Embedded Fuzzy-Wavelet Neural Networks Identification Learning Machine Integrated to Model Predictive Control Algorithm of Highly Uncertain Mechanical Systems

In this paper, a method for identification and control of nonlinear uncertain switched dynamical systems with average dwell-time constraints of switching signal has been proposed. An input-output based fuzzy wavelet neural network (FWNN) structure is used for addressing the identification problem. The combination of wavelet core and fuzzy antecedents allow the FWNN network toeffectively deal with system nonlinearity and modelling uncertainty in various operating conditions. Considering convergent characteristics of the identification scheme, FWNN is used for constructing a model predictive control (FWNN-MPC) system whose closed-loop stability and optimality is verified over the prediction horizon. Subsequently, the novel contribution of the paper is showing that the FWNN-MPC scheme is stable if the switching signal satisfies the necessary conditions of average dwell time. No additional constraints is exerted on the switching signal, i.e. it is considered to be arbitrarily generated within the permitted switchingspeed. Finally, numerical examples are used to demonstrate the effectiveness and accuracy of the proposed control scheme considering a nonlinear dynamical system with time and sampling delay.

Keywords:

Uncertain Switched Systems , Average Dwell Time , Model Predictive Control , Fuzzy Wavelet Neural Network , On-Line Identification.