A Neural Pattern Analyzer for Adaptive Process Control

This work details an adaptation strategy based on an analysis of patterns exhibited in the recent history of the controller error and manipulated process input variable. The focus of the strategy is on adaptation requirements due to load disturbances. A quantizing network is studied in the role of a pattern analysis tool for implementing the method. The strategy is limited to a single parameter adaptation where the gain of the controller´s internal model is the adjustable parameter. Details and a demonstration of the method are presented using a simulated process constructed to challenge the strategy. A model based PI algorithm is the controller employed in this work.