Short-term Runoff Prediction Optimization Method Based on BGRU-BP and BLSTM-BP Neural Networks

Abstract

Runoff forecasting is one of the important non-engineering measures for flood prevention and disaster reduction. The accurate and reliable runoff forecasting mainly depends on the development of science and technology, many machine learning models have been proposed for runoff forecasting in recent years. Considering the non-linearity and real-time of hourly rainfall and runoff data. In this study, two runoff forecasting models were proposed, which were the combination of the bidirectional gated recurrent unit and backpropagation (BGRU-BP) neural network and the bidirectional long short-term memory and backpropagation (BLSTM-BP) neural network. The two models were compared with the gated recurrent unit (GRU), long short-term memory (LSTM), bidirectional gated recurrent unit (BGRU), and bidirectional long short-term memory (BLSTM) models. The research methods were applied to simulate runoff in the Yanglou hydrological station, Northern Anhui Province, China. The results show that the bidirectional models were superior to the unidirectional model, and the backpropagation (BP) based bidirectional models were superior to the bidirectional models. The bidirectional propagation was conducive to improving the generalization ability of the model, and BP neural network could better guide the model to find the optimal nonlinear relationship. The results also show that the BGRU-BP model performs equally well as the BLSTM-BP model. The BGRU-BP model has few parameters and a short training time, so it may be the preferred method for short-term runoff forecasting.