Improving adaptive fuzzy control performance by speeding up identification: Application to an electric furnace

The development and application to a physical system of an adaptive predict ive fuzzy controller is presented. The target process is a small electric f urnace used in ceramic manufacturing. The pr,posed controller attempts to m inimize a multi-step quadratic cost under the assumption that the control a ctions are all free over the prediction horizon. The control law relies on a simplified fuzzy relational model identified on-line. A convenient select ion of the triangular norms used in the composition operator is made for al lowing the application of Recursive Least Squares (RLS) to fuzzy relational structures, thus speeding up identification. It is shown that for a partic ular selection of norms, relational structures require less parameters to b e described, and they can be interpreted as a set of simplified fuzzy rules . Examples are presented in which the controller developed outperforms othe r controllers with identification procedures based on gradient.