煤矿采场瓦斯分布与分布场重构技术研究.pdf

博士学位论文 煤矿采场瓦斯分布与分布场重构技术研究 Study on Gas Distribution and Its Distribution Field Reconstruction Technology in Mine Stope 作 者董海波 导 师童敏明 教授 中国矿业大学 二○一二年五月 学位论文使用授权声明 学位论文使用授权声明 本人完全了解中国矿业大学有关保留、使用学位论文的规定，同意本人所 撰写的学位论文的使用授权按照学校的管理规定处理 作为申请学位的条件之一，学位论文著作权拥有者须授权所在学校拥有学 位论文的部分使用权，即①学校档案馆和图书馆有权保留学位论文的纸质版 和电子版，可以使用影印、缩印或扫描等复制手段保存和汇编学位论文；②为 教学和科研目的，学校档案馆和图书馆可以将公开的学位论文作为资料在档案 馆、图书馆等场所或在校园网上供校内师生阅读、浏览。另外，根据有关法规， 同意中国国家图书馆保存研究生学位论文。 （保密的学位论文在解密后适用本授权书） 。 作者签名 导师签名 年 月 日 年 月 日 中图分类号 TP23 学校代码 10290 UDC 681.5 密 级 公开 中国矿业大学 博士学位论文 煤矿采场瓦斯分布与分布场重构技术研究 Study on Gas Distribution and Its Distribution Field Reconstruction Technology in Mine Stope 作 者 董海波 导 师 童敏明 申请学位 工学博士 培养单位 信电学院 学科专业检测技术与自动化装置 研究方向 数据处理 答辩委员会主席 刘明 评 阅 人 二○一二年五月 论文审阅认定书 论文审阅认定书 研究生 董海波 在规定的学习年限内，按照研究生培养方案的 要求，完成了研究生课程的学习，成绩合格；在我的指导下完成本 学位论文,经审阅，论文中的观点、数据、表述和结构为我所认同， 论文撰写格式符合学校的相关规定，同意将本论文作为学位申请论 文送专家评审。 导师签字 年 月 日 致谢 致谢 首先向导师童敏明教授致以最诚挚的敬意和感谢五载时光， 在老师身边耳 濡目染，您严谨治学的工作作风、勤勉忘我的工作态度和不断创新的研究思想 使我受益匪浅。您从论文的选题、课题的研究进展直至论文的编写到完成都给 予了耐心充分的指导和支持，对我的教学工作也予以了很大的帮助，或许万千 话语都无法表达内心的谢意，只有在此向童老师真诚道一声老师，谢谢您 读博期间，唐守锋博士、蔡丽博士、王莹博士及硕士刘彬、刘鹏、梁良、 王晶晶、邓玉娇、牛洁茹等给予了很多帮助，谢谢你们也感谢 208 研究室各 位师弟师妹们，和你们在一起的交流，使我开拓了思路，有了研究的创新思想。 感谢学院领导和同事的支持和帮助。在职攻读博士学位困难很多，正是有 了你们的大力帮助，使我顺利完成了论文。 感谢我的爱人与家人，给予我最坚定的支持和默默付出。 感谢审阅本文的各位专家和参与答辩的各位老师的辛勤付出，谢谢您们 谨以默默的祝福和深深的谢意献给支持和帮助我的人们 董海波 二零一二年四月于文昌 I 摘 要 摘 要 瓦斯事故是煤矿最严重的事故形式，往往造成大量的人员伤亡和重大的经 济损失。煤矿瓦斯爆炸事故中，绝大多数发生在回采掘进区域。回采掘进区域 巷道是主要的瓦斯涌出区域，容易发生瓦斯积聚。对采掘巷道风流内的瓦斯分 布主要采用计算机数值模拟，能够认识瓦斯的分布规律，优化通风设计，但无 法根据现场瓦斯浓度对瓦斯涌出及积聚区域做出判断及预测，不能根据煤矿实 测瓦斯浓度数据对采场的瓦斯分布做出有效分析。利用现场瓦斯浓度监测数据 进行瓦斯分布场的重构，可以实现对采场瓦斯分布的定量化认识，实现快速确 定瓦斯涌出与积聚区域。因此，本文主要就采场瓦斯分布规律进行理论和试验 分析、瓦斯分布场重构技术进行研究。 论文完成的主要工作有 （1）建立回采工作面巷道中通风物理模型和风流紊流流动的数学模型。根 据现有湍流理论，基于 CFD 分析软件计算平台，对井下巷道内风流运移过程进 行了数值模拟分析，掌握了采场巷道内的风流流场和瓦斯分布的一般规律，为 重构瓦斯分布场提供了理论依据。研究显示采煤机机头截割部和上隅角是容易 发生瓦斯积聚区域，应加强该区域的瓦斯浓度监测。 （2）对采掘工作面瓦斯监测的基本要求做了介绍，阐述了采场瓦斯监测网 络的实现，提出通过改进单元法对采场瓦斯浓度的分布进行测定，对测量获得 的瓦斯浓度进行了趋势面拟合和数学插值分析。通过对现场瓦斯监测数据的分 析研究，结果显示采场瓦斯分布的现实规律与理论仿真分析一致，为重构瓦斯 分布场提供了现实依据。 （3）提出基于空间信息统计方法的采场瓦斯分布场的重构技术。传统插值 方法由于自身的缺点难以实现稀疏数据下的瓦斯分布场的重构。空间信息统计 方法以区域化变量理论为基础，以变异函数为基本工具，对研究具有随机性和 结构相关性的数据可以实现最佳无偏内插估计。论文通过对测量数据的实验变 异函数计算及对采用多种模型拟合结果的交叉验证比较分析，确定采用幂函数 模型拟合进行克里金插值的瓦斯分布场重构方法。 （4）提出基于神经网络算法的瓦斯分布场重构技术。神经网络插值具有良 好的非线性逼近能力，介绍了神经网络的基本理论，通过对实测数据分别利用 BF 网络算法和 RBF 网络算法进行训练，结果显示，广义回归神经网络方法重 构的瓦斯分布场具有最好的重构特性，重构瓦斯分布场光滑连续性好，且具有 理想的拟合精度。 （5）完成了对上述两种重构技术的实践和数据适应性的比较分析。两种瓦 II 斯重构方法均具有扎实的理论基础和应用案例，因此通过对一组采场实测的瓦 斯浓度监测数据进行了重构比较分析，并通过逐渐减少参加重构的样本数据的 方式，检验两种重构方法的数据适应性及预测性能。结果显示基于神经网络算 法的重建技术需要较多的样本数据来保证重建精度，而基于空间信息统计方法 的重建技术则在样本数量变化较大的情况下具有相对较好的重建精度，重建效 果稳定性好。 该论文有图 63 幅，表 13 个，参考文献 169 篇。 关键词关键词瓦斯分布；场重构；空间信息统计学；神经网络；克里金插值 III Abstract Gas accident is the most serious accidents in the coal mine which causes a large number of casualties and economic losses. Most of gas explosion accidents occur in the mining and tunneling area in mine. Mining and tunneling tunnel is the main gas emission area, where it is easy to come up with the gas accumulation. Using computer numerical simulation of the gas distribution of airflow within the mining tunnel, which is able to know the regularity of gas distribution and optimize the design of ventilation. But this can not achieve the quantification grasp and prediction of gas emission and gas accumulation, and also can not make an effective analysis of gas distribution in the stope with the measured data of the gas concentration. The reconstruction of the gas distribution field use of gas concentration monitoring data for, can realize the quantitative understanding for the stope gas distribution, and can realize the fast determine of the gas emission and accumulation area. So this dissertation mainly researches that the theory and test analysis of stope gas distribution rules, gas distribution field reconstruction technique. Main works in this dissertation are as follows 1 To establish the ventilation physical model and airflow turbulence mathematical model of the mining face tunnel. Under the existing turbulence theory, the numerical simulation analysis of air migration process in the underground tunnel is made. The general rules of stope tunnel airflow field and gas distribution is mastered to provide a theoretical basis for the reconstruction of gas distribution field. It is shown that the areas of the cutting unit of the shearer head and the upper corner are most prone to gas accumulation, so the gas concentration monitoring in these areas must be reinforced. 2 Basic requirements of the mining face gas monitoring are introduced, and to achieve the stope gas monitoring network is expounded. Improving element is proposed to measure the distribution of the stope gas concentration, and the trend surface fitting and mathematical interpolation analysis are completed. Through the research of onsite gas monitoring data, the result shows that the reality rules of the stope gas distribution is in accord with the results of the theoretical simulation analysis, and to provide a realistic basis for the reconstruction of gas distribution field. IV 3 A reconstruction technique of stope gas distributed field based on the spatial ination statistics is presented. Because of its own shortcomings of traditional interpolation s, they are difficult to achieve the reconstruction of gas distribution under the sparse data. Spatial ination statistics s based on the regionalized variable theory and the variogram tool, it can realize the best non-biased interpolation estimation to study on the randomness and structure data. Through the experimental variogram calculation of the measurement data and comparative analysis with the cross-validation of a variety of model fitting results, the power function model fitting is determined to be used for Kriging interpolation of gas distribution field reconstruction s. 4 Proposed a reconstruction techniques based on the neural network algorithm for gas distribution field. The neural network interpolation has good nonlinear approximation capability. The basic theory of the neural networks is introduced. Through the neural network training to the measured data with the BF network algorithm and the RBF network algorithm, the results show that the generalized regression neural network algorithm has the best perance for reconstruction of gas distribution field. The reconstructed gas distribution field has smooth continuity and has best the fitting precision. 5 Completed a comparative analysis of their practice perance and data adaptation with these two reconstruction s. Both of the gas reconstruction s have good theoretical basis and application cases. In dissertation，through the reconstruction practice and comparative analysis of a group of measured gas concentration monitoring data, and by gradually reducing the reconstruction sample data, their data adaptive and predictive perance of two reconstruction s were tested. The results show that the reconstruction technique based on the neural network algorithm requires more sample data to ensure the reconstruction accuracy, the reconstruction technique based on spatial ination statistics has good reconstruction accuracy and stability under in the case of large changes of the sample quantity. This dissertation contains 63 figures, 13 tables, and 169 references. Keywords gas distribution; field reconstruction; spatial ination statistics; neural network; Kriging V Extended Abstract Gas accidents is the main accidents in the coal mine, and a large number of casualties and economic losses will occur when accidents happens.To reinforce the research of gas distribution in the mine gas emission region is the key to prevent gas accident happen. Most of gas explosion accidents occur in the mining and tunneling area in mine. Mining and tunneling tunnel is the main gas emission area, where it is easy to come up with the gas accumulation. Using computer numerical simulation of the gas distribution of airflow within the mining tunnel, which is able to know the regularity of gas distribution. But this can not achieve the quantification grasp and prediction of gas emission and gas accumulation, and also can not make an effective analysis of gas distribution in the stope with the measured data of the gas concentration. This dissertation takes to determine the coal mine gas emission and accumulation of regional as research background, based on the measured data obtained by the mine gas monitoring system, to achieve the reconstruction technology of gas distribution field in the stope as the core of the research. The dissertation used the computer numerical simulation to analyze the stope tunnel airflow situation, mastered the general rules of gas accumulation in the stope. At the same time, a full analysis of the gas distribution characteristics of the stope is done with the measured data of gas concentration in the onsite. On this basis, the Kriging reconstruction technique of the gas distribution field using the spatial ination statistics-based theory is presented, which is the first time that the spatial ination statistics is used in the stope gas distribution study. Because of good nonlinear approximation ability of neural network technology, the dissertation researches the gas distribution field reconstruction based on the neural network algorithm. Finally, a comparative analysis of the practice test and date adaptive is made between the two kinds of reconstruction s on the base of spatial ination statistics and the neural network algorithm. Main contributions and innovations in this dissertation are as follows 1 To establish the ventilation physical model and airflow turbulence mathematical model of the mining face tunnel. Under the existing turbulence theory, the numerical simulation analysis of air migration process in the underground tunnel is made. The general rules of stope tunnel airflow field and gas distribution is mastered to provide a theoretical basis for the reconstruction of gas distribution field. VI It is shown that the areas of the cutting unit of the shearer head and the upper corner are most prone to gas accumulation, so the gas concentration monitoring in these areas must be reinforced. 2 Basic requirements of the mining face gas monitoring are introduced, and to achieve the stope gas monitoring network is expounded. Improving element is proposed to measure the distribution of the stope gas concentration, and the trend surface fitting and mathematical interpolation analysis are completed. Through the research of onsite gas monitoring data, the result shows that the reality rules of the stope gas distribution is in accord with the results of the theoretical simulation analysis, and to provide a realistic basis for the reconstruction of gas distribution field. 3 A reconstruction technique of stope gas distributed field based on the spatial ination statistics is presented. Because of its own shortcomings of traditional interpolation s, they are difficult to achieve the reconstruction of gas distribution under the sparse data. Spatial ination statistics s based on the regionalized variable theory and the variogram tool, it can realize the best non-biased interpolation estimation to study on the randomness and structure data. Dissertation introduces the basic rationale of the spatial ination statistics, and conducts the variogram analysis and function fitting of stope gas distribution. Through the experimental variogram calculation of the measurement data and comparative analysis with the cross-validation of a variety of model fitting results, the power function model fitting is determined to be used for Kriging interpolation of gas distribution field reconstruction s. 4 Proposed a reconstruction techniques based on the neural network algorithm for gas distribution field. The neural network interpolation has good nonlinear approximation capability. The basic theory of the neural networks is introduced. Through the neural network training to the measured data with the BF network algorithm and the RBF network algorithm, the results show that the generalized regression neural network algorithm has the best perance for reconstruction of gas distribution field and its contour. The reconstructed gas distribution field has smooth continuity and has best the fitting precision. 5 Completed a comparative analysis of their practice perance and data adaptation with these two reconstruction s. Both of the gas reconstruction s have good theoretical basis and application cases. In dissertation，through VII the reconstruction practice and comparative analysis of a group of measured gas concentration monitoring data, and by gradually reducing the reconstruction sample data, their data adaptive and predictive perance of two reconstruction s were tested. The results show that the reconstruction technique based on the neural network algorithm requires more sample data to ensure the reconstruction accuracy, the reconstruction technique based on spatial ination statistics has good reconstruction accuracy and stability under in the case of large changes of the sample quantity。 The results also show, the gas concentration monitoring of the outlet area of the coal face should be strengthened to obtain a good reconstruction results, and stope gas monitoring data should not less than 10. Keywords gas distribution; field reconstruction; spatial ination statistics; neural network; Kriging VIII IX 目 录 目 录 摘摘 要要 I 目目 录录 IX 图清单图清单XIII 表清单表清单XVII 变量注释表变量注释表 XVIII 1 绪论绪论1 1.1 课题背景及选题意义1 1.2 国内外研究现状