【摘要】 为了提高径流预测的准确性，以澄碧河流域坝首站1979-2019年共41 a的实测月径流序列为例，在优选Elman神经网络模型、支持向量机模型、BP单一预测模型的基础上，分别耦合经验模态分解(EMD)、集合经验模态分解(EEMD)和经验小波变换分解(EWT),选取纳什效率系数(NSE)、平均相对误差绝对值(MAPE)和均方根误差(RMSE)对测试集的预测结果进行评价与分析。结果表明：相对于Elman神经网络模型和SVM模型，BP模型的预测效果较好；耦合预测模型预测精度都优于单一模型。耦合模型中，EWT-BP的纳什效率系数为0.91,预报等级为甲级，预测精度优于EMD-BP和EEMD-BP。采用数据预处理技术生成平稳序列，可有效减少原序列存在非线性和不稳定性特征的影响，并有利于提高流域水文模型的径流预测能力。更多还原

【Abstract】 To improve the accuracy of runoff prediction, the paper takes the measured monthly runoff series from 1979-2019 for a total of 41 a at the dam head station in the Chengbi River basin as an example, on the basis of the preferred Elman neural network model, SVM model, BP single prediction model, coupled empirical modal decomposition(EMD), ensemble empirical modal decomposition(EEMD) and empirical wavelet transform decomposition(EWT), respectively, Nash efficiency coefficient(NSE), mean relative error absolute(MAPE) and root mean square error(RMSE) were selected to evaluate and analyze the prediction results of the test set.The results show that the BP model has better prediction results than the Elman neural network model and the SVM model. Among the coupled forecasting models, Nash efficiency coefficient of EWT-BP is 0.91 and the forecast grade is A, the forecasting accuracy is better than that of EMD-BP and EEMD-BP. The use of data pre-processing techniques to generate smooth series can effectively reduce the effects of the existence of nonlinear and instability characteristics of the original series and help improve the runoff prediction capability of the basin hydrological model.更多还原