Short term water level prediction based on C-Stacking ensemble model

This paper proposes a C-Stacking ensemble model for short-term prediction of water level time series. First, the method of complete empirical mode decomposition with adaptive noise (CEEMDAN) is used to decompose the original water level sequence into high, low frequency and residual sequences, and the method of wavelet threshold denoising is used to eliminate the noise of the high frequency sequence, and then support vector regression (SVR), back propagation neural network (BP) and long short-term memory neural network (LSTM) models are used as the primary learner to predict the denoised water level sequence, and finally the linear model is used to integrate the predicted values of the three primary learners to obtain the final predicted value. The experimental analysis of Hongze Lake water level data shows that the ensemble model has better prediction accuracy and applicability than a single model.