|
ИСТИНА |
Войти в систему Регистрация |
Интеллектуальная Система Тематического Исследования НАукометрических данных |
||
Software tool for neural network control algorithm researchтезисы доклада
- Авторы: Eliseev V.L., Filaretov G.F.
- Сборник: 45th International Scientific Colloquium
- Тезисы
- Год издания: 2000
- Место издания: Ilmenau Technical University
- Аннотация: Well known capabilities of artificial neural networks for arbitrary function interpolation promise many benefits in area of automated control. Keeping in mind indirect information extraction from the training data it’s promising to use neural networks for object and noise identification and for controller synthesis on the basis of given experimental data. It’s worth while to compare traditional stochastic methods of optimal identification and control with neural network approach. It’s evident that neural network application can be useful only if its effectiveness will be not worse than traditional methods. While doing research on different approaches of neural network application in automated control system a software tool NeuCon was developed. It allows to model control loop with noise in observation output, has facilities to generate determined and stochastic set point, remembers time series of signals in different check points of the control loop, gives description of object and controller in terms of linear control theory and, especially, allows to use integrated neural network modules for solving different research tasks. Software tool has capabilities to import data from text files and files of STATISTICA package. It exports data to the same file formats plus to files of scientific chart drawing utility DPLOT. Wide capabilities of data import/export allow to use NeuCon in collaboration with other application programs and doing research with experimental data from real control systems. Some computation experiments of applied neural networks for identification and control synthesis were made using described software tool. Methods for solving such tasks were developed. Especially, the optimal controller synthesis with RMS error criteria using neural network doesn’t need knowledge of formal models for an object and a noise and doesn’t need symbolic operations over formulas. In addition the resulted controller is always physically accomplished and usable for practice (having specific hardware for implementation). The computation experiments gave data for comparison between neural network and traditional approaches of optimal control. Some limitations of neural network methods were revealed and useful advises for different classes of control systems were given.
- Добавил в систему: Елисеев Владимир Леонидович
Прикрепленные файлы
| № | Имя | Описание | Имя файла | Размер | Добавлен |
|---|---|---|---|---|---|
| 1. | Полный текст | Programma_NS-upravleniya_eng.pdf | 751,2 КБ | 1 марта 2020 [vlad-eliseev] | |
| 2. | SlideShow.pdf | SlideShow.pdf | 559,1 КБ | 1 марта 2020 [vlad-eliseev] |