Fuzzy Sarsa Learning of Operational Instructions to Schedule Water Distribution and Delivery

Operational instructions have a major role in improving water delivery performance in irrigation canals. Of different delivery systems, on-request systems have higher flexibility than rotational ones and can be applied to present irrigation networks with minor changes. The main challenge in on-request systems is determining appropriate operational instructions. The main objective of this paper is formulating a fuzzy Sarsa learning (FSL) system for learning operational instructions. As a result of its properties, including being model free and generalizable, FSL can learn in a limited number of situations and suggest operational instructions for a wide range of conditions. The FSL steps in an irrigation system include present depths observation and flows, selection and execution of the operational instructions, observing new depths and flows, and assigning reward to the applied operational instructions and learning accordingly. The physical and hydraulic data of the East Aghili Canal are used for evaluation of our FSL. The results support the suitability of FSL in terms of performance with respect to generalization and convenience of deployment; the total operation time decreased from 23.1 h in conventional operation to 5.76 h in FSL, the efficiency and adequacy indicators obtained were close to desired values, and water level deviations were less than 10%. Copyright © 2016 John Wiley & Sons, Ltd.