Tuning of Predictive Controllers for Drinking Water Networked Systems

Abstract

In this paper, two tuning strategies for a multi-objective predictive controller applied to a drinking water network (DWN) are proposed. A control-oriented DWN model is briefly reviewed, together with its management objectives. A comparison of methods to explore the Pareto front of the multi-objective optimisation (MOO) problem behind the predictive controller is presented with an effective normalisation method for the model predictive control (MPC) objectives. The proposed tuning strategies, applied to a real-life case study, are compared. Finally, simulation results show that the proposed MPC tuning strategies outperform the baseline results.

Modeling and real-time control of urban drainage systems: A review

“…An MPC controller is composed by four main elements: a mathematical (control-oriented) model of the system, a cost function that expresses the control objective to be achieved, a set of system constraints (of bounding and operational nature), and a finite-horizon open-loop optimization problem, which is solved at each time instant [51]. Figure 1 shows the basic scheme of an MPC controller in a closed-loop topology.…”

“…MPC is a model-based control strategy that uses a prediction of the system response to establish an appropriate control action [51,52]. This strategy makes an explicit use of a mathematical model of the process to generate a sequence of future actions within a finite prediction horizon.…”

Evolutionary Game-Based Dynamical Tuning for Multi-objective Model Predictive Control

“…The selection of a management point as in [22] implies to have to compute several different prioritization weights at each iteration in order to find the proper combination of weights. This procedure must be made every iteration since conditions in the system vary over time as disturbances in the system also vary.…”

“…A region is selected over the Pareto front instead of a point as reported in [22]. The selection of a management point as in [22] implies to have to compute several different prioritization weights at each iteration in order to find the proper combination of weights.…”

“…For instance, a self-tuning terminal cost approach is applied in [13] for an economic MPC controller. In [22], a normalization procedure and a computation of the minimum distance from the Pareto front to a management point have been proposed as a tuning strategy. Other approaches use learning systems.…”