Abstract: River flow is an important ecological indicator for monitoring hydrological issues and managing water resources. Flow forecast is also significant for providing guidance and reference for water conservancy construction, navigation planning, and water resources dispatching. In the present research, a VMD-BP (variational mode decomposition and back propagation) model of forecasting river flow is proposed and constructed by combining the advantages of VMD in dealing with non-stationary sequences and the ability of BP neural network in tackling nonlinear fitting problems. The model is verified by using daily water level and flow data in 1998 and 1999 at Yichang Hydrological Station of the Yangtze River. Results indicate the VMD-BP model solves the multi-value relations between water level and flow to some extent and mitigates the volatility of data. The predicted result is better than those of linear fitting regression model and BPNN model, and the prediction error is merely 1.61%. Therefore, the VMD-BP model can be considered as an effective method for river flow prediction.

Key words: river flow, variational mode decomposition, BP neural network, prediction model, stage-discharge relation, Yichang hydrologic station