Generalized predictive control with dynamic filtering for process control applications

Open loop gains in process control applications often vary due to deterioration in control surfaces and actuator performance. This results in substandard product quality in spite of periodic retuning of conventional PI controller gains. In addition, sensors used for the measurement f of process parameters are subject to high frequency measurement noise. Compensation of this noise by the control algorithm results in excessive actuator activity. A control algorithm that provides optimal control, at the same time capable of incorporating changes in the process gains is then desired. This paper describes the implementation of GPC for the control of mold level in the continuous casting process. This involves development of a suitable model of the process and changes made to the identification algorithm in order to enhance accuracy of the identified model. A modification to the GPC cost function is suggested to account for measurement disturbances. This is done by dynamically filtering the predicted free response of the process model before the total future response of the model is computed and then weighted in the GPC cost function. Experimental results that compare performance of the modified GPC with that of PI control are presented.