NONLINEAR RULE-BASED MODEL-PREDICTIVE CONTROL OF CHEMICAL PROCESSES

Several difficulties are still encountered in the direct use of a nonl inear model in the area of model based control. A time series rule bas ed model is employed in this work to perform nonlinear control in case s where linear approaches have failed. The rule based model is basical ly comprised of a set of rules which are related to the time series of input and output data. The proposed control approach filtered out the high-frequency disturbances using possibility theory. An on-line iden tification phase is required if persistent changes of some parameters frequently occur. The identification algorithm maximizes the membershi p of the disturbance parameter in the immediate past. The control obje ctive minimizes the square errors of the output and set point in a tim e horizon projected into the immediate future. Physical examples are s imulated to demonstrate the implementation of this approach.