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

Public water supply facilities are vulnerable to intentional intrusion. In particular, Water Distribution Network (WDN) has become one of the most important public facilities that are prone to be attacked because of its wide coverage and constant open operation. In recent years, water contamination incidents happen frequently, causing serious losses and impacts to the society. Various measures have been taken to tackle this issue. Pollution or contamination isolation by localizing the contamination via sensors and scheduling certain valves have been regarded as one of the most promising solutions. The main challenge is how to schedule water valves to effectively isolate contamination and reduce the residual concentration of contaminants in WDN. In this paper, we are motivated to propose a reinforcement learning based method for valve real time scheduling by treating the sensing data from the sensors as state, and the valve scheduling as action, thus we can learn scheduling policy from uncertain contamination events without precise characterization of contamination source. Simulation results show that our proposed algorithm can effectively isolate the contamination and reduce the risk exclosure to the customers.