Adaptive wastewater flow control using AquaFlowNet with reinforcement learning for urban drainage optimization

Abstract Conventional wastewater flow-control strategies often fail to respond effectively to rapidly changing hydraulic conditions, resulting in increased overflow events, excessive energy consumption, and reduced operational resilience. Existing AI-based approaches primarily focus on prediction or process-level optimization, offering limited capability for autonomous, system-wide flow management. To overcome these limitations, this study introduces AquaFlowNet, a reinforcement learning–driven framework designed for real-time wastewater flow optimization across interconnected sewer networks. The proposed RL-DFC agent dynamically learns optimal valve and pump actions from environmental feedback, enabling adaptive control under uncertain inflow conditions. Simulation results demonstrate that AquaFlowNet achieves substantial improvements over existing rule-based and machine learning controllers, including a 52% reduction in total overflow volume, up to 90% reduction in overflow events, 35–40% improvement in peak-flow attenuation, and 230 kWh reduction in aeration-related energy demand. These findings indicate that AquaFlowNet can transform wastewater management from prediction-centric approaches into autonomous, resilient, and scalable network-level control, supporting the development of next-generation smart wastewater infrastructure.