Genetic design of robust predictive control systems
Despite the fact that PID controllers are undoubtedly the most popular controllers used in industrial control processes for decades, they do not perform well when applied to systems with significant time-delay. Consideration of this problem led to the development of predictive control strategies in the 1950's and lately in the 1980's. Perhaps the best-known predictive control techniques that has received the most attention in the control of long time-delayed plants and used almost exclusively in the process industries are the Smith Predictor Controller (SPC) and more recently the Internal Model controller (IMC). These two controllers are amenable to conventional robustness measures such as Gain-margin, Phase-margin, Delay-margin and Modulus-margin, which make them suitable for comparison with conventional PID type controllers. The application of Evolutionary Algorithms to process control systems constitutes a new methodology within the CACSD. In recent years, the controls research community have become increasingly interested in the use of genetic algorithms as a means to control various classes of systems, however, the technique has concentrated on the unity feedback control system design problem in both the SISO and MIMO cases and there are still a lot of research topics that are not addressed (or properly addressed), in the application of genetic algorithms. In this research, the genetic algorithm has been adopted as a major control design tool for designing robust predictive control systems for process plants. As much work on predictive control has been done using low order Pade' approximatios (Morari, 1989), it is interesting to further investigate the robustness of IMC and SPC by not using Pade' approximations and deploying the concept of delay-margin to assess robustness. In this context, a new design methodology is proposed for designing robust IMC controllers under pre-prescribed gain-margin and delay-margin constraints where controller tuning is replaced by simple design curves. The robustness analysis of a single parameter Smith predictor controller (SP-SPC) in comparison with a proportional plus integral Smith predictor controller (PI-SPC) revealed that under robustness consideration, TMC controller and Pi-Smith predictor controller are almost identical, however, IMC is preferred because the design curves developed in this thesis can be used to automate the design. hi this thesis also, Competitive Co-evolutionary Algorithm is proposed as a new technique to design robust IMC controllers that can cope with large parametric uncertainty. The co- evolutionary technique proposed is capable of determining the most difficult plants to control, both in terms of stability and performance and can simultaneously select the optimal nominal parameter values for the model used in the IMC controller. A new gain-scheduled controller based on Internal model controller architecture is proposed. This controller is novel in the sense that no controller design phase is needed as the case with conventional gain-scheduled controller design. The only task required in the gain-scheduled IMC controller design is the identification and mapping a set of locally linearised models to fit the non-linear process. The design curves can be used to assure robustness in terms of GM and DM for all the locally linearised control systems. In this context, such gain-scheduled controllers have been designed for a non-linear water tank system and a non-linear heating exchanger (laboratory process trainer).