A Water Level Prediction Model Based on ARIMA-RNN

Abstract

Accurate water level prediction has important guiding value for scientific decision-making and planning. In order to fully excavate the information in the water level and improve the accuracy of water level prediction, a new water level prediction method is proposed. On the basis of ARIMA and RNN model, a new scheme based on ARIMA-RNN combined model for water level prediction is proposed. This method solves the problem that a single forecasting model can't take into account both linear and nonlinear components in data, and also solves the problem of precision reduction caused by simple addition of linear and nonlinear components in traditional combination schemes. In this scheme, the linear correlation components in the data are predicted by the Autoregressive Integrated Moving Average Model (ARIMA), and the nonlinear components are predicted by the Recurrent Neural Network (RNN), and the relationship between the two components is constructed by RNN. The model uses the data of daily water level and environmental factors related to water level as input vectors, and the water level in the next 30 days as output vectors. Experiment in Taihu Lake proves the validity of the model. ARIMA and RNN models were established to predict water level, and the results were compared with the results of the model proposed in this paper. It was found that the RMSE of ARIMA-RNN model was the smallest. The experimental results proved that the prediction model proposed in this paper could achieve better results.