基于CEEMDAN-WNN耦合模型的黄河入海输沙量预测研究

Prediction of sediment transport from the Yellow River to the Bohai Sea based on the CEEMDAN-WNN coupled model

河流水沙过程影响因素复杂,是非线性、非平稳的时间序列,具有周期性、突变性、趋势性等特征,传统的水沙预测模型难以准确识别水沙过程的上述特征,存在预测误差大、稳定性差等问题,为了实现对河流水沙的准确预测,发展了一种基于自适应噪声完备经验模态分解耦合小波神经网络的输沙量预测模型,首先利用自适应噪声完备经验模态分解的方法将输入因子径流量进行分解,然后对各个分量IMF以及残差项Res分别进行小波神经网络建模和预测,最后将各分量预测结果重构作为最终预测结果,结果表明,CEEMDAN-WNN耦合预测模型在黄河利津站输沙量预测过程中均方根误差(RMSE)和平均绝对误差(MAE)分别为0.075、0.065,相比于BP神经网络和小波神经网络,精度更高,能够实现对水沙序列的准确预测。

The influencing factors of river water and sediment process are complex,nonlinear and non-stationary time series with characteristics of periodicity,abruptness and trend.The traditional water and sediment prediction model is difficult to accurately identify the above characteristics.The prediction error is large and the stability is poor.In order to improve the accuracy,this paper develops a sediment transport prediction model based on adaptive noise complete empirical mode decomposition coupled with wavelet neural network.Firstly the adaptive noise complete experience is used to decompose the input factor runoff,and to perform wavelet neural network modeling.The prediction for each component IMF and the residual term Res are worked out respectively.Finally,the component prediction results are reconstructed as the final prediction result.The result shows that the CEEMDAN’s root mean square error(RMSE)and mean absolute error(MAE)of the-WNN combined forecasting model in the sediment transport of the Yellow River are 0.075 and 0.065 at Lijin Station,respectively.Compared with the BP neural network and the wavelet neural network,its accuracy is higher.Accurate prediction of water and sand sequences can be obtained.