Multivariable adaptive fuzzy TSK model-based predictive control with feedforward

Abstract

In multivariable control the study of loop interactions is of prime importance. The P or V-canonical form structures, where loop interactions are dealt with as feed-forward couplings, are popular transfer function representations used to describe multivariable processes. Another alternative design consists of a bank of several Single Input Single Output (SISO) controllers linked by feed-forward terms representing in fact the interactions as disturbances. This is the focus of this paper which proposes a new long-range predictive control algorithm for multivariable processes with feed-forward based on the popular Generalised Predictive Control (GPC) algorithm but using a Takagi-Sugeno Kang (TSK) piece-wise fuzzy modelling approach. To improve estimation a Long Range Predictive Identification (LRPI) algorithm for fuzzy modelling is also integrated within the approach. The performance of the adaptive control scheme is assessed using a series of experiments on the binary distillation column. The new proposed algorithm with a fuzzy model is shown to be more robust than the standard GPC algorithm which uses a crisp CARIMA model in terms of handling loop interactions.