煤矿瓦斯场分布演化规律及其时空建模研究
发布时间：2018-08-20 12:16
【摘要】：瓦斯灾害是煤矿最严重的灾害形式，往往造成大量的人员伤亡和重大的经济损失。回采工作面巷道是主要的瓦斯涌出区域，容易发生瓦斯积聚，煤矿瓦斯事故绝大多数发生在回采工作面区域。当前，对工作面的瓦斯监测主要采用监测站点超限报警的方式，采集的数据为孤立的点数据，监测范围有限，不能有效地描述整个工作面的瓦斯分布和安全状况。如何通过合理部署瓦斯传感器节点并利用节点间的内在联系，实现对瓦斯涌出及积聚区域做出分析判断，对未来瓦斯场的运移态势进行预测预警具有重要现实意义。针对瓦斯数据的预测主要以时间序列方法为主，对瓦斯分布场的构建主要以空间信息学作为主要手段，而瓦斯的运移和分布规律与时间、空间密切相关，避开两者关系孤立地处理问题将造成重要先验知识的丢失。因此本文从充分利用时间和空间内在关系的角度出发，采用研制的无线瓦斯传感器阵列作为研究工具，通过时空建模的方式对瓦斯浓度预测、瓦斯分布场构建、瓦斯传感器布点优化三个渐进问题进行理论和试验研究，，实现对瓦斯分布场的实时重构、对瓦斯场运移态势的预测分析、对瓦斯异常行为的时空反演和对瓦斯传感器阵列的优化布置。 论文完成的主要工作有： （1）对煤矿通风系统进行分析，研究工作面巷道风流特性，根据对风流性质的数学模型的分析和工作面巷道实际参数确定巷道中的风流状态。然后对工作面瓦斯的主要运移形式进行分析，在菲克扩散定律和守恒方程的基础上，推导出瓦斯在巷道湍流风流作用下的运移扩散方程，理论分析表明瓦斯的运移和扩散与时间和空间变量有关，证明采用时空建模的方式监测瓦斯的运移行为是合理的，为工作面瓦斯浓度的时空预测和瓦斯分布场时空重构提供理论依据。利用数值分析的方法对工作面瓦斯时空规律进行分析，研制瓦斯传感器阵列设备，对瓦斯在工作面的分布进行试验测试，测试结果验证了理论分析的正确性。 （2）选取ELM作为基本学习模型并对其进行时空扩展，增加空间位置信息作为先验知识，提出一种新的时空极限学习机模型STELM。以空间关联性作为输入权重，邻接站点的时间面板数据作为时空神经元的输入，简化了时空建模的复杂性，使算法仅需要两个输入参数：空间延迟算子边界值和时间延迟算子边界值。在理论仿真数据以及现场监测数据中的应用表明，较仅依托于时间维度信息的预测方法在泛化能力上有较大的提升。在此基础上，结合选择性集成学习思想，提出了一种基于L1正则化的STELM选择性集成学习方法SERSTELM，通过L1正则化稀疏加权组合多个STELM学习机从而避开了定义和度量多样性的问题，可以直接获得选择性稀疏解，该方法在STELM的基础上进一步提升了预测精度和泛化性能。 （3）提出基于时空克里金模型的瓦斯分布场重构技术。空间信息统计方法采用区域化变量为基础，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估计，借鉴空间克里金方法，对其进行时间扩展，采用积和模型通过拟合空间半变异函数和时间半变异函数构建时空半变异函数，获得时间和空间之间的内在关系，利用交叉验证的方式对算法进行分析比较，各项评价指标均表明，时空克里金模型能够获得更好的瓦斯场重构效果。同时，结合时空克里金方法和STELM能够实现对未来瓦斯运移态势预测，为研究减少和预防瓦斯灾害事故提供了新的研究手段和思路。 （4）提出了兼顾已知监测点瓦斯浓度预测和未知监测点瓦斯场插值效果的瓦斯传感器布置方式。监测系统的性能在很大程度上取决于传感器的布置数量和位置，相同数目的传感器，不同的布置方式将产生不同的监测效果。在瓦斯传感器数量一定的情况下，为了使本文提出的时空建模方法能够在实际现场获得更好的效果，采用多目标粒子群算法进行建模并对其进行改进，引入增量比率占优作为适应度优胜策略，增加精英保留和被动更新机制，使其适用于煤矿传感器布点场景，试验结果表明其布局方式随着节点数量的减少而呈现明显规律性，寻优结果可以在两种算法中进行有效折衷。
[Abstract]:Gas disaster is the most serious form of coal mine disasters, often resulting in a large number of casualties and major economic losses. Mining face roadway is the main gas emission area, easy to occur gas accumulation, the vast majority of coal mine gas accidents occur in the mining face area. At present, gas monitoring in the working face mainly uses monitoring stations. The method of point-out alarm can not effectively describe the gas distribution and safety condition of the whole working face because the data collected are isolated point data and the monitoring range is limited. It is of great practical significance to predict and warn the migration situation of gas field.The prediction of gas data is mainly based on time series method,and the construction of gas distribution field is mainly based on spatial informatics.The migration and distribution law of gas is closely related to time and space,so it is necessary to deal with the problem in isolation from the relationship between them. Therefore, this paper makes full use of the inherent relationship between time and space, uses the wireless gas sensor array as a research tool, and carries out theoretical and Experimental Research on three progressive problems of gas concentration prediction, gas distribution field construction and gas sensor distribution optimization through space-time modeling. The research realizes the real-time reconstruction of gas distribution field, the prediction and analysis of gas migration situation, the space-time inversion of abnormal gas behavior and the optimization of gas sensor array.
The main tasks of the thesis are as follows:
(1) Analyze the ventilation system of coal mine, study the airflow characteristics of the roadway, determine the airflow state in the roadway according to the analysis of the mathematical model of the airflow properties and the actual parameters of the roadway. Then analyze the main migration forms of the gas in the working face, deduce the gas on the basis of Fick diffusion law and conservation equation. The equation of gas migration and diffusion under the action of turbulent airflow in roadway shows that gas migration and diffusion are related to time and space variables, which proves that the method of space-time modeling is reasonable to monitor gas migration behavior, and provides theoretical basis for the space-time prediction of gas concentration and the space-time reconstruction of gas distribution field in working face. The method of analysis is to analyze the space-time law of gas in working face, develop gas sensor array equipment, and test the distribution of gas in working face. The test results verify the correctness of theoretical analysis.
(2) Selecting ELM as the basic learning model and extending it to space-time, adding spatial location information as a prior knowledge, a new spatio-temporal limit learning machine model STELM is proposed. Spatial correlation is used as the input weight, and temporal panel data of adjacent stations is used as the input of spatio-temporal neurons, which simplifies the complexity of spatio-temporal modeling and makes calculation easier. The method only needs two input parameters: the spatial delay operator boundary value and the time delay operator boundary value. The application in theoretical simulation data and field monitoring data shows that the generalization ability of the prediction method based on time dimension information is improved greatly. A selective ensemble learning method for STELM based on L1 regularization, SERSTELM, combines multiple STELM learning machines by L1 regularization and sparse weighting to avoid the problem of definition and metric diversity, and obtains the selective sparse solution directly. This method further improves the prediction accuracy and generalization performance based on STELM.
(3) The reconstruction technology of gas distribution field based on spatio-temporal Kriging model is proposed. Spatial information statistical method is based on regionalized variables and variogram is used as a basic tool to realize the best unbiased estimation of the data with randomness and structural correlation. Spatial Kriging method is used for reference to expand its time and product is adopted. The spatio-temporal semi-variogram is constructed by fitting the spatio-temporal semi-variogram and the spatio-temporal semi-variogram, and the intrinsic relationship between time and space is obtained. The algorithm is analyzed and compared by cross-validation. All the evaluation indexes show that the spatio-temporal Kriging model can obtain better gas field reconstruction effect. Spatio-temporal Kriging method and STELM can predict the future gas migration situation, and provide new research means and ideas for reducing and preventing gas disaster accidents.
(4) Presents a gas sensor layout method which takes into account the prediction of gas concentration at known monitoring points and the interpolation effect of gas field at unknown monitoring points. In order to make the spatio-temporal modeling method proposed in this paper achieve better results in the actual scene, the multi-objective particle swarm optimization (MPSO) algorithm is used to model and improve it. Incremental ratio dominance is introduced as the fitness strategy to increase the elite retention and passive update mechanism, which makes it suitable for coal mine sensor distribution. In the point scenario, the experimental results show that the placement pattern is obviously regular as the number of nodes decreases, and the optimization results can be effectively compromised between the two algorithms.
【学位授予单位】：中国矿业大学
【学位级别】：博士
【学位授予年份】：2014
【分类号】：TD712
本文编号：2193581
[Abstract]:Gas disaster is the most serious form of coal mine disasters, often resulting in a large number of casualties and major economic losses. Mining face roadway is the main gas emission area, easy to occur gas accumulation, the vast majority of coal mine gas accidents occur in the mining face area. At present, gas monitoring in the working face mainly uses monitoring stations. The method of point-out alarm can not effectively describe the gas distribution and safety condition of the whole working face because the data collected are isolated point data and the monitoring range is limited. It is of great practical significance to predict and warn the migration situation of gas field.The prediction of gas data is mainly based on time series method,and the construction of gas distribution field is mainly based on spatial informatics.The migration and distribution law of gas is closely related to time and space,so it is necessary to deal with the problem in isolation from the relationship between them. Therefore, this paper makes full use of the inherent relationship between time and space, uses the wireless gas sensor array as a research tool, and carries out theoretical and Experimental Research on three progressive problems of gas concentration prediction, gas distribution field construction and gas sensor distribution optimization through space-time modeling. The research realizes the real-time reconstruction of gas distribution field, the prediction and analysis of gas migration situation, the space-time inversion of abnormal gas behavior and the optimization of gas sensor array.
The main tasks of the thesis are as follows:
(1) Analyze the ventilation system of coal mine, study the airflow characteristics of the roadway, determine the airflow state in the roadway according to the analysis of the mathematical model of the airflow properties and the actual parameters of the roadway. Then analyze the main migration forms of the gas in the working face, deduce the gas on the basis of Fick diffusion law and conservation equation. The equation of gas migration and diffusion under the action of turbulent airflow in roadway shows that gas migration and diffusion are related to time and space variables, which proves that the method of space-time modeling is reasonable to monitor gas migration behavior, and provides theoretical basis for the space-time prediction of gas concentration and the space-time reconstruction of gas distribution field in working face. The method of analysis is to analyze the space-time law of gas in working face, develop gas sensor array equipment, and test the distribution of gas in working face. The test results verify the correctness of theoretical analysis.
(2) Selecting ELM as the basic learning model and extending it to space-time, adding spatial location information as a prior knowledge, a new spatio-temporal limit learning machine model STELM is proposed. Spatial correlation is used as the input weight, and temporal panel data of adjacent stations is used as the input of spatio-temporal neurons, which simplifies the complexity of spatio-temporal modeling and makes calculation easier. The method only needs two input parameters: the spatial delay operator boundary value and the time delay operator boundary value. The application in theoretical simulation data and field monitoring data shows that the generalization ability of the prediction method based on time dimension information is improved greatly. A selective ensemble learning method for STELM based on L1 regularization, SERSTELM, combines multiple STELM learning machines by L1 regularization and sparse weighting to avoid the problem of definition and metric diversity, and obtains the selective sparse solution directly. This method further improves the prediction accuracy and generalization performance based on STELM.
(3) The reconstruction technology of gas distribution field based on spatio-temporal Kriging model is proposed. Spatial information statistical method is based on regionalized variables and variogram is used as a basic tool to realize the best unbiased estimation of the data with randomness and structural correlation. Spatial Kriging method is used for reference to expand its time and product is adopted. The spatio-temporal semi-variogram is constructed by fitting the spatio-temporal semi-variogram and the spatio-temporal semi-variogram, and the intrinsic relationship between time and space is obtained. The algorithm is analyzed and compared by cross-validation. All the evaluation indexes show that the spatio-temporal Kriging model can obtain better gas field reconstruction effect. Spatio-temporal Kriging method and STELM can predict the future gas migration situation, and provide new research means and ideas for reducing and preventing gas disaster accidents.
(4) Presents a gas sensor layout method which takes into account the prediction of gas concentration at known monitoring points and the interpolation effect of gas field at unknown monitoring points. In order to make the spatio-temporal modeling method proposed in this paper achieve better results in the actual scene, the multi-objective particle swarm optimization (MPSO) algorithm is used to model and improve it. Incremental ratio dominance is introduced as the fitness strategy to increase the elite retention and passive update mechanism, which makes it suitable for coal mine sensor distribution. In the point scenario, the experimental results show that the placement pattern is obviously regular as the number of nodes decreases, and the optimization results can be effectively compromised between the two algorithms.
【学位授予单位】：中国矿业大学
【学位级别】：博士
【学位授予年份】：2014
【分类号】：TD712
【参考文献】
相关期刊论文 前10条
1 付华;王宏云;;基于特征分析模型的瓦斯监测传感器优化布点[J];传感器与微系统;2009年11期
2 李恩良,王秉权;井巷污染物横向紊流扩散系数的研究[J];东北工学院学报;1986年01期
3 李恩良,王秉权,王振诚;井巷紊流扩散与弥散的实验研究[J];东北工学院学报;1986年02期
4 黄学满;;井下风流中瓦斯扩散模型研究[J];矿业安全与环保;2011年06期
5 李恩良;井巷紊流传质过程的纵向弥散模型及纵向弥散系数[J];阜新矿业学院学报;1989年03期
6 孙继平;唐亮;陈伟;张欢;;煤矿井下长巷道瓦斯传感器间距设计[J];辽宁工程技术大学学报(自然科学版);2009年01期
7 龚昌来;;一种基于数据融合的图像插值算法[J];光电工程;2010年02期
8 杨芳;高峰;马艺闻;赵会娟;阮平巧;;一种基于测量空间样条插值扩展的扩散光学层析图像改善方法[J];光学学报;2009年09期
9 王登婷;;Direct Numerical Simulation of Interaction Between Wave and PorousBreakwater Based on N-S Equation[J];China Ocean Engineering;2012年04期
10 陆秋琴;;采场瓦斯运移的三维LBM仿真[J];安全与环境学报;2010年04期
相关博士学位论文 前6条
1 侯少杰;煤矿瓦斯治理过程控制理论及应用研究[D];中国矿业大学;2010年
2 董海波;煤矿采场瓦斯分布与分布场重构技术研究[D];中国矿业大学;2012年
3 刘年平;煤矿安全生产风险预警研究[D];重庆大学;2012年
4 韦善阳;瓦斯异常涌出气体运移规律及影响范围研究[D];中国矿业大学（北京）;2013年
5 王泽璞;电容层析成像图像重建与气力输送两相流可视化测量研究[D];华北电力大学;2013年
6 梁双华;矿井瓦斯传感器优化选址研究[D];中国矿业大学;2013年
本文编号：2193581
本文链接：https://www.wllwen.com/kejilunwen/anquangongcheng/2193581.html
