Modified Algorithm of Neural Network Control for Non-stationary Object

Filaretov, G.F.; Eliseev, V.L.

Авторы: Filaretov G.F., Eliseev V.L.
Сборник: 55th International Scientific Colloquium
Тезисы
Год издания: 2010
Место издания: Ilmenau University of Technology
Аннотация: A task of neural network control over non-stationary plant is observed. It's supposed that dynamic behaviour of the plant may change at arbitrary moment. Reference signal and noise are considered as random series. A traditional approach in such conditions is to have permanently active adoption algorithms for neural network controller and neural network for plant identification. This leads to additional expenses during periods of steady plant condition and may decrease quality of control in general. To avoid mentioned disadvantages it's proposed to activate adoption algorithms only if a meaningful change of plant behaviour is detected. It can be done by comparison of observed plant output and signal from neural network plant model . There is a well known statistical algorithm of cummulative sum to detect such change as an increase of dispersion of the identification error ( - ). The proposed approach was implemented and successfully tested in computer simulation experiment. Conditions of efficient application of designed appoarch were revealed and it was stated when it can compete with traditional approach. Questions of optimal cumulative sum and learning algorithms tuning are observed. A method of adaptive neural network controller synthesis using described idea is formulated as a result of the research.
Добавил в систему: Елисеев Владимир Леонидович

	ИСТИНА	Войти в систему Регистрация
	ИСТИНА ИНХС РАН
	Главная Поиск Статистика О проекте Помощь

ИСТИНА

ИСТИНА ИНХС РАН

Modified Algorithm of Neural Network Control for Non-stationary Objectтезисы доклада

Прикрепленные файлы