Optimal Control for Water Distribution Networks with Unknown Dynamics

Abstract

Optimal control for Water Distribution Networks (WDN) is subject to complex system models. Typically, detailed models are not available or the implementation is too expensive for small utilities. Reinforcement Learning (RL) methods are well known techniques for model-free control. This paper proposes a model-free controller for WDNs based on RL methods and presents experimental evidence of the practicality of the design.