An Elementary Pattern Recognition Self-Tuning PI-Controller

Abstract This paper deals with the modelling of an adaptive control system from an informational point of view in general and the design of an adaptive pattern recognition Pi-controller in particular. The design starts off with a state-and-event oriented overall adaptation strategy so that the resulting adaptive contoller can be modelled as a finite state machine. The presented closed-loop adaptation strategy is based on pattern recognition control which yields essentially process model-free tuning methods. The three basic tasks in the design of adaptive pattern recognition controllers (APRC) are formulated and treated sepatately: (1) pattern description, (2) definition of control objectives in terms of patterns and (3) mapping the patterns to controller settings. In this context, the powerful Elementary Pattern Description (EPD) technique is presented solving the problem of information loss adherent to the other known pattern description techniques. A parameter-adaptive PI controller based on this EPD approach yields good results in simulation studies for linear time-varying stable minimum/nonminimumphase processes with dead time. The application to laboratory plants for the control of flow, pressure and temperature was also sucessful.