A Neurofuzzy-Based Adaptive Predictor for Control of Nonlinear Systems

This paper proposes an adaptive predictor for general nonlinear systems based on the use of a class of neurofuzzy models. The neurofuzzy-based predictor can be interpreted as a linear predictor network consisting of a global linear predictor and several local linear predictors with interpolation. It has some distinctive features as well as good prediction ability: its parameters have explicit meanings useful for initial value setting in parameter adjustment; it may be transformed into a form linear for the variables synthesized in control systems, which makes deriving a control law straightforward. Simulations on applying it to adaptive control of nonlinear systems demonstrate its usefulness.