Disturbance pattern classification and neuro-adaptive control

An adaptation strategy is given that is 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. The manipulated input pattern associated with a disturbance is analyzed to determine whether the disturbance has the potential for changing process character. If so, the corresponding controller error pattern is analyzed to determine the appropriate adaptation to the controller´s internal model. A vector quantizing neural network is studied as 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 employed.<>