REINFORCEMENT LEARNING IN THE OPERATIONAL MANAGEMENT OF A WATER SYSTEM

摘要

In this paper we present a variant of the classic Q-learning algorithm to design an operating policy for the management of a multi-purpose water reservoir. The model-free approach is applied only to a part of the model of the water system, the catchment, while the remainder, the reservoir, is described by mathematical models based on physical characteristics. This algorithm has been tested on the case of the regulation of Lake Como.