基于可变时序移位Transformer−LSTM的集成学习矿压预测方法

李泽西

doi:10.13272/j.issn.1671-251x.18142

基于可变时序移位Transformer−LSTM的集成学习矿压预测方法

doi: 10.13272/j.issn.1671-251x.18142

李泽西

西安科技大学通信与信息工程学院, 陕西西安　710054

详细信息

作者简介:
李泽西（2001—），男，湖北孝感人，研究方向为人工智能、电子信息技术在煤矿领域的应用，E-mail：1632982226@qq.com

中图分类号: TD323
计量
- 文章访问数: 117
- HTML全文浏览量: 155
- PDF下载量: 36
- 被引次数: 0
出版历程
- 收稿日期: 2023-05-10
- 修回日期: 2023-07-10
- 网络出版日期: 2023-08-03

Ensemble learning mine pressure prediction method based on variable time series shift Transformer-LSTM

LI Zexi

College of Communication and Information Technology, Xi'an University of Science and Technology, Xi'an 710054, China

摘要

摘要: 现有的矿压预测模型多为依赖固定长度时序序列特征的单一预测模型，难以准确捕捉矿压时序数据的复合特征，影响矿压预测的准确度。针对该问题，提出一种基于可变时序移位Transformer−长短时记忆（LSTM）的集成学习矿压预测方法。基于拉依达准则和拉格朗日插值法，剔除矿压监测数据中的异常值，插入缺失值，并进行归一化预处理；提出可变时序移位策略，划分不同尺度的矿压时序数据，避免固定长度时序序列可能存在的数据偏移问题；在此基础上，构建基于Transformer−LSTM的集成学习矿压预测模型，通过结合注意力机制和准确的时序特征表示能力，多层次捕捉矿压变化规律的动态特征，采用集成学习的投票算法，联合预测矿压数据，克服单一预测模型的局限性。实验结果表明：采用集成学习的投票算法可降低矿压预测平均绝对误差（MAE）和均方根误差（RMSE）的波动性，有效减小不同尺度特征序列对矿压预测结果的敏感性影响；Transformer−LSTM模型在2个综采工作面顶板矿压数据集上预测结果的MAE较Transformer模型分别提高了8.9%和9.5%，RMSE分别提高了12.7%和16.5%，且高于反向传播（BP）神经网络模型和LSTM模型，有效提升了矿压预测准确度。
- 矿压预测 /
- 可变时序移位 /
- Transformer−LSTM模型 /
- 集成学习 /
- 投票算法
Abstract: The existing mine pressure prediction models are mostly single prediction models that rely on fixed length time series features. It is difficult to accurately capture the composite features of mine pressure time series data, which affects the accuracy of mine pressure prediction. To solve this problem, an ensemble learning mine pressure prediction method based on variable time series shift Transformer-long short-time memory (LSTM) is proposed. Based on the Laida criterion and Lagrange polynomial method, the outlier values in the mine pressure monitoring data are eliminated, and the missing values are inserted. Then normalized preprocessing is performed. The paper proposes a variable time series shift strategy to divide mine pressure time series data at different scales. It avoids potential data shift issues that may exist in fixed length time series. On this basis, the ensemble learning mine pressure prediction model based on Transformer-LSTM is constructed. By combining the attention mechanism and the accurate time series feature representation capability, the dynamic features of the mine pressure change law are captured at multiple levels. The voting algorithm of ensemble learning is used to jointly predict the mine pressure data to overcome the limitations of a single prediction model. The experimental results show that the voting algorithm of ensemble learning can reduce the volatility of mean absolute error (MAE) and root mean square error (RMSE) of mine pressure prediction. It effectively reduces the sensitivity impact of different scale feature series to the mine pressure prediction results. Compared with the Transformer model, the MAE of the Transformer-LSTM model's prediction results on two roof mine pressure datasets of fully mechanized working faces improves by 8.9% and 9.5% respectively, and the RMSE has increased by 12.7% and 16.5% respectively. The above indexes are also higher than those of back propagation (BP) neural network model and LSTM model. The method proposed in the paper effectively improves the accuracy of mine pressure prediction.
- mine pressure prediction /
- variable time series shift /
- Transformer-LSTM model /
- ensemble learning /
- voting algorithm

HTML全文

图 1 基于可变时序移位Transformer−LSTM的集成学习矿压预测方法

Figure 1. Ensemble learning mine pressure prediction method based on variable time series shift Transformer-LSTM

下载: 全尺寸图片幻灯片

图 2 Transformer−LSTM模型结构

Figure 2. Transformer-LSTM model structure

下载: 全尺寸图片幻灯片

图 3 综采工作面矿压测站局部布置

Figure 3. Local layout of mine pressure measuring stations in fully mechanized working face

下载: 全尺寸图片幻灯片

图 4 不同矿压序列的预测结果

Figure 4. Prediction results of different mine pressure series

下载: 全尺寸图片幻灯片

图 5 不同模型的部分矿压预测结果

Figure 5. Part prediction results of mine pressure by different models

下载: 全尺寸图片幻灯片

表 1 13号测站部分矿压监测数据

Table 1. Part of measured mine pressure data of No.13 station

时间	支架编号	采集器编号	左柱压力/MPa	右柱压力/MPa	整架压力/MPa
2020−05−06T12:25:00	61	13	22.40	19.30	20.85
2020−05−06T12:30:00	61	13	22.10	18.60	20.35
2020−05−06T12:35:00	61	13	22.10	18.10	20.10
2020−05−06T12:40:00	61	13	21.90	17.20	19.55
2020−05−06T12:45:00	61	13	21.90	16.70	19.30

下载: 导出CSV

表 2 不同模型预测结果的评价指标

Table 2. Evaluation index comparison of prediction results of different models

数据集	模型	MAE/MPa	RMSE/MPa
数据集1	BP神经网络	2.214	3.756
	LSTM	1.847	3.632
	Transformer	1.313	2.447
	Transformer−LSTM	1.196	2.136
数据集2	BP神经网络	1.333	2.755
	LSTM	1.205	2.764
	Transformer	0.820	1.810
	Transformer−LSTM	0.742	1.512

下载: 导出CSV

参考文献(23)

[1]	钱鸣高,许家林,王家臣. 再论煤炭的科学开采[J]. 煤炭学报,2018,43(1):1-13. QIAN Minggao,XU Jialin,WANG Jiachen. Further on the sustainable mining of coal[J]. Journal of China Coal Society,2018,43(1):1-13.
[2]	冀汶莉,刘艺欣,柴敬,等. 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(3):71-81. JI Wenli,LIU Yixin,CHAI Jing,et al. Mine pressure prediction method based on random forest[J]. Journal of Mining and Strata Control Engineering,2021,3(3):71-81.
[3]	曹富荣,吴学松,传金平,等. 冲击危险区域震动波CT反演探测技术应用[J]. 煤炭科技,2021,42(2):67-71,76. CAO Furong,WU Xuesong,CHUAN Jinping,et al. Application of CT inversion detection technology for shock wave in hazardous area of rock burst[J]. Coal Science & Technology Magazine,2021,42(2):67-71,76.
[4]	常峰. 基于GA−BP神经网络的工作面顶板矿压预测模型应用研究[D]. 徐州: 中国矿业大学, 2019. CHANG Feng. Application research of the prediction model for the coal working face roof pressure based on GA-BP neural networks[D]. Xuzhou: China University of Mining and Technology, 2019.
[5]	FENG Xiating,WANG Yongjia,YAO Jianguo. A neural network model for real-time roof pressure prediction in coal mines[J]. International Journal of Rock Mechanics and Mining Science & Geomechanics Abstracts,1996,33(6):647-653.
[6]	柴敬,王润沛,杜文刚,等. 基于XGBoost的光纤监测矿压时序预测研究[J]. 采矿与岩层控制工程学报,2020,2(4):64-71. CHAI Jing,WANG Runpei,DU Wengang,et al. Study on time series prediction of rock pressure by XGBoost in optical fiber monitoring[J]. Journal of Mining and Strata Control Engineering,2020,2(4):64-71.
[7]	程敬义,万志军,PENG Syd S,等. 基于海量矿压监测数据的采场支架与顶板状态智能感知技术[J]. 煤炭学报,2020,45(6):2090-2103. CHENG Jingyi,WAN Zhijun,PENG Syd S,et al. Technology of intelligent sensing of longwall shield supports status and roof strata based on massive shield pressure monitoring data[J]. Journal of China Coal Society,2020,45(6):2090-2103.
[8]	徐刚,范志忠,张春会,等. 宏观顶板活动支架增阻类型与预测模型[J]. 煤炭学报,2021,46(11):3397-3407. doi: 10.13225/j.cnki.jccs.2020.1951 XU Gang,FAN Zhizhong,ZHANG Chunhui,et al. Support increasing resistance forms of macroscopic roof support and its prediction model[J]. Journal of China Coal Society,2021,46(11):3397-3407. doi: 10.13225/j.cnki.jccs.2020.1951
[9]	惠艳荔. 电磁辐射技术在采煤工作面矿压预测中的应用[J]. 煤炭技术,2013,32(12):73-75. HUI Yanli. Application of electromagnetic radiation technology in mine pressure observation at coal face[J]. Coal Technology,2013,32(12):73-75.
[10]	李静,何江,巩思园,等. 上覆遗留煤柱作用下冲击矿压预测预警案例研究[J]. 煤炭学报,2016,41(增刊2):305-310. doi: 10.13225/j.cnki.jccs.2015.1835 LI Jing,HE Jiang,GONG Siyuan,et al. Case study on prediction and warning of rock burst under overlying pillar[J]. Journal of China Coal Society,2016,41(S2):305-310. doi: 10.13225/j.cnki.jccs.2015.1835
[11]	LI Ye,SHI Xiaohu. Mine pressure prediction study based on fuzzy cognitive maps[J]. International Journal of Computational Intelligence and Applications,2020,19(3):1-13.
[12]	霍丙杰,荆雪冬,于斌,等. 坚硬顶板厚煤层采场来压强度分级预测方法研究[J]. 岩石力学与工程学报,2019,38(9):1828-1835. HUO Bingjie,JING Xuedong,YU Bin,et al. A classification forecasting method for the weighting intensity of stopes of thick coal seams with hard roofs[J]. Chinese Journal of Rock Mechanics and Engineering,2019,38(9):1828-1835.
[13]	赵铭生,刘守强,纪润清,等. 基于遗传算法优化BP神经网络的华北型煤田矿压破坏带深度预测[J]. 矿业研究与开发,2020,40(6):89-93. ZHAO Mingsheng,LIU Shouqiang,JI Runqing,et al. Depth prediction of mining pressure failure zone in north China coalfield based on BP neural network optimized by genetic algorithm[J]. Mining Research and Development,2020,40(6):89-93.
[14]	赵毅鑫,杨志良,马斌杰,等. 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(1):54-65. ZHAO Yixin,YANG Zhiliang,MA Binjie,et al. Deep learning prediction and model generalization of ground pressure for deep longwall face with large mining height[J]. Journal of China Coal Society,2020,45(1):54-65.
[15]	曾庆田,吕珍珍,石永奎,等. 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(7):16-23. doi: 10.13199/j.cnki.cst.2021.07.002 ZENG Qingtian,LYU Zhenzhen,SHI Yongkui,et al. Research on prediction of underground coal mining face pressure based on Prophet+LSTM model[J]. Coal Science and Technology,2021,49(7):16-23. doi: 10.13199/j.cnki.cst.2021.07.002
[16]	李泽萌. 基于LSTM的采煤工作面矿压预测方法研究[D]. 西安: 西安科技大学, 2020. LI Zemeng. Research on prediction method of mining pressure in coal face based on LSTM[D]. Xi'an: Xi'an University of Science and Technology, 2020.
[17]	SHEN Cheng,BAO Xuejing,TAN Jiubin,et al. Two noise-robust axial scanning multi-image phase retrieval algorithms based on Pauta criterion and smoothness constraint[J]. Optics Express,2017,25(14):16235-16249. doi: 10.1364/OE.25.016235
[18]	HIGHAM N J. The numerical stability of barycentric Lagrange interpolation[J]. IMA Journal of Numerical Analysis,2004,24(4):547-556. doi: 10.1093/imanum/24.4.547
[19]	HAN Kai,XIAO An,WU Enhua,et al. Transformer in transformer[J]. Advances in Neural Information Processing Systems,2021,34:15908-15919.
[20]	YU Yong,SI Xiaosheng,HU Changhua,et al. A review of recurrent neural networks:LSTM cells and network architectures[J]. Neural Computation,2019,31(7):1235-1270. doi: 10.1162/neco_a_01199
[21]	苏振国,邓志刚,李国营,等. 顶板深孔爆破防治小煤柱冲击地压研究[J]. 矿业安全与环保,2019,46(4):21-25,29. doi: 10.3969/j.issn.1008-4495.2019.04.005 SU Zhenguo,DENG Zhigang,LI Guoying,et al. Study on prevention and control of small coal pillar rock burst by roof deep hole blasting[J]. Mining Safety & Environmental Protection,2019,46(4):21-25,29. doi: 10.3969/j.issn.1008-4495.2019.04.005
[22]	田原,庞骁,赵文祎,等. 基于Transformer的滑坡短期位移预测模型[J]. 北京大学学报(自然科学版),2023,59(2):197-210. TIAN Yuan,PANG Xiao,ZHAO Wenyi,et al. A transformer-based model for short-term landslide displacement prediction[J]. Acta Scientiarum Naturalium Universitatis Pekinensis,2023,59(2):197-210.
[23]	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(6):40-43. WEN Bingmei,ZHAO Lianwen,HUANG Lei. Proof of the asymptotic equivalence between AIC criterion and LOOCV[J]. Statistics & Decision,2022,38(6):40-43.