Abstract: The optimal operation of a single pumping station is crucial for managing long-distance water transfer projects.Current methodologies inadequately address the intricate constraints and uncertainties inherent in the operation and maintenance of such facilities.This study introduces an enhanced Soft Actor-Critic (SAC) algorithm featuring an adaptive discount factor and an improved cost function,grounded in the maximum entropy reinforcement learning framework.This refined approach redefines the optimal operation of the pumping station into a Constrained Markov Decision Process (CMDP) that prioritizes both reliability and economic efficiency.Real-time optimization of the pumping strategy is conducted in response to the environmental conditions of the station.Through analysis of historical data from the Denglou Pumping Station,part of the East Route of the South-to-North Water Diversion Project,this method’s effectiveness and superiority are validated against other advanced deep reinforcement learning algorithms and traditional optimization techniques.The results reveal that the proposed method achieves greater convergence stability and operational efficacy.It ensures more consistent and cost-effective performance with minimal adjustments to blade angles and reduced frequency of unit start-stop actions,thereby offering a robust decision-making framework for pumping station management.