Hydrological time series forecast model based on wavelet de-noising and ARIMA-LSTM

Abstract

Hydrological time series is affected by many factors and it is difficult to be forecasted accurately by traditional forecast models. In this paper, a hydrological time series forecast model based on wavelet de-noising and ARIMA-LSTM is proposed. The model first removes the interference factors in the hydrological time series by wavelet de-noising, and then uses ARIMA model to fit and forecast the de-noised data to obtain the fitting residuals and forecast results. Then we use the residuals to train LSTM network. Next, the forecast error of the ARIMA model is forecasted by LSTM network and used to correct the forecast result of ARIMA model. In this paper, we use the daily average water level time series of a hydrological station in Chuhe River Basin as the experimental data and compare this model with ARIMA model, LSTM network and BP-ANN-ARIMA model. Experiment shows that this model can be well adapted to the hydrological time series forecast and has the best forecast effect.