Research on Reservoir Operation Rules Extraction Based on Decision Tree and Its Ensemble Model

Abstract

Reservoir operation rules, as an important tool to guide reservoir operation, are not only the decision-making reference in the period of reservoir planning and design, but also one of the key technologies affecting the comprehensive benefits of the reservoir in the stage of operation and management. Therefore, the reservoir operation rules extraction method based on decision tree and its ensemble model was proposed in this paper. The time index, early water level, inflow, outflow and current inflow were selected to form the input factor set with the reservoir scheduling principle. Comprehensively considering the characteristics of operation data and the principle of decision tree, reservoir water level at the end of period was determined as the model output and then the corresponding simulated water level evaluation index was proposed based on the reservoir regulation capacity. Furthermore, in view of the difficulties in selecting input factors and calibrating the hyperparameters of models, the F test method and mutual information were used as the correlation evaluation index of model input factors, and the Tree-structured Parzen Estimator was introduced. Finally, the proposed methods were applied in Ahai and Ertan Reservoirs, the results showed that the extracted operation rules have high precision, which demonstrated the proposed methods is practicable and effective in the reservoir operation rule extraction.