Daily Runoff Prediction in Xijiang River Basin Based on FOA‐TCN‐BiLSTM Model

ABSTRACTAccurate and reliable daily runoff forecasting plays a vital role in water resource management, flood warning and operational scheduling. However, runoff prediction is challenging due to its nonlinear and non‐stationary nature, influenced by climate change, topography and human activities. To improve forecasting accuracy, this study proposes a hybrid TCN‐BiLSTM model optimised by the Fruit Fly Optimization Algorithm (FOA) for daily runoff prediction in the Xijiang River basin. The model first utilises the Temporal Convolutional Network (TCN) to extract temporal features, then employs the Bidirectional Long Short‐Term Memory (BiLSTM) network to capture temporal dependencies, and finally optimises key hyperparameters of the model using the FOA to enhance overall performance. Taking the four hydrological stations in the Xijiang River basin, including WX, WZ, DHJK and GG, as examples, the model exhibits outstanding performance in both single‐step and multi‐step prediction tasks. Taking the WX station as a representative example, the model achieved an MSE, MAE, and R2 of 0.888 × 106 m3/s, 0.530 × 103 m3/s and 0.960 on the test set, respectively. Compared with the BiLSTM model, the MSE and MAE decreased by 63.27% and 40.69%, while the R2 increased by 7.49%. Compared with the TCN model, the MSE and MAE decreased by 59.60% and 39.71%, with an R2 improvement of 6.31%. Relative to the TCN‐BiLSTM model, the MSE and MAE were reduced by 43.15% and 26.38%, and the R2 increased by 3.11%. Moreover, the R2 values for the test sets at all four stations reached 0.955 or higher, further confirming the model's stability and generalisation capability across multiple regions. The results indicate that the FOA‐TCN‐BiLSTM model demonstrates significant advantages in enhancing runoff prediction accuracy and generalisation, making it particularly suitable for practical engineering applications such as flood forecasting, water resource management, and regional hydrological risk assessment, thus holding promising application prospects.