DRL-Based Scheduling Optimization Strategy for the Pumping System of Ninghai Xinghai Wastewater Treatment Plant

Against the backdrop of global energy shortages and stricter environmental-protection requirements, the energy-consumption issue in wastewater treatment plants has become prominent, especially in pumping systems. Traditional pump-group scheduling methods struggle to adapt to complex working conditions, leading to severe energy waste. Reinforcement learning holds great potential in system optimization. This paper focuses on the pumping system of Ninghai Xinghai Wastewater Treatment Plant, integrates MS-LSTM and ARWA-PPO, and proposes a DRL-based pump control strategy. The contributions of this study are as follows: It can precisely adapt to working conditions and improve the efficiency of pump groups; enhance the scientific nature of decision-making; outperform expert scheduling and other methods in performance, with a 3.3% reduction in energy consumption, zero water-level violations, and an 8.3% reduction in the number of pump switches. This research achieves intelligent scheduling, contributes to energy conservation in the industry, and provides a reference for similar plants.