Fuzzy-neural predictive control using fast optimisation polices

This paper describes the development of fast optimisation polices based on Newtonian approaches, as effective algorithms to solve the on-line optimisation task, during the operation of a predictive controller. To simplify the calculation of the control actions, an iterative solutions based on Newton-Raphson and Levenberg-Marquardt approaches, are proposed. To avoid the computational load related to Hessian inversion, a simple Gaussian elimination in a form of matrix decomposition is applied. As plant response predictor, a Takagi-Sugeno fuzzy-neural network, with global and local (after the rules layer) recurrent nodes, is used. The efficiency of the proposed optimisation strategies is demonstrated by simulation experiments in MATLAB environment to control a continuous stirred tank reactor.