Predictive control of a water distribution system based on process historian data

Abstract

In this paper, a heuristic historian data-based predictive control strategy is presented and used to control a water distribution system simulated using the EPANET software, in particular, the Richmond water distribution system. The control actions are computed based on past historian data. The historian stores closed-loop operation data of the process with different controllers used in the past, which may not provide sufficient information for a precise system nor controller identification. The proposed predictive controller computes the current control actions as a weighted sum of past control actions so that an estimation of the performance cost over a prediction horizon is minimized. Only a subset of the past control actions in the historian close to the current state of the process is considered in the current control computations to carry out a local linearization. This predictive strategy is well suited to control applications of large and complex processes for which it is difficult to carry out identification experiments such as water distribution systems. In the application example, the trajectories of a set of relay controllers are used through the proposed approach to take into account pressure constraints and periodic references.