我国是世界煤炭生产大国，煤炭工业为我国经济快速发展提供了重要保障。然而，矿井作业安全问题始终是影响煤炭工业发展的一大隐患。在众多煤矿安全事故中，顶板事故位居各类煤矿事故危害之首。因此，将大数据及人工智能技术应用于智慧矿山建设，对于解决实际安全生产问题具有非常重要的现实意义。如何利用大数据及人工智能技术进行采煤工作面矿压预测以保障煤矿安全、经济、高效开采，是当今针对煤矿安全问题研究的重中之重。然而，相比于传统的数值型建模方式以追求模型精确度作为目标，通过数据建模挖掘出数据内在信息关联，从而制定出更符合人类思维认知的策略，是目前数据挖掘与大数据分析领域面临的极具挑战性的难题。本文旨在将粒计算引入传统数据分析模型，建立一种新型的粒度矿压预测模型。主要研究内容如下：

(1)矿压序列的粒度子集选择

在实际矿压数据分析中，当数据集规模较大时，会导致建模过程的计算复杂度呈指数级增长。同时，数据集中某些元素对于表征数据集整体特征的作用较小。因此，在进行系统建模时，仅使用能够代表数据特征的数据子集进行建模，可以有效降低计算复杂度。此外，为了弥补由部分数据进行建模造成的系统性能遗失，通过将粒计算引入数据建模与分析中，可以在一定程度上保证建模的准确性。本文通过设计一种基于优化子集的粒度数据建模方法，从粒计算视角对数据进行分析。实验结果表明，该方法可以有效降低模型计算复杂度，并且获得具有较高精确度的数据分析模型。

(2)基于矿压序列的时空信息粒构建

由综采工作面支架监测系统所得到的矿压序列数据不仅在时间上具有关联关系，同时，由于受煤层结构空间动态变化影响，其在空间上也具有关联性。本文充分考虑矿压数据的时空关联特性，首先采用最大信息系数法获得紧密关联支架，然后通过卷积神经网络结合双向长短期记忆神经网络提取时空特征，并结合注意力机制以提升模型性能，从而获得具有时空特征的序列。最终，通过构造时空信息粒，建立一种新的矿压数据描述与表征方法。

(3)基于粒计算的矿压序列预测方法

为进一步提高矿压预测性能，本文提出一种新型的粒度矿压序列预测模型。该模型通过建立不同的粒度优化配置策略，以一系列具有语义特征的区间值信息粒作为输出结果。通过将信息粒作为基本分析单元，可以进一步解决矿山顶板压力的动态预测问题。实验证明，本文方法能够有效预测合理的矿压置信区间，比传统数值型预测可靠度更高。

China is a major coal producer in the world, and the coal industry has provided an important guarantee for the rapid development of China’s economy. However, the safety issues of mine operations are always major hidden dangers for the development of the coal industry. Among a variety of coal mine accidents, the roof accident ranks the first place. Therefore, the applications of big data and artificial intelligence technology in the construction of intelligent mine show very important practical significance for solving practical safety production problems. How to use big data and artificial intelligence technology to predict the coal mining face pressure to ensure the safety, economy, and efficient mining of coal mines is the top priority of current research. However, compared with traditional numerical modeling methods which aim to pursue the modeling accuracy, in big data era, how to mine the inherent information association through data modeling, so as to develop a strategy of supporting decision making which is more conforming to human thinking and cognition, becomes a challenging problem in data mining and data analytics. The main purpose of this thesis is to introduce Granular Computing (GrC) to traditional data analysis modeling, thus to establish a novel granular ore pressure prediction model. The thesis includes the following main research contents:

(1) Granular subset selection based on the ore pressure sequences

In practical data analysis area, a commonly encountered situation can be described as follows: due to the large scale of data size, the computational complexity of data modeling increases in an index form. In the meanwhile, some data elements in the data set have less impact on the integral characteristics of the data set. Therefore, to effectively reduce the computational complexity in system modeling, it is considered to use some subsets of data to represent the main features of the entire data set. In addition, in order to make up for the performance loss which is caused by system modeling based on the subsets of data, the modeling accuracy can be ensured to a certain extent by introducing granular computing into data modeling and analysis. This thesis designs a granular model for data analyzes based on the optimized subsets of data, which can help with the data analytics from a new perspective of granular computing. The experimental results indicate that the proposed method can effectively reduce the computational complexity and improve the accuracy of system modeling.

(2) Construction of spatio-temporal information granules based on the ore pressure sequences 

The ore pressure sequence obtained by the integrated working surface stent monitoring system are not only with time relationships, but also with spatial relationships because of the effection of the spatial dynamic changes of the coal seam structure. In this thesis, the spatio-temporal association characteristics of the ore pressure data are fully considered. Firstly, the tightly associated bracket is obtained by using the maximum information coefficient method. Then, the spatio-temporal characteristics are extracted by combining the convolutional neural network with the bidirectional long-short-term memory (BiLSTM), and the attention mechanism is involved to improve the entire performance of the model, so as to obtain the sequence with spatio-temporal features. Finally, the description and characterization of a new ore pressure data is established by constructing the spatio-temporal information granules.

(3) The prediction of ore pressure sequence based on granular computing

In order to further improve the prediction performance of ore pressure, this thesis proposes a new framework of granular ore sequence prediction model. By designing different granular optimization allocation strategy, a collection of interval-valued information granules with semantic characteristics performs as the outputs of system modeling. By taking the information granules as the basic units of data analytics, the problem of dynamic prediction of mine roof pressure can be further solved. As the experimental results shown, the ground pressure prediction based on granular computing can effectively predict a collections of reasonable confidence intervals, which have higher reliabilities than traditional parameter numerical prediction models.

[1]肖新建. 改革开放40年煤炭支撑我国经济社会高速发展[J]. 中国能源, 2018, 40(11): 10-16.

[2]张维宸. 能源安全视角下煤炭大国开采政策对比[J]. 国土资源情报, 2021(03): 16-27.

[3]牛超, 施龙青, 肖乐乐, 翟培合, 吴树磊, 赵燕红. 2001-2013年煤矿生产事故分类研究[J]. 煤矿安全, 2015, 46(03): 208-211.

[4]Maiti J, Khanzode V. Development of a relative risk model for roof and side fall fatal accidents in underground coal mines in India[J]. Safety Science，2009，47(8):1068-1076.

[5]宁小亮. 2013—2018年全国煤矿事故规律分析及对策研究[J]. 工矿自动化, 2020, 46(07): 34-41.

[6]康红普, 张镇, 黄志增. 我国煤矿顶板灾害的特点及防控技术[J]. 煤矿安全, 2020, 51(10): 24-33+38.

[7]孙超. 煤炭开采中存在的问题及其对策探讨[J]. 能源与节能, 2021(07): 151-152.

[8]王海波, 贺耀宜. 深耕煤矿安全生产领域 助力煤矿智能化建设——中煤科工集团常州研究院服务煤矿智能化建设之路[J]. 智能矿山, 2022, 3(02): 2-10.

[9]侯刚. 煤矿智能化开采系统功能及典型模式分析研究[J]. 中国煤炭, 2022, 48(02): 55-60.

[10]尹希文, 徐刚, 刘前进, 卢振龙, 于秋鸽, 张震. 基于支架载荷的矿压双周期分析预测方法[J]. 煤炭学报, 2021, 46(10): 3116-3126.

[11]巩师鑫, 任怀伟, 杜毅博, 赵国瑞, 文治国, 周杰. 基于MRDA-FLPEM集成算法的综采工作面矿压迁移预测[J]. 煤炭学报, 2021, 46(S1): 529-538.

[12]屈世甲, 李鹏. 基于支架工作阻力大数据的工作面顶板矿压预测技术研究[J]. 矿业安全与环保, 2019, 46(02): 92-97.

[13]冀汶莉, 刘艺欣, 柴敬, 王斌. 基于随机森林的矿压预测方法[J]. 采矿与岩层控制工程学报, 2021, 3(03): 71-81.

[14]Tan T, Yang Z, Chang F, Zhao K. Prediction of the first weighting from the working face roof in a coal mine based on a GA-BP neural network[J]. Applied Sciences, 2019, 9(19):4159.

[15]王国胤, 代劲, 李昊. 基于多粒度认知计算的生产安全管理与决策[J]. 中国科学基金, 2021, 35(05): 752-758.

[16]李金海, 王飞, 吴伟志, 徐伟华, 杨习贝, 折延宏. 基于粒计算的多粒度数据分析方法综述[J]. 数据采集与处理, 2021, 36(03): 418-435.

[17]Bargiela A, Pedrycz W. Toward a Theory of Granular Computing for Human-Centered Information Processing[J]. IEEE Transactions on Fuzzy Systems, 2008, 16(2):320-330.

[18]焦李成, 杨淑媛, 刘芳, 王士刚, 冯志玺. 神经网络七十年: 回顾与展望[J]. 计算机学报, 2016, 39(08): 1697-1716.

[19]贺超峰, 华心祝, 杨科, 马菁花. 基于BP神经网络的工作面周期来压预测[J]. 安徽理工大学学报(自然科学版), 2012, 32(01): 59-63.

[20]Feng X, Wang Y, Yao J. A neural network model on real-time prediction of roof pressure in coal mines[J]. International Journal of Rock Mechanics and Mining Science, 1996, 33(6):647-653.

[21]Alya Ev S, Shahriari M, Pardo D, et al. Modeling extra-deep EM logs using a deep neural network[J]. Geophysics, 2021, 86(3):1-47.

[22]Han Z, Pedrycz W, Zhao J, et al. Hierarchical Granular Computing-Based Model and Its Reinforcement Structural Learning for Construction of Long-Term Prediction Intervals[J]. IEEE Transactions on Cybernetics, 2020, 52(1):1-11.

[23]Qin Y, Yang H, Deng Z, et al. A simplified model for computing pollutants release from granular pavement base to local aquifer[J]. Environmental Earth Sciences, 2014, 72(5):1533-1540.

[24]ZADEH L. Fuzzy sets and information granularity[G]. Advances in Fuzzy Set Theory and Applications, North-Holland, 1979: 3-18.

[25]王灵俊. 周期来压期间对综采工作面顶板管理影响探析[J]. 中国石油和化工标准与质量, 2020, 40(24): 86-88.

[26]Razani M, Yazdani-Chamzini A, Yakhchali S H. A novel fuzzy inference system for predicting roof fall rate in underground coal mines[J]. Safety Science, 2013, 55(2):26-33.

[27]蒋金泉, 曲华, 刘传孝. 巷道围岩弱结构灾变失稳与破坏区域形态的奇异性[J]. 岩石力学与工程学报, 2005(18): 3373-3379.

[28]谢广祥, 杨科, 常聚才. 非对称综放开采煤层三维应力分布特征及其层厚效应研究[J]. 岩石力学与工程学报, 2007(04): 775-779.

[29]Song X, Gu T, Yan Z. Effects of increasing working face’ s length on underground pressure behaviors of mining super-high faces under shallow coal seam[J]. Chinese Journal of Rock Mechanics and Engineering, 2007,0(S2):4007-4012.

[30]Zhao Y, Wang S, Hagan P, Ren L, Zou Z. Pressure-Arching Characteristics in Roof Blocks during Shallow Coal Mining[J]. Advances in Civil Engineering, 2018(1):1-12.

[31]郝海金, 吴健, 张勇, 袁宗本. 大采高开采上位岩层平衡结构及其对采场矿压显现的影响[J]. 煤炭学报, 2004(02): 137-141.

[32]弓仲标, 冯晓斌, 贺斌. 特厚煤层下分层综放面初采期间矿压显现规律分析[J]. 陕西煤炭, 2021, 40(05): 41-45.

[33]Zhang T, Zhao Y, Zhu G, Wang S, Jiao S, Engineering S. A multi-coupling analysis of mining-induced pressure characteristics of shallow-depth coal face in Shendong mining area[J]. Journal of China Coal Society, 2016, 41(S2), 287-296.

[34]乔懿麟, 黄克军, 耿耀强, 丁国峰. 采动影响下沿空留巷底板非对称性底鼓变形研究[J]. 煤炭技术, 2017, 36(07): 94-96.

[35]刘卫东, 孟晓静, 丁恩杰, 窦林名. 模糊逻辑在矿井冲击矿压预测中的应用[J]. 煤矿安全, 2008(01): 66-68.

[36]闫吉太, 梁广锋, 安满林, 卫建清, 冯增强. “孤岛”综采放顶煤工作面矿压预测预报[J]. 中国矿业大学学报, 1996(04): 98-103.

[37]Toma, Ambroi, Goran, Turk. Prediction of subsidence due to underground mining by artificial neural networks[J]. Computers & Geosciences, 2003, 29(5):627-637.

[38]Mohammed M , Watanabe K , Takeuchi S. Grey model for prediction of pore pressure change[J]. Environmental Earth Sciences, 2010, 60(7):1523-1534.

[39]周宣赤, 白春华, 林大超, 王仲琦. 顶板压力预报机制研究[J]. 煤炭学报, 2011, 36(S2): 299-303.

[40]杨硕, 王胄. 用粒子群优化BP神经网络的顶板矿压预测[J]. 陕西煤炭, 2010, 29(04): 83-86.

[41]Zhang R, Lowndes I. The application of a coupled artificial neural network and fault tree analysis model to predict coal and gas outbursts[J]. International Journal of Coal Geology, 2010, 84(2):141-152.

[42]赵毅鑫, 杨志良, 马斌杰, 宋红华, 杨东辉. 基于深度学习的大采高工作面矿压预测分析及模型泛化[J]. 煤炭学报, 2020, 45(01): 54-65.

[43]贾澎涛, 苗云风. 基于堆叠LSTM的多源矿压预测模型分析[J]. 矿业研究与开发, 2021, 41(08): 79-82.

[44]曾庆田, 吕珍珍, 石永奎, 田广宇, 林泽东, 李超. 基于Prophet+LSTM模型的煤矿井下工作面矿压预测研究[J]. 煤炭科学技术, 2021, 49(07): 16-23.

[45]Swpu P. Bottom hole pressure estimation using hybridization neural networks and grey wolves optimization[J]. Petroleum, 2018, 4(4):419-429.

[46]ZADEH L. Towards a theory of fuzzy information granulation and its centrality in human reasoning and fuzzy logic[J]. Fuzzy Sets and Systems, 1997, 90(2): 111-117.

[47]Pedrycz W. The Principle of Justifiable Granularity and an Optimization of Information Granularity Allocation as Fundamentals of Granular Computing[J]. Journal of Information Processing Systems, 2014, 7(3):397-412.

[48]Lin T. Granular computing rough set perspective[J]. The Newsletter of the IEEE Computational Intelligence Society, 2005, 2(4):1543-4281.

[49]Lin T and Young T, Granular computing I: The concept of granulation and its formal model[J]. International Journal of Granular Computing, 2009, 1(1):21-42.

[50]Panda M, and Tripathy B. Soft granular computing-based classification using hybrid fuzzy-KNN-SVM[J]. Intelligent Decision Technologies, 2016, 10(2):115-128.

[51]Mehmood F， Haider Z， Farooq U and Yin B. Droplet movement control using fuzzy logic controller and image processing along with pin valves inside microfluidic device[J]. Measurement and Control, 2019, 52(9-10): 1-12.

[52]吴珺, 王春枝. 面向大数据的多维粒矩阵关联分析及应用[J]. 计算机科学, 2017, 44(S2): 407-410+421.

[53]覃希, 苏一丹, 张雯. 商空间框架下的大规模SVM数据集约减法[J]. 计算机科学, 2013, 40(12): 104-107.

[54]Chen L F，Chen L，Liu B X . Research of SVM Multiclass Model Based on Granular Computing & Huffman Tree[J]. Computer Science, 2016, 43(1):64-68.

[55]W. Pedrycz and G. Vukovich. Granular neural networks[J]. Neurocomputing, 2011, 36(1):205-224.

[56]Han X, Yan C. GrC and ANN based fault diagnosis method of distribution network[C]//2011 International Conference on Electronic & Mechanical Engineering & Information Technology. Harbin, Heilongjiang, China: IEEE, 2011.

[57]Shen Z，Yang W. BP Neural Network Via Granular Computing for Prediction of Financial Trends[J]. Journal of Chinese Computer Systems, 2019, 40(03):527-532.

[58]Mao Z, Su Y, Xu G, Wang X, Huang Y, Yue W, Sun L, Xiong N. Spatio-temporal deep learning method for ADHD fMRI classification[J]. Information Sciences, 2019, 51(499):1-11.

[59]朱修彬. 用于系统建模和数据挖掘的粒度数据描述[D]. 西安电子科技大学, 2018.

[60]Moore R. Interval analysis[M]. Englewood Cliffs, NJ: Prentice-Hall, 1966.

[61]Zadeh L. From computing with numbers to computing with words: From manipulation of measurements to manipulation of perceptions[J]. Annals of the New York Academy of Sciences, 2001, 929(243):221-252.

[62]Pan M, Xun Y, Gu B, Huang J, Chen Y. Fuzzy-Set Theoretic Control Design for Aircraft Engine Hardware-in-the-Loop Testing: Mismatched Uncertainty and Optimal Design[J]. IEEE Transactions on Industrial Electronics, 2021, 69(7):7223-7233.

[63]Pawlak Z. Rough sets [J]. International journal of computer & information sciences, 1982, 11(5):341-356.

[64]Pedrycz W. From fuzzy sets to shadowed sets: Interpretation and computing[J]. International Journal of Intelligent Systems, 2009, 24(1):48-61.

[65]Pedrycz W. The principle of justifiable granularity and an optimization of information granularity allocation as fundamentals of granular computing[J]. Journal of Information Processing Systems, 2011, 7(3):397-412.

[66]Hochreiter S, Schmidhuber J. Long Short-Term Memory[J]. Neural Computation, 1997, 9(8):1735-1780.

[67]Razavian A, Azizpour H, Sullivan J, Carlsson S. CNN Features off-the-shelf: an Astounding Baseline for Recognition[J]. IEEE, 2014, 3(11):806-813.

[68]Baek J, Chung K. Accident risk prediction model based on attention-mechanism LSTM using modality convergence in multimodal[J]. Personal and Ubiquitous Computing, 2021, 4(02):1-11.

[69]Blatt M, Wiseman S, Domany E. Data clustering using a model granular magnet[J]. Neural Computation, 2014, 9(8): 1805-1842.

[70]Zhu X, Pedrycz W, Li Z. Development and analysis of neural networks realized in the presence of granular data[J]. IEEE Transactions on Neural Networks and Learning Systems, 2019, 31(9):1-14.

[71]Wang D, Zhu X, Pedycz W, Yu Z, Li Z. Development of Granular Fuzzy Relation Equations Based on a Subset of Data[J]. IEEE/CAA Journal of Automatica Sinica, 2021, 8(08):1416-1427.

[72]文冰梅, 赵联文, 黄磊. AIC准则与留一法交叉验证渐近等价的证明[J]. 统计与决策, 2022, 38(06): 40-43.

[73]梁宇轩. 基于多层注意力神经网络的地理传感器时间序列预测[D]. 西安电子科技大学, 2019.

[74]Reshef D, Reshef Y A, Sabeti P C. Detecting Novel Associations in Large Data Sets[J]. Science, 2011, 334(6062):1518-1524.

[75]Park H, Pedrycz W, Oh S. Granular Neural Networks and Their Development Through Context-Based Clustering and Adjustable Dimensionality of Receptive Fields[J]. IEEE Transactions on Neural Networks, 2009, 20(10):1604.

[76]Syeda M, Zhang Y, Yi P. Parallel granular neural networks for fast credit card fraud detection[C]//2022 IEEE International Conference on Fuzzy Systems. Honolulu, Hawaii, USA:C IEEE, 2002: 572-577.

[77]Sanchez D, Melin P. Optimization of modular granular neural networks using hierarchical genetic algorithms for human recognition using the ear biometric measure[J]. Engineering Applications of Artificial Intelligence, 2014, 27(01):41-56.

Wang D, Pedrycz W, Li Z. Granular Data Aggregation: An Adaptive Principle of the Justifiable Granularity Approach[J]. IEEE Transactions on Cybernetics, 2019, 49(2):417-426.