DATA-BASED MODELLING OF NONLINEAR SYSTEMS USING A MODIFIED LOLIMOT ALGORITHM

Modelling and identification of nonlinear dynamic systems is a challenging task because nonlinear processes are unique in the sense that they do not share many properties. Fuzzy local linearization (FLL) is a useful divide-and-conquer method for coping with complex problem such as data based nonlinear process modelling. In this paper first a modified local linear model tree (LOLIMOT) algorithm for nonlinear system modelling are proposed. Expectation maximization (EM) algorithm is used for local model estimation which improves the model performance and provides covariance information about the model mismatch which is essential in the consequent state estimation, data fusion or control applications. The proposed algorithm is presented, with an implementation on a land vehicle to validate the proposed method for modelling. Also we implement LOLIMOT+LS, ANFIS, MLP and NRBF algorithms on above mentioned land vehicle. Results demonstrate the optimum performance of proposed algorithm.