ISSN 1817-2172, рег. Эл. № ФС77-39410, ВАК

Differential Equations and Control Processes
(Differencialnie Uravnenia i Protsesy Upravlenia)

Algorithms for Compensation Controls Realized On Artificial Neural Networks

Author(s):

A. A. Pervozvanski

Russia, 195251, St.-Petersburg, Polytechnicheskaja st. 29,
St.-Petersburg
State Technical University,
Department of Mechanic and control proceses,

control1@citadel.stu.neva.ru

T. V. Varjadchenko

Russia, 195251, St.-Petersburg, Polytechnicheskaja st. 29,
St.-Petersburg
State Technical University,
Department of Mechanic and control proceses,

control1@citadel.stu.neva.ru

Abstract:

For a compensation of nonlinear interactions between subsystems in a complex control system, a neuroalgorithm is suggested. This algorithm is based on the local approximation method for a control function of phase variables. We consider the case of unknown dynamics where the values of "observations" used in the local estimates cannot be calculated a priori. A simple algorithm of the coefficient adjustment is suggested. As a basic example, control problems for simple robots are considered. Numerical experiments demonstrate that the performance index characterizing a difference between the "ideal control" and the ANN control can be diminished in many times for not too large number of elements.

Full text (pdf)