Neural Networks based Adaptive Inverse Control for Nonlinear Systems

Abstract

Adaptive Inverse Control (AIC) is a very significant approach for control of unknown linear and nonlinear plants. Neural Networks (NNs) based AIC for dynamical systems has acquired much attention these days due to its compliant characteristics. The AIC problem can be used to obtain stable controllers for nonlinear plants. Foremost aspiration of the thesis is to put forward a new scheme for intended tracking of nonlinear systems. Radial Basis Function Neural Networks (RBFNNs) have the ability of mapping nonlinearities effectively and a new AIC scheme based on RBFNN is proposed. In this scheme, update of controller parameters is acquired by passing the tracking error through estimated Jacobian of the system model. The implementation of proposed scheme is simple as compared to existing techniques for AIC and to substantiate the results of the proposed scheme, it is compared with an existing neural networks based AIC technique. Simulation results of three different nonlinear systems are presented in this thesis to authenticate the proposed scheme. Primarily, the simulation of a nonlinear plant model is presented and the results are shown with and without the affect of disturbance induced in the plant. The results validate finer tracking with diminished disturbance and error convergence of the proposed scheme. Subsequently, to further corroborate the proposed scheme it is implemented on another class of nonlinear systems known as Hammerstein type systems, the presented scheme is simulated on heat exchanger plant model and binary distillation column process plant model.Finer tracking performance and error convergence are clearly discernible from the results, both in the presence of disturbance and without any disturbance in the plant. The results manifest that proposed scheme is significantly adaptive and efficient with minimal rise time, slighter overshoots, lesser settling time and is also capable to restrain the affect of disturbance in the plant. Moreover, the presented scheme is ratified by the mathematical proofs of error and parameters’ convergence, delivered in the thesis. Hence, the results affirm that proposed scheme is pertinent for control of nonlinear systems and Hammerstein type systems as well.