Flood flow prediction based on combined CNN-GRU-XGBoost model

Flood flow prediction can prevent flood disasters and reduce flood losses, which is of great significance for ensuring the safety of people's lives and property. Aiming at the problem that a single prediction model is difficult to effectively analyze the pattern of historical hydrological data leading to low prediction accuracy, a multi-model hybrid CNN-GRU-XGBoost flood flow prediction model is proposed. By constructing a GRU model with a feature pre-extraction function module, and combining it with XGBoost by assigning weights using the reciprocal error method, flood discharge prediction is achieved. Based on the flow, water level, and rainfall data of Yongzhou section of the Xiangjiang River basin, a combined model was used to predict the flow. The experimental results show that the RMSE (Root Mean Square Errors), MAPE (Mean Absolute Percentage Error), and NSE (Nash-Sutcliffe Efficiency) of the combined CNN-GRU-XGBoost model are 23.742m³/s The combined model structure achieves a significant reduction in the error index compared with the single network model, verifying that the flood flow prediction method based on the combined CNN-GRU-XGBoost model has a lower prediction error and higher prediction accuracy.