinear approximation function of the Projection Pursuit Auto Regression （ PPAR）. In this model, the runoff is firstly decomposed into several high-frequency signals and a low-frequency signal by Mallat algorithm, and then, each high-frequency signal is processed and reordered by wavelet de-noising technology to eliminate noise effects, which enables the processed signals to be reconstructed in a new runoff series. The application of this new combined model in the runoff forecast of a hydrologic station shows that PPAR-WDT model has higher forecasting accuracy and better stability comparing with other three different kinds of forecasting models.