Fuzzy model-based predictive control by instantaneous linearization (Q5937160)

A multi-step model-based predictive control (MBPC) algorithm for complex or badly defined processes is constructed. The algorithm is based on a local linear model, which is calculated by instantaneous linearization of the fuzzy process model around the current operating point. The product-sum crisp-type fuzzy process model is exposed as a nonlinear autoregressive (NARX) model with an exogenous input variable. Instead of recursive estimation, a linear time-invariant ARX model is extracted from the original NARX model at each sample instant. It is argued that by applying the proposed model linearization, it is possible to use any long-range predictive linear controller design method. The control of pH in a continuous stirred tank reactor is chosen as a realistic nonlinear case study for the proposed MBPC algorithm.