Use of Reinforcement Learning in the Modeling of Ring-Type Water Networks

Abstract

In this paper it is proposed to use reinforcement learning (RL) based predictive control of ring-type water supply network for finding the optimal path from point of water supply to point of water consumption avoiding some obstacles that may arise along the way. An example of an obstruction may be a defect in a section point (consumer) through which the water cannot flow. Ring-type water supply network consists of a number of closed rings surrounding the consumers while supplying them with water through sections. In this paper, a computer program was developed in LabVIEW in order to calculate the optimal path from point of water supply to point of water consumption based on reinforcement learning techniques. In order to achieve the goal, Markov Decision Processes, Q-Learning, Bellman Equations and Policy Gradients were used. The results show the optimal path.