Developing the Automatic Control System Based on Neural Controller

Abstract

Mostly the dynamics of controlled objects is often described by nonlinear equalizations. Last years themethodology of neural networks is engaged into designing the systems controlling such objects, in particular due to theinfluence of nonlinearities can be taken into account by nonlinear functions of the activation. Such methodology brings someintelligence to the designed system.Authors proposed the purposeful procedure of forming the structure of the neural controller according the desired lawof the control using the discrete transformation of the motion equation. Requirements to the mathematical model of thereference and method of network training are determined, and the control quality is estimated at traditional passing thedisagreement error in the controller input and for the proposed new configuration of its input circuit, namely with separatedinputs. Simulation results confirmed providing the better quality of the system control.DOI: http://dx.doi.org/10.5755/j01.itc.44.3.7717