Experimental Evaluation of a Kalman Filter based Multistep Adaptive Predictive Controller

A Kalman-Filter based Predictor is used for Long Range Predictive Control (LRPC). Special attention is focussed on the performance of the predictor in the presence of time-delay mismatch. The ratio of the measurement and process noise covariances is found to be a very important tuning parameter for the predictor. By increasing the value of this parameter, the predictor gives good future predictions even in the presence of significant time-delay mismatch. The predictor was coupled with a LRPC scheme and the controller was tested on a pilot-scale stirred tank heater, where time-delay mismatches can be easily contrived. The controller not only gave satisfactory servo and regulatory performances in the presence of relatively large time-delay, but it is also very simple to use.