LONG-RANGE PREDICTIVE CONTROL OF A POLYMERIZATION REACTOR

This paper examines the application of generalized predictive control (GPC), one in the class of long-range predictive algorithms, to the control of conversion of methyl methacrylate (MMA) monomer in a simulated CSTR, and to the control of temperature in a pilot plant batch polymer reactor. The control objective is regulation in the presence of (i) stochastic disturbances due to impurities (in the case of the CSTR), and (ii) pulse disturbances from the addition of cold solvent and initiator (in the case of the batch reactor). The role of the observer polynomial as a detuning parameter for trading off performance against variability in the control action is emphasized. Also, the role of data prefiltering, prior to model parameter estimation, is examined. A frequency domain interpretation of the least squares estimation algorithm is used to clarify the role of the filter.