FORECASTING STREAMFLOW USING A HYBRID CNN-BILSTM MODEL WITH ATTENTION MECHANISM: A CASE STUDY FOR TUYEN QUANG HYDROPOWER RESERVOIR

Abstract

Forecasting the streamflow to hydropower reservoirs plays a crucial role in optimizing the operation and management of hydropower plants, ensuring efficient energy production and rational water resource utilization. This study proposes a hybrid deep learning model that combines a convolutional neural network with a bidirectional long short-term memory network integrated with an attention mechanism for hydropower inflow forecasting. To evaluate the forecasting capability in real-time scenarios, the data was resampled at a daily frequency, and predictions were made for the next 1 day, 1 week, and 1 month. The proposed model was compared with various deep learning models, including CNN, LSTM, CNN-LSTM, and CNN-BiLSTM. The results demonstrate that the proposed model outperforms the others, particularly in the 1-week forecast, achieving MAE, RMSE, and N-RMSE values of 176.52 m³/s, 427.29 m³/s, and 7.47%, respectively. These findings highlight the model’s potential to support more informed and efficient reservoir operation decisions. By doing so, it contributes to improving the reliability of water resource management and optimizing hydropower output under dynamic operating conditions.