Sophisticated Dynamic Adaptive Control of a Polymerization Process

The chemical process considered serves as an appropriate paradigm of multivariable dynamic systems of strong non-linear coupling in the control of which the state propagation of various internal degrees of freedom can neither be measured nor directly controlled. The desired output is a single real, nonlinear function of these quantities. In the present example only a single input variable is used for control purposes. In the paper quantitative model of the process is presented. For controlling this process various approaches were applied: genetic programming for the identification of the process, an ARMAX-type floating basis vector approach in the quasi-stationary limit, and fuzzy-type adaptive control with fixed and with variable speed in which the adaptation rule was determined on qualitative considerations. In the present approach the adaptation rule is determined in a sophisticated manner, on the basis of a modified version of the renormalization transformation, in which the system is observed real time. The quality of the control is investigated via simulation from the points of view of its robustness with respect to setting its free parameters, and sensitivity to the measurement noises. It is concluded that at the time-scale of about 0.067 s sampling time the ‘dynamics’ of the controlled process can well be traced, and on the basis of a simple planning method quite accurate dynamic control can be achieved.