Interpretable hybrid artificial intelligence model for predicting daily hydropower generation of cascade hydropower reservoirs

To tackle the challenges posed by low prediction accuracy and the elusive internal mechanisms of traditional models in daily hydropower generation prediction, this paper introduces a hybrid artificial intelligence method. Initially, the MIC method is employed to ascertain the optimal number of input data. Secondly, the CNN model is leveraged to extract potential features from the input data. Subsequently, the ResLSTM model is used to capture the dependencies between the processed data and make predictions. Cascade hydropower reservoirs experimental results across multiple forecast periods demonstrate that the MIC-CNN-ResLSTM method surpasses comparison models in terms of stability and robustness. Furthermore, this paper introduces SHAP theory to elucidate the impact of input data on the daily hydropower generation predictions. Results indicate that historical hydropower generation and reservoir inflow have a significant influence on the prediction outcomes. In conclusion, this paper presents an effective and interpretable daily hydropower generation prediction method, providing valuable insights for power system dispatch operations and the rational development and utilization of water resources.