基于EMD-ATT-BiLSTM的径流量预测模型研究

Research on Runoff Prediction Model Based on EMD-ATT-BILSTM

准确可靠的径流预报在水资源的优化管理中发挥着越来越重要的作用。为了提高预测精度,提出了一种神经网络模型,来进行日径流预报。此模型将经验模态分解(EMD)方法、注意力机制、BiLSTM神经网络相结合,并且对输入数据采用了插值方法来提升精确度。EMD方法能够将非稳态非线性的径流时间序列分解成多组本征模态分量和趋势项,实现输入时间序列的稳态化,再经过注意力机制赋予时间序列不同关注度,然后通过BiLSTM分别预测再重构。将该模型应用于四川省宣汉县的清溪河站点的每日径流数据上,与另外三种神经网络模型即LSTM、ATT-LSTM和ATT-BiLSTM模型进行对比,其结果证实了该模型的优越性。结果表明,提出的组合模型具有更好的性能,其纳什效率系数为0.957,平均绝对误差为1.73,均方根误差为2.88。因此,EMD-ATT-BiLSTM模型是一种可行的日径流预报方法。

Accurate and reliable runoff forecast plays an increasingly important role in the optimal management of water resources.In order to improve the prediction accuracy,a neural network model is proposed to forecast daily runoff.The model combines the empirical mode decomposition(EMD)method,the attention mechanism,and the BILSTM neural network.The interpolation method is used to improve the accuracy of the input data.EMD method can decompose the unsteady nonlinear runoff time series into multiple groups of eigenmode components and trend terms to realize the steady state of the input time series,and then give different degrees of attention to the time series through the attention mechanism,and then forecast and reconstruct the time series respectively through BILSTM.The proposed model was applied to the daily runoff data of Qingxi River Station in Xuanhan County,Sichuan Province,and compared with other three neural network models,namely LSTM,ATT-LSTM and ATT-BILSTM,the results show the superiority of the proposed model.The results show that the proposed combined model has better performance,and its NSE=0.957,MAE=1.73 and RMSE=2.88.Therefore,EMD-ATT-BILSTM model is a feasible daily runoff forecast method.