Deep reinforcement learning based valve scheduling for pollution isolation in water distribution network.

Area of Science:

• Environmental Engineering
• Water Resource Management
• Artificial Intelligence in Infrastructure

Background:

• Public water supply facilities, particularly water distribution networks (WDN), are vulnerable to intentional contamination.
• Frequent water contamination incidents pose significant societal risks and economic losses.
• Effective contamination isolation and contaminant reduction in WDN are critical challenges.

Purpose of the Study:

• To propose a novel reinforcement learning (RL) based method for real-time valve scheduling in WDN.
• To address the challenge of effectively isolating contamination and minimizing residual contaminant concentrations.
• To develop a system capable of learning optimal scheduling policies without precise source characterization.

Main Methods:

• Utilizing sensor data as the state input for the RL agent.
• Defining valve scheduling as the action space for the RL agent.
• Training the RL agent to learn an effective contamination isolation policy.

Main Results:

• The proposed RL algorithm demonstrated effective isolation of contamination events in simulations.
• The method successfully reduced the residual concentration of contaminants within the WDN.
• The approach proved robust in handling uncertain contamination events.

Conclusions:

• Reinforcement learning offers a promising solution for real-time valve scheduling to enhance water safety.
• The developed method can significantly mitigate risks associated with intentional water contamination.
• This approach provides a scalable and adaptive strategy for protecting public water supplies.