350 rub
Journal Neurocomputers №11 for 2011 г.
Article in number:
Research of neuro-fuzzy control system survivability of reactor installations for butanol hydrogenation
Authors:
A. I. Koldaev, L. B. Kopytkova, A. N. Lavrinenko, M. M. Pogosov, A. V. Gladkov
Abstract:
In the given work the problem of construction of a control system by reactor installation of hydrogenation butanol by means of an indistinct neural network is considered. The controler which would watch observance of norms of technological process is necessary for similar manufacture and if necessary reacted to arising fluctuations of temperature and pressure, submitting operating influences. The wide range of values of parameters of manufacture allows to use advantages of indistinct logic to control system construction. In the field of management neural networks have appeared suitable means for the decision of difficult nonlinear problems. Thanks to ability of neural networks to self-training, neuro-controller are suitable for management in the conditions of essential uncertaintyly. The algorithm of work of an indistinct regulator is formed by means of simple and clear linguistic rules. By a corresponding choice of functions of an accessory the regulator characteristic can be simplified, and the increase in number of entrance and target functions of an accessory allows to raise degree of complexity of the characteristic of a regulator. Accuracy of target result depends on quality of fine tuning of functions of an accessory. For simplification of this process it is recommended information on object to present in a kind neuro- indistinct networks. One of important qualities neuro-networks control systems is safety of work in the conditions of various failures. One of methods of increase of survivability is the method of increase in number of neurons of an indistinct network. In article the method of increase of survivability of the given system is offered at possible refusals. Functioning of indistinct neural system at doubling of number of the neurons forming functions of an accessory to entrance values has been investigated. Authors of work had been drew following conclusions. 1. After failure more than third of all communications in neuro-controller, quality of work indistinct neuro-system still remains comprehensible. Sharp deterioration comes when damaged there are the most critical communications. 2. Various neurons of the latent layers play neuro-controller different roles. The layer forming functions of an accessory has the greatest sensitivity. The layer forming rules which allows to support quality of management at comprehensible level at degradation to 50% of neurons is less sensitive.
Pages: 3-8
References
  1. Ротштейн А.П., Митюшкин Ю.И. Нейролингвистическая идентификация нелинейных зависимостей // Кибернетика и системный анализ. 2000. № 2. С. 37-44.
  2. Дьяконов В., Круглов В. Математические пакеты расширения MATLAB. СПб.: Питер. 2001.
  3. Колдаев А.И.Интеллектуальная система поддержания принятия решений при управлении технологическим процессом // «Математические методы в технике и технологиях» (ММТТ-20). Сб. трудов ХХ Междунар. науч. конф. Ярославль: ЯГТУ. 2007.
  4. Деменков Н. П.Нечеткое управление в технических системах: Учебное пособие. М.: Изд-во МГТУ им. Н.Э. Баумана. 2005.
  5. Пегат А.Нечеткое моделирование и управление. М.: БИНОМ. Лаборатория знаний. 2009.