Multivariate Identification and Stochastic Control of A Pilot Plant Packed Bed Reactor

Abstract

<p>The modelling and control of a pilot plant packed bed catalytic reactor is investigated. The reactions considered are those of the hydrogenolysis of n-Butane over a nickel on silica gel catalyst. These involve several independent species and are highly exothermic. A multivariate transfer function model for process and a time series model for the disturbances are constructed through an iterative procedure of process identification, parameter estimation and diagnostic checking. Using exit concentrations inferred from reactor bed temperatures and inlet flow measurements, a multivariate stochastic controller, designed to minimize a quadratic function of the production rates of various intermediate products subject to constraints on the variances of the manipulated input flowrates is implemented on-line using a process minicomputer. The parameters of the inferential concentration equation are updated on-line using infrequent data from a process gas chromatograph. The performance of this feedback controller is found to be quite satisfactory.</p>