基于LVQ神经网络的煤自然发火预报系统研究

发布时间：2018-05-02 15:39

本文选题：煤矿安全 + 指标气体　；参考：《华北理工大学》2017年硕士论文

【摘要】：煤自然发火是导致煤矿火灾的重要原因之一,不仅制约了煤炭经济的发展,也为煤矿正常生产埋下了极大的安全隐患。做好煤炭自然发火早期预测预报至关重要。在研究了煤炭自然发火机理作用,以及煤炭自然发火预测领域的现状后,提出了一套基于LVQ神经网络的煤层自然发火预报系统。LVQ神经网络的煤自然发火预报系统大体可分为如下几个环节:第一,采取束管集气系统将矿井下检测到的气体传送于色谱仪,第二,应用色谱仪对采集到的气体进行分析,得到各检测点的指标气体浓度数据,第三,经过色谱工作站对指标气体浓度数据进行强化处理后,将数据以Excel文件的形式通过局域网,结合数据文件同步传输软件,实时自动传输于训练好的LVQ神经网络中。第四,通过网络对指标气体浓度数据的分析,评判该监测点煤层是否自然发火,与此同时结合温度传感器测得的温度值,对监测点煤层是否自然发火做出综合评判。基于唐山东欢坨煤样氧化实验数据,通过矩阵实验室(MATLAB)创建LVQ网络并进行仿真分析。选取CH4/CO、O2/CO2、CO/O2三组标志气体浓度比来作为网络输入,选取标志气体浓度比作为网络输入可以最大限度的降低矿井通风散热对标志气体浓度的影响,从而能在一定范围内提高煤自然发火预报准确度。未发火与发火作为网络的输出分别用1和2来表示。通过仿真分析采用LVQ神经网络比BP网络模型在处理束管集气装置从煤层中采集的指标气体浓度并作出自然发火预测预报中,更具快速性、稳定性及准确性。当向网络输入训练样本以外的其他指标气体浓度数据时,网络会迅速对煤层自然发火与否做出准确预报。采用LVQ神经网络模型能够有效克服指标气体浓度数据难以采取复杂数学建模的问题,可以有力解决束管系统易受通风条件影响的缺陷,明显提高了煤自然发火预报的正确率,为煤层的自然发火的预报提供了新的有力依据。
[Abstract]:The spontaneous combustion of coal is one of the important reasons leading to the coal mine fire, which not only restricts the development of coal economy, but also buries a great hidden danger of safety for the normal production of coal mine. It is very important to make early prediction and forecast of coal spontaneous combustion. After studying the mechanism of coal spontaneous combustion and the present situation in the field of coal spontaneous combustion prediction, In this paper, a coal spontaneous combustion prediction system based on LVQ neural network is put forward. The system can be divided into the following steps: first, The beam tube gas collecting system is adopted to transmit the gas detected under the mine to the chromatograph. Secondly, the collected gas is analyzed by using the chromatograph, and the index gas concentration data of each detection point are obtained. Third, After the chromatographic workstation strengthened the index gas concentration data, the data was transferred automatically and real-time to the trained LVQ neural network in the form of Excel file through LAN, combined with the data file synchronous transfer software. Fourth, through the network analysis of the index gas concentration data, the coal seam of the monitoring point is judged whether the spontaneous combustion, at the same time, combined with the temperature sensor measured temperature value, the monitoring point coal seam spontaneous combustion or not to make a comprehensive evaluation. Based on the experimental data of Donghuantuo coal sample oxidation in Tangshan, LVQ network was created and simulated by matrix lab. Selecting three groups of marked gas concentration ratios Ch _ 4 / COO _ 2 / CO _ 2 / O _ 2 as the network input, selecting the mark gas concentration ratio as the network input can minimize the influence of mine ventilation and heat dissipation on the mark gas concentration. Therefore, the prediction accuracy of coal spontaneous combustion can be improved within a certain range. Unfired and unfired are represented as network outputs by 1 and 2, respectively. Through simulation analysis, LVQ neural network is more rapid, stable and accurate than BP neural network model in dealing with index gas concentration collected from coal seam by beam tube collecting device and making prediction of spontaneous combustion. When the data of gas concentration other than training samples are inputted into the network, the network will make an accurate prediction of the spontaneous combustion of coal seams. The LVQ neural network model can effectively overcome the complex mathematical modeling problem of the index gas concentration data, can solve the defects of the beam tube system easily affected by the ventilation conditions, and obviously improves the correct rate of coal spontaneous combustion prediction. It provides a new powerful basis for the prediction of spontaneous combustion of coal seam.
【学位授予单位】：华北理工大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TD752.2;TP183

【参考文献】