Please use this identifier to cite or link to this item: http://ea.donntu.edu.ua:8080/jspui/handle/123456789/28048
Title: РАЗРАБОТКА СПОСОБА ПРОГНОЗА СОДЕРЖАНИЯ ВЗРЫВООПАСНЫХ ГАЗОВ В ГОРНЫХ ВЫРАБОТКАХ
Other Titles: Розробка способу прогнозу вмісту вибухонебезпечних газів в гірничих виробках.
Development of the method of prediction of content of explosive gases in mines.
Authors: Е.Е. Федоров, E.E. Fedorov
Ю.Л. Дикова, Y.L. Dikova
Keywords: прогноз
forecast
содержание взрывоопасных газов
content of explosive gases
нейронная сеть
neural network
идентификация структуры и параметров сети
selection of a network architecture
среднеквадратичная ошибка
mean square error
Issue Date: May-2015
Publisher: Донецький національний технічний університет
Citation: Наукові праці Донецького національного технічного університету. Серія: Обчислювальна техніка та автоматизація. Випуск 1 (28). - Красноармійськ, ДонНТУ, 2015.
Series/Report no.: Наукові праці Донецького національного технічного університету. Серія: Обчислювальна техніка та автоматизація. Випуск 1 (28). - Красноармійськ, ДонНТУ, 2015. - C. 97-104.;
Abstract: В статье рассмотрены и проанализированы существующие методы прогноза содержания взрывоопасных газов. Исходя из основных преимуществ и недостатков, разработан и реализован нейросетевой способ прогноза содержания взрывоопасных газов в горных выработках. В основу способа заложена нейронная сеть NARX, архитектура которой определена на основе проведенных экспериментов. Критерием выбора архитектуры было минимальное значение MSE. Для оценки эффективности предложенного способа были проведены численные исследования, которые доказывают эффективность выбранной сети и ее архитектуры.
Description: Despite the rapid development of computer systems in coal mines emergencies caused by high concentrations of explosive gases continue to occur. Therefore, the development of methods for forecasting the content of combustible gases in mines, used to improve the quality of air and gas situation assessment is urgent. To solve the problem of forecasting the article analyzed the most common methods of forecasting - regression and autoregressive methods; methods based on exponential smoothing methods based on Markov chains; methods based on the classification and regression trees; neural network forecasting methods. The main criteria for the choice of the method were: the relationship between the factors examined for the finished model; it does not require any assumptions about the distribution of the factors, a priori information about the factors may be absent; the original data can be highly correlated, be incomplete or noisy; systems can be analyzed with a high degree of nonlinearity; the rapid development of the model; high adaptability; systems can be analyzed with many factors; it does not require an exhaustive search of all possible models; it is possible to analyze systems with heterogeneous factors. On the basis of the comparative characteristics the choice was made in favor of the neural network method. We examined neural networks, which are designed to meet the challenges of the forecast. The main criteria for selecting a particular neural network were such as the presence of feedback delay in the input layer and the accuracy of the forecast. Among the considered networks the most appropriate one was NARX. To determine the selected network architecture we carried out numerical experiments. Architecture selection criterion was the minimum value of MSE. Experiments have shown that with the increase of neurons in the hidden layer MSE value decreases rapidly. According to the results of the study we chose network architecture with the number of neurons at 10. As a tool for neural network training three types of genetic algorithm and back-propagation algorithm were examined. As it was shown by the experimental results, the combined genetic algorithm that uses the combination of the search direction to research all of the search space, is a more effective tool for training the neural network. To evaluate the effectiveness of the proposed method we carried out numerical studies that proved the effectiveness of the selected network, its architecture and learning algorithm.
URI: http://ea.donntu.edu.ua/handle/123456789/28048
ISSN: 2075-4272
Appears in Collections:Випуск 1(28)

Files in This Item:
File Description SizeFormat 
Npdntu_ota_2015_1_14.pdf271,47 kBAdobe PDFView/Open


Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.