ON PRACTICAL IMPLEMENTATION OF ADAPTIVE ALGORITHMS IN COMPUTER PROCESS CONTROL

Abstract

This thesis investigate the problem of using self-tuning control in the presence of a load disturbance and a periodic disturbance. Two versions of the Clarke and Gawthrop generalized minimum variance algorithm are considered. The first algorithm is developed from an ARMA model where a dc term is used to model the output offset. This dc term is estimated separately from the dynamic parameters. The second algorithm is developed from an ARIMA model where the output offset is modelled by an integration of a white noise. In the presence of a periodic disturbance, the unmodified generalized minimum variance self-tuning algorithm does not work well. A pseudo feedforward method is proposed to overcome the problem. The method begins with an estimate of the disturbance frequency. A second stage allows the estimate of the disturbance frequency to be improved. The method has been shown to reduce the effect of the periodic disturbance. A simple scalar variable is proposed as an indicator of load disturbance. Since appreciable errors appear in the parameter estimates when severe load disturbance occurs, an intuitive idea to reduce the error is to freeze the estimation when load change occurs. This idea is evaluated in this thesis