(240d) Decentralized Direct Adaptive Control for Process Networks

An approach to decentralized direct adaptive control using normalized recursive least squares parameter estimation is presented. A novel extension to an infinite time horizon of Ydstie's switching lemma is used to show boundedness of each local closed-loop adaptive system. This allows for the phenomenon of bursting encountered in adaptive algorithms to be accounted for. Motivated by application to chemical process networks, the overall system is modelled as set of nodes interconnected via a network with known and time-invariant topology. The nodes themselves are modelled as discrete-time linear single input single output systems, possibly with time-varying coefficients. Additional signals are introduced to represent interconnections and unmodelled dynamics. Stability of the overall system is then shown by applying dissipative systems theory. The dissipativity properties of the individual closed-loop adaptive systems are used in conjunction with the process network topology to determine the dissipativity of the overall system. This is then used to show stability and minimum performance, in the form of L2-gain bounds, of the overall system. This approach facilitates a scalable approach to control of large-scale chemical systems. In the presentation simulation results for some academic examples will be presented to exemplify the theoretical results which have been proposed.