智慧化矿山是我国煤炭高质量开采的发展战略，由于井下生产环境复杂导致传感器 测量数据出现不完整、失真现象，无法获得准确的瓦斯运移状态信息，难以建立有效的 采空区卸压瓦斯抽采状态（抽采与上隅角瓦斯浓度）预测模型，对瓦斯抽采状态智能 识别造成影响，为煤层智能开采带来安全隐患。针对该问题，研制了大尺度瓦斯运移三 维物理模拟自动测控系统，开展了物理模拟过程控制与物理模型可靠性分析，获得了煤 层开采过程中覆岩裂隙演化、矿山压力分布、卸压瓦斯运移、瓦斯抽采等问题一体化研 究方法；提出了基于遗传算法与自适应卡尔曼滤波算法的传感器测量数据处理方法，解 决了因传感器疲劳损伤与测量环境不一致引发的测量数据不完整与失真问题；提出了时 空序列下覆岩垂直应力、气体浓度变化趋势分析方法，获得了煤层开采与瓦斯抽采状态 下瓦斯运移时空演化规律；基于机器学习算法分析并获得了覆岩裂隙演化特征与瓦斯运 移主控因素之间的内在联系，利用裂隙网络与瓦斯运移演化之间信息的互补性，建立了 多维数据下采空区卸压瓦斯抽采状态预测模型。论文主要研究工作如下：

1）设计了大尺度瓦斯运移三维物理模拟实验测控系统。该系统集煤层开采、柔性 加载、瓦斯涌出、矿井通风、瓦斯抽采过程控制与数据采集于一体，可实现大尺度（模 拟最大采深 2105 m，采高 0~12 m，工作面面长 200m，面宽 160m）物理模型密闭环境 搭建与煤层开采过程模拟，并基于山西和顺天池高瓦斯煤矿与物理模拟相似准则设计了 物理模拟实验方案。

2）基于煤层开采条件控制原理与关键参数变化特征建立物理模拟自动控制系统。 提出了基于风量环与风压环反馈回路的双闭环通风过程控制方法，并结合电磁开关建立 了 U 型、U+L 型通风方式转换控制系统。提出了以气压、气体流量监测为基础，涌出 量为反馈回路的瓦斯涌出过程控制方法，通过对涌出气体压力及余量的实时监控，保障了气体涌出过程控制的可靠性。建立了以气压环反馈回路为核心，PI 控制算法为手段的瓦斯抽采控制系统。

3）基于测量环境特征与物理模型可靠性分析搭建物理模拟测量系统。开展了传感 器、线缆、岩层组合体静应力分析，获得了传感器体积与通讯线缆设计参数。提出了 “带电安装”的安装方法，研制了低功耗多源自适应传感器状态监测系统，系统总功耗 小于 22w，实现了物理模型搭建至开展实验全过程的 7*24 小时传感器状态实时监测与 故障自动诊断，覆岩垂直应力与气体浓度等其他关键参数实时同步测量。

4）针对因物理模型搭建（相似材料装填及固化）时间过长（近一年）所引发的传 感器测量数据不完整与失真问题，开展了测量数据误差分析，提出了基于遗传算法与自 适应卡尔曼滤波算法的测量数据处理方法。获得了当覆岩垂直应力测量点数量为 11 个 时，重构覆岩垂直应力分布曲线效果最佳，最少测量点数量为 6 个，建立覆岩垂直应力 测量关键数据集并重构测量缺失数据。提出了基于气体运动等效模型的气体浓度估计方 程，获得了不精确系统与量测误差下的气体浓度测量数据处理方法。开展验证实验，获 得样本组与对照组气体浓度变化幅值与趋势保持一致。

5）提出了时空序列下覆岩裂隙与瓦斯运移演化规律分析方法。开展了模型剖切与 皮尔逊相关性实验与分析，获得了覆岩垂直应力变化与覆岩裂隙演化趋势正相关，并基 于时空序列下覆岩垂直应力与气体浓度变化趋势数据集，获得了采动状态下瓦斯运移时 空演化规律。

6）基于机器学习算法建立多维数据下气体运动状态预测模型。建立了内含气体运 移主控因素与覆岩裂隙演化信息的采空区覆岩应力与气体浓度估计模型，挖掘多维数据 的内在联系。建立了多维数据下上隅角与抽采气体浓度预测模型，模型预测值与真实值 之间均方根误差（RMSE）为 0.06。为开展采空区卸压瓦斯抽采状态预测、智能评价等 问题的研究提供一种新的方法。 以上研究成果通过物理模拟原型山西和顺天池煤矿与陕西建新煤矿采空区卸压瓦斯 抽采工程实践验证进行了验证。为开展瓦斯运移、覆岩裂隙演化等问题的研究提供可 靠、精确的大尺度三维物理模拟实验平台，提出了一种新的采空区卸压瓦斯抽采状态 预测模型，获得了更加全面的瓦斯抽采状态信息，为煤炭智能开采提供了安全保障。

Intelligent mining is the development strategy of China's high quality coal mining. However, due to complex underground production environment, sensor measurement data are prone to incomplete and distorted phenomenon, which makes it difficult to establish the prediction model of accurate goaf pressure relief gas extraction indicator (drainage and upper corner gas concentration), and has a certain influence on the gas extraction effect recognition, bringing some hidden troubles for safety mining of coal seam. In view of the problem, this paper developed an automatic measurement and control system for three-dimensional physical simulation of large-scale gas migration, focusing on the parameter control and physical model reliability analysis amid physical simulation process, obtained an integrated research method of overburden fracture evolution, mine pressure distribution, pressure relief gas migration and gas extraction in the process of coal seam mining; proposed a sensor measurement data processing method based on genetic algorithm and adaptive Kalman filter algorithm, which eliminates incomplete and distorted measurement data caused by the change of measurement environment; raised the information analysis method of vertical stress, gas concentration distribution and change trend of overburden under spatial-temporal sequence, solved the problem of sparse sensor distribution, and acquired the spatial-temporal evolution law of gas migration under coal seam mining and gas extraction. In addition, based on machine learning algorithm, the internal relationship between the evolution of overburden fractures and the main control factors of gas migration is analyzed and obtained, and by using the information complementarity between fracture network and gas migration and evolution, theindex prediction model of pressure relief gas drainage in goaf under multi-dimensional data is established. The main research work of this paper is as follows: 1) A three-dimensional physical simulation measurement and control system for large- scale gas migration is designed. The system integrates process control and data collection amid coal seam mining, flexible loading, gas emission, mine ventilation, and gas extraction as a whole, which achieves large-scale (the maximum simulated mining depth of 2105 m, mining height of 0~12 m, working surface length of 200m, surface width of 160m) closed environment construction and coal seam mining process simulation of physical model, and designs a physical simulation experiment scheme based on Shanxi and Shuntianchi high gas coal mines and physical simulation similarity criterion. 2) An automatic control system for physical simulation is established based on the control principle of coal seam mining conditions and the variation characteristics of key parameters. A double closed-loop ventilation process control method pursuant to air volume loop and air pressure loop feedback circuits is proposed, and a U type and U+L type ventilation mode conversion control system is established combined with electromagnetic switch. A gas emission process control method based on air pressure and gas flow monitoring is proposed, which takes emission as feedback circuit, realizes real-time monitoring of gas emission pressure and allowance1, and guarantees the stability of gas emission process control. Meanwhile, a gas extraction control system based on pneumatic loop feedback circuit and PI control algorithm is built. 3) A physical simulation measurement system pursuant to measurement environment characteristics and physical model reliability analysis is constructed. The static stress analysis of the complex integrating sensor, cable and rock stratum is carried out, and the volume of sensor and design parameters of communication cable are obtained. the arrangement of "live mounted" is proposed, and a low-power multi-source adaptive sensor condition monitoring system is developed. The monitoring system has the total power consumption less than 22w, and realizes 7*24 hours real-time monitoring and automatic fault diagnosis of the sensor state from building the physical model to carrying out the experiment, and real-time synchronous acquisition of other key parameters such as vertical stress and gas concentration of overburden. 4) In view of incomplete and distorted sensor measurement data caused by the long time (nearly one year) of building physical model (similar material loading and curing), a measurement data processing method based on genetic algorithm and adaptive Kalman filter algorithm is raised. By solving the position information of key monitoring points of overburden vertical stress, it is summarized that when the number of sensor monitoring points is 11, thereconstructed vertical stress distribution curve has the best effect; when the number of minimum sensor monitoring points reaches 6. Besides, the gas concentration estimation equation is obtained based on the gas motion equivalent model, and the gas concentration measurement data processing method under the imprecise system and measurement error is obtained. According to small-scale physical simulation verification experiment, the gas concentration change amplitude of the sample group and the control group is consistent. 5) An analysis method of overburden fractures and gas migration evolution under spatiotemporal sequence is put forward. Following the model cutting and Pearson correlation experiment and analysis, it is concluded that the vertical stress change of overburden is positively correlated with the evolution trend of overburden fractures. Then, the method for analyzing the spatio-temporal evolution of gas migration under spatio-temporal sequence is proposed, and the temporal and spatial evolution law of gas migration under coal seam mining and gas extraction is obtained by combining with the evolution state information of overburden fractures. 6) A prediction model of gas motion state under multidimensional data based on machine learning algorithm is constructed. An estimation model of overburden stress and gas concentration in goaf containing the main control factors of gas migration and the evolution information of overburden fractures is established to explore the internal relationship of multi- dimensional data. The prediction model aiming at extraction and gas concentration in upper corner under multi-dimensional data is built, and the root mean square error between the predicted value of it and the real value is 0.06. To sum up, this paper developed a new method for the research of index prediction and intelligent evaluation of gas emission in goaf. The above research results are proved by the practical verification of the physical simulation prototype of Shanxi Heshun Tianchi Coal Mine and Shaanxi Jianxin Coal Mine goaf pressure relief gas extraction project. To this end, it can be concluded that this paper provides a reliable and accurate large-scale three-dimensional physical simulation experiment platform for the study of gas migration and overburden fracture evolution, offers a new prediction method for pressure relief gas extraction index, and obtains more comprehensive gas extraction state information, which guarantees the intelligent coal mining.