降雨时间序列降噪及预测系统建模研究

Noise reduction and prediction system of rainfall chaotic time series

以解决降雨混沌时序预测精度较低的问题为目的,基于相空间重构思想,应用混沌降雨时序奇异值分解技术对混沌时序的噪声进行了剥离。采用定性和定量的混沌特性判定方法,计算指出降雨时序具有明显的混沌特性。在此基础上构建了基于最小二乘支持向量机的降噪、预测一体化模型并进行了多步预测实验。结果表明降噪前后预测精度相差很大,证实了噪声是造成混沌预测方法预测精度较低的主要原因。通过与其他预测方法比较,验证了所建立的混沌预测模型能够捕捉到原混沌系统的动力学特征,预测误差较小,泛化能力较强,其预测效果较好。

Not only the noise reduction methods of rainfall chaotic time series with noise and its reconstruction techniques were studied, but also prediction techniques of rainfall chaotic time series were discussed based on chaotic data noise reduction. First , the measurement of rainfall time series were respectively applied for noise reduction using singular value decomposition; Then , nonlinear dynamics character of this time series is analyzed, the results of calculating show that there＇ s chaotic time series. Based on this, a LLSVM model was built by phase space reconstruction. The influent of rainfall has been disposed can be predicted by this model. The results indicate that reasonable forecasting results have been achieved through such method.