煤自燃指标气体分析与分级预警

Index gas analysis and grading early warning of coal spontaneous combustion

煤自然发火严重制约矿井安全生产。为实现煤自然发火危险性精准预测预报,采用程序升温试验系统测试分析了沙曲一号煤矿不同粒径煤样各种气体产物及其浓度变化规律,进一步引入随机森林集成学习方法建立了煤自然发火危险性分级预警模型,并通过大佛寺煤矿自然发火试验进行了验证分析。结果表明:粒径越小,煤氧接触面积越大,煤氧反应越激烈,气体产物浓度越大;C_(2)H_(6)气体属于煤体赋存气体,在试验初始阶段就出现了,但C_(2)H_(4)气体在温度升高至120℃左右才出现,是煤氧化裂解的产物,可以作为沙曲一号煤矿自然发火指标气体;基于随机森林建立的煤自然发火危险性分级预警模型训练样本预测准确率达到100%,在默认参数条件下,测试样本预测准确率高达96.7%,通过自然发火试验数据验证分析得到测试集预测准确率为98.9%,变量重要度评估结果为CO和C_(2)H_(4)气体对煤自然发火危险性影响最大,这与现场实际情况吻合。随机森林用于处理煤自然发火危险性与气体产物之间的复杂非线性关系十分理想,适合于煤自然发火危险性预测预报。

Coal spontaneous combustion seriously restricts the safety production of mines.In order to accurately predict the risk of coal spontaneous combustion,a temperature-programmed experimental system was adopted to test and analyze various gas products and their concentration changes in coal samples with different particle sizes in Shaqu No.1 Coal Mine.Furthermore,a random forest ensemble learning method was introduced to establish a grading and early warning model for the risk of coal spontaneous combustion,and the verification analysis was carried out by the spontaneous combustion test in Dafosi Coal Mine.The results show that the smaller the particle size,the larger the contact area between coal and oxygen,the more intense the coal oxygen reaction,and the higher the concentration of gas products.C_(2)H_(6) belongs to the gas existing in the coal body,which appeared in the initial stage of the test.However,C_(2)H_(4) only appears when the temperature rises to around 120℃,which is a product of coal oxidation cracking and can be used as an index gas for coal spontaneous combustion in Shaqu No.1 Coal Mine.The risk grading and early warning model for coal spontaneous combustion based on random forest has achieved 100%accuracy for training samples,and under the default parameter condition,the prediction accuracy of test samples is as high as 96.7%.The prediction accuracy of the test set is 98.9%through the validation and analysis of the spontaneous combustion test data,and the evaluation results of the importance of variables show that CO and C_(2)H_(4) gases have the greatest impact on the risk of coal spontaneous combustion,which is consistent with the actual situation on site.These indicate that random forest is ideal for dealing with the complex nonlinear relationship between the risk of coal spontaneous combustion and gas products,and is suitable for predicting the risk of coal spontaneous combustion.