基于小波包分解的 LS-SVM-ARIMA 组合降水预测

针对降水量影响因素众多, 是一种复杂的非平稳、非线性且存在噪声问题的时间序列的特点, 提出一种基于小波包分解的 LS-SVM 与 ARIMA 组合模型的年降水量预测方法。利用小波包将降水序列分解成低频趋势序列和高频细节序列; 应用 LS-SVM 模型预测低频趋势序列, ARIMA 模型预测高频细节序列; 将两个模型的预测结果叠加, 得到年降水量的预测值。实例验证表明: 小波包对时间序列的分解比小波分解更精细, 组合模型预测能够全面的提取降水序列中所包含的信息, 更好地反映年降水量随时间变化规律, 提高了年降水量预测的精准度, 为降水量预测提供一种新方法。

Precipitation prediction using LS-SVM and ARIMA combined model based on wavelet packet decomposition

An annual precipitation prediction method is proposed based on wavelet packet decomposition of LS-AVM and ARI- MA combined model because the precipitation has a complex non-stationary, nonlinear, and noisy time series. The wavelet packet is used to decompose the precipitation sequence into a low-frequency trend sequence and high-frequency detail sequence. The LS-SVM model is used to predict the low-frequency trend sequence, and the ARIMA model is used to predict the high-frequency detail sequence. The prediction results of the two models are superimposed to get the predicted value of annual precipitation. The case study shows that: the decomposition of time series by wavelet packet is more precise than the wavelet decomposition, the combined model prediction can comprehensively extract the information contained in the precipitation sequence, better reflect the change of precipitation with time, and improve the annual precipitation forecast which provides a new met hod for the prediction of precipitation.