Water level prediction using long short-term memory network with residual structure based on ensemble empirical mode decomposition

Abstract

To solve the problem about low accuracy of water level prediction caused by its nonlinearity and high noise, a water level prediction model based on EEMD-RESNET-LSTM is proposed. First, EEMD (Ensemble Empirical Modal Decomposition) is used for data noise reduction. Then, the RESNET-LSTM model is employed to predict the multiple components and residual term obtained after EEMD decomposition. Finally, the individual prediction results are aggregated to obtain the actual water level predictions. Compared to the three models, LSTM, RESNET-LSTM and EEMD-LSTM, the performance of this model is improved. Its RMSE is 0.127 m, MAE is 0.102 m, and R ² reaches 94.5% on water level prediction of Hongze Lake.