Data-driven linear parameter-varying predictive control for process systems

Abstract

In this paper, we propose a method for model predictivecontrol of linear parameter-varying (LPV) systems describedin an input-output (IO) representation and subject to inputandoutput constraints. By exploiting knowledge of the futuretrajectory of the scheduling variable, the on-line computationsrequired are the computation of a nominal parameter-varyingprediction model and the solution of a constrained quadraticprogram. Closed-loop asymptotic stability is guaranteed by extendingthe on-line optimization problem with a suitable terminalcost and terminal set constraint. After converting the LPV-IOrepresentation to an equivalent linear fractional representation(LFR), the terminal cost is obtained by solving a robust controlproblem formulated in terms of linear matrix inequalities. Theterminal set is computed as the maximal level set of the Lyapunovfunction associated with the robust controller. In practice, thefuture scheduling sequence of the LPV system can be unknown.We propose to estimate the future trajectory using a modelidentified on-line from measured data. By using this estimation inthe prediction model, improved control performance is obtained.The effectiveness of the proposed approach is demonstrated onseveral simulation examples, including a continuous stirred-tankreactor and a simplified polymerization reactor.Index Terms-Linear parameter-varying systems, model predictivecontrol, input-output representation, linear fractionalrepresentation, stability, support vector machine

Date of Award	30 Aug 2014
Original language	English
Supervisor	Roland Tóth (Supervisor 1)