基于多种混合模型的径流预测研究

变化环境下径流的波动不断加大,给径流的精准预报带来新的挑战。基于"分解-合成"策略的混合径流预报模型来提高预报精度是当前研究的热点之一。以往研究聚焦在单一的混合预报模型而忽视了它们的适用性研究。基于此,以渭河流域为例,在优选多元线性回归(MLR)、人工神经网络(ANN)和支持向量机(SVM)单一预报模型的基础上,分别基于经验模态分解(EMD)、集合经验模态分解(EEMD)和小波分解(WD)构建了多种混合模型,并融合了大气环流异常因子的信息。结果表明:(1)SVM模型预测精度高于ANN和MLR;(2)混合预测模型预测精度均高于单一模型,混合模型中WD-SVM的预测精度优于EMD-SVM和EEMD-SVM;(3)融合大气环流异常因子后WD-SVM模型预测精度最高,对极值预报精度的提高较为明显。

Runoff prediction based on multiple hybrid models

The high variability of runoff has brought a new challenge to accurate runoff forecasting in a changing environment. Hybrid runoff forecasting models based on the "decomposition-synthesis" strategy to improve the accuracy of forecasting has become one of the research hotspots. Most previous studies focused on the application of single hybrid model in runoff forecasting however ignoring the applicability of different hybrid models. To this end, the Weihe River Basin is taken as case study, multiple linear regression (MLR), artificial neural network (ANN) and support vector machine (SVM) single prediction models were optimally selected. Then, the results of the hybrid prediction model using wavelet decomposition (WD) were compared with empirical mode decomposition (EMD) and ensemble empirical mode decomposition (EEMD) techniques. The atmospheric circulation anomaly factors was incorporated into the hybrid models for further improve the accuracy of runoff forecasting. The achieved results demonstrate that:(1) the prediction accuracy of SVM model is higher than that of ANN and MLR;(2) the prediction accuracy of hybrid models is higher than that of single models;(3) the prediction accuracy of WD-SVM is better than EMD-SVM and EEMD-SVM, which has been further improved by integrated with the information of atmospheric circulation anomaly factors. Especially, the improvement of prediction accuracy is more obvious at the extreme point.