基于Transformer的矿井内因火灾时间序列预测方法

王树斌; 王旭; 闫世平; 王珂

doi:10.13272/j.issn.1671-251x.2023100084

基于Transformer的矿井内因火灾时间序列预测方法

doi: 10.13272/j.issn.1671-251x.2023100084

1.
陕煤集团神木柠条塔矿业有限公司，陕西榆林　719300
2.
中国矿业大学计算机科学与技术学院，江苏徐州　221116

基金项目: 国家自然科学基金重点项目（52130411）。

详细信息

作者简介:
王树斌（1967—），男，陕西蒲城人，高级工程师，主要从事煤矿智能化工作，E-mail：793978271@qq.com

中图分类号: TD75
计量
- 文章访问数: 183
- HTML全文浏览量: 36
- PDF下载量: 25
- 被引次数: 0
出版历程
- 收稿日期: 2023-10-27
- 修回日期: 2024-03-15
- 网络出版日期: 2024-04-11

Transformer based time series prediction method for mine internal caused fire

1.
Shanmei Group Shenmu Ningtiaota Mining Co., Ltd., Yulin 719300, China
2.
School of Computer Science & Technology, China University of Mining and Technology, Xuzhou 221116, China

摘要

摘要: 传统的基于机器学习的矿井内因火灾预测方法尽管具备一定的预测能力，然而在处理复杂的多变量数据时不能有效捕捉数据间的全局依赖关系，导致预测精度较低。针对上述问题，提出了一种基于Transformer的矿井内因火灾时间序列预测方法。首先，采用Hampel滤波器和拉格朗日插值法对数据进行异常值检测和缺失值填补。然后，利用Transformer的自注意力机制对时间序列数据进行特征提取及趋势预测。最后，通过调节滑动窗口的大小与步长，在不同的时间步长和预测长度下对模型进行不同时间维度的训练。结合气体分析法将矿井火灾产生的标志性气体（CO，O₂，N₂，CO₂，C₂H₂，C₂H₄，C₂H₆）作为模型输入变量，其中CO作为模型输出的目标变量，O₂，N₂，CO₂，C₂H₂，C₂H₄，C₂H₆作为模型输入的协变量。选取陕煤集团柠条塔煤矿S1206回风隅角火灾预警的束管数据进行实验验证，结果表明：① 对CO进行单变量预测和多变量预测，多变量预测相比单变量预测有着更高的预测精度，说明多变量预测能通过捕捉序列间的相关性提高模型的预测精度。② 当时间步长固定时，基于Transformer的矿井内因火灾预测模型的预测精度随着预测长度的增加而下降。当预测长度固定时，模型的预测精度随时间步长增加而提高。③ Transformer算法的预测精度较长短时记忆（LSTM）算法和循环神经网络（RNN）算法分别提高了7.1%～12.6%和20.9%～24.9%。
- 矿井内因火灾 /
- Transformer /
- 时间序列 /
- 标志性气体 /
- 自注意力机制
Abstract: Although traditional machine learning based methods for predicting mine internal caused fire have certain predictive capabilities, they cannot effectively capture global dependencies between complex multivariate data, resulting in low prediction precision. In order to solve the above problems, a transformer based time series prediction method for mine internal caused fire is proposed. Firstly, the Hampel filter and Lagrange interpolation method are used to detect outliers and fill in missing values in the data. Secondly, the self attention mechanism of Transformer is utilized to extract features and predict trends from time series data. Finally, by adjusting the size and step size of the sliding window, the model is trained in different time dimensions at different time steps and prediction lengths. Combining gas analysis method, the iconic gases generated by mine fires (CO, O₂, N₂, CO₂, C₂H₂, C₂H₄, C₂H₆) are used as input variables for the model, with CO as the target variable for model output and O₂, N₂, CO₂, C₂H₂, C₂H₄, C₂H₆ as covariates for model input. Selecting the bundle data of S1206 return air corner fire warning in Ningtiaota Coal Mine of Shanmei Coal Group for experimental verification, the results show the following points. ① Univariate prediction and multivariate prediction of CO show that multivariate prediction has higher prediction precision than univariate prediction, indicating that multivariate prediction can improve the prediction precision of the model by capturing the correlation between sequences. ② When the time step is fixed, the prediction precision of the Transformer based mine internal caused fire prediction model decreases with the increase of prediction length. When the prediction length is fixed, the prediction precision of the model improves with the increase of time step. ③ The prediction accuracy of the Transformer algorithm is improved by 7.1%-12.6% and 20.9%-24.9% over the long short-term memory (LSTM) algorithm and recurrent neural network (RNN) algorithm, respectively.
- mine internal caused fire /
- Transformer /
- time series /
- iconic gas /
- self attention mechanism

HTML全文

图 1 基于Transformer的矿井内因火灾预测模型

Figure 1. Mine internal caused fire prediction model based on Transformer

下载: 全尺寸图片幻灯片

图 2 Transformer算法模型

Figure 2. Transformer algorithm model

下载: 全尺寸图片幻灯片

图 3 测试样本中预测值与真实值的拟合曲线

Figure 3. Fitting curve of predicted values and true values in test samples

下载: 全尺寸图片幻灯片

图 4 对CO进行单变量预测和多变量预测的拟合曲线

Figure 4. Fitted curves for univariate and multivariate predictions for CO

下载: 全尺寸图片幻灯片

表 1 基于 Transformer的矿井内因火灾预测模型在不同时间步长下的误差

Table 1. Errors of mine internal caused fire prediction model based on Transformer under different time dimensions

预测长度	时间步长20		时间步长30		时间步长40
预测长度	MAE	RMSE	MAE	RMSE	MAE	RMSE
5	0.012 3	0.0015	0.0119	0.0147	0.0113	0.0139
10	0.012 7	0.0160	0.0124	0.0156	0.0119	0.0144
15	0.014 1	0.0183	0.0126	0.0161	0.0128	0.0166

下载: 导出CSV

表 2 不同算法预测结果比较

Table 2. Comparison of the prediction results of the different algorithms

时间步长	MAE
时间步长	Transformer	LSTM	RNN
20	0.0160	0.0183	0.0213
30	0.0156	0.0178	0.0200
40	0.0144	0.0155	0.0182

下载: 导出CSV

参考文献(21)

[1]	孙继平,孙雁宇. 矿井火灾监测与趋势预测方法研究[J]. 工矿自动化,2019,45(3):1-4. SUN Jiping,SUN Yanyu. Research on methods of mine fire monitoring and trend prediction[J]. Industry and Mine Automation,2019,45(3):1-4.
[2]	雷忠青. 矿井火灾预警技术研究[J]. 煤炭与化工,2015,38(10):148-149,153. LEI Zhongqing. Study on mine fire warning technology[J]. Coal and Chemical Industry,2015,38(10):148-149,153.
[3]	李军,张书林,龚仲强,等. 基于多传感信息融合的输送带运输巷道火灾预警方法[J]. 煤矿机械,2017,38(10):132-135. LI Jun,ZHANG Shulin,GONG Zhongqiang,et al. Fire alarm method of conveying belt transportation roadway based on multi-sensor information fusion[J]. Coal Mine Machinery,2017,38(10):132-135.
[4]	贝伟仰,江虹. 基于红外测温的无线温度监测系统的研究[J]. 计算机测量与控制,2011,19(10):2397-2400. BEI Weiyang,JIANG Hong. Study of temperature wireless monitoring system based on infrared technology[J]. Computer Measurement & Control,2011,19(10):2397-2400.
[5]	张箫剑. 基于光纤测温技术的采空区煤温监测研究[D]. 淮南:安徽理工大学,2015. ZHANG Xiaojian. Research on coal temperature monitoring in the goaf based on optical fiber pyrometer[D]. Huainan:Anhui University of Science and Technology,2015.
[6]	任慧,孙继平,刘晓阳. 矿用电缆火灾图像识别方法[J]. 辽宁工程技术大学学报,2007,26(1):85-88. REN Hui,SUN Jiping,LIU Xiaoyang. Image recognition method of mine cable fires[J]. Journal of Liaoning Technical University,2007,26(1):85-88.
[7]	唐杰,周洋,杨萌,等. 采用颜色混合模型和特征组合的视频烟雾检测[J]. 光电子·激光,2017,28(7):751-758. TANG Jie,ZHOU Yang,YANG Meng,et al. A smoke detection algorithm using color mixture model and feature combination[J]. Journal of Optoelectronics·Laser,2017,28(7):751-758.
[8]	PREMAC E,VINSLEY S S,SURESH S. Multi feature analysis of smoke in YUV color space for early forest fire detection[J]. Fire Technology,2016,52(5):1319-1342. doi: 10.1007/s10694-016-0580-8
[9]	胡云. 基于红外热成像技术的矿井火灾识别系统研究[D]. 淮南:安徽理工大学,2016. HU Yun. The research of mine fire recognition system based on infrared thermal imaging technology[D]. Huainan:Anhui University of Science and Technology,2016.
[10]	蔡冬雷. 基于AFSA−FCM的火灾预测与控制系统的研究[D]. 阜新:辽宁工程技术大学,2017. CAI Donglei. Research of fire prediction and control system based on AFSA-FCM clustering algorithm[D]. Fuxin:Liaoning Technical University,2017.
[11]	魏超,童敏明,任子晖,等. 基于激光气体分析的矿井火灾预警装置[J]. 软件,2011,32(4):77-78,83. doi: 10.3969/j.issn.1003-6970.2011.04.026 WEI Chao,TONG Minming,REN Zihui,et al. Mine fire alarm system based on the analysis of laser gas[J]. Software,2011,32(4):77-78,83. doi: 10.3969/j.issn.1003-6970.2011.04.026
[12]	侯毛伟,马海林. 光纤传感系统在煤矿火灾预测预警的应用[J]. 科技风,2013(17):116-117,119. doi: 10.3969/j.issn.1671-7341.2013.17.096 HOU Maowei,MA Hailin. Application of optical fiber optic sensing system in coal mine fire prediction and early warning[J]. Technology Wind,2013(17):116-117,119. doi: 10.3969/j.issn.1671-7341.2013.17.096
[13]	陈雅,蒋仲安,谭聪. 基于危机征兆的煤矿内因火灾预测模型的研究[J]. 矿业安全与环保,2015,42(1):56-59. doi: 10.3969/j.issn.1008-4495.2015.01.015 CHEN Ya,JIANG Zhong'an,TAN Cong. Study on prediction model of coal mine spontaneous combustion based on crisis symptoms[J]. Mining Safety & Environmental Protection,2015,42(1):56-59. doi: 10.3969/j.issn.1008-4495.2015.01.015
[14]	刘永立,刘晓伟,王海涛. 基于LSTM神经网络的煤矿火灾预测[J]. 黑龙江科技大学学报,2023,33(1):1-5. doi: 10.3969/j.issn.2095-7262.2023.01.001 LIU Yongli,LIU Xiaowei,WANG Haitao. Coal mine fire prediction based on LSTM neural network[J]. Journal of Heilongjiang University of Science and Technology,2023,33(1):1-5. doi: 10.3969/j.issn.2095-7262.2023.01.001
[15]	叶咪. 基于Transformer的多特征期货价格趋势预测研究[D]. 南京:南京财经大学,2023. YE Mi. Research on multi-feature futures price trend prediction based on Transformer model[D]. Nanjing:Nanjing University of Finance & Economics,2023.
[16]	张帅,刘文霞,唐浩洋,等. 一种基于Transformer多特征融合的短期负荷预测方法[J/OL]. 华北电力大学学报(自然科学版):1-9 [2023-06-07]. http://kns.cnki.net/kcms/detail/13.1212.TM.20230607.0918.002.html. ZHANG Shuai,LIU Wenxia,TANG Haoyang,et al. A short-term load forecasting method based on multi-feature fusion using Transformer[J/OL]. Journal of North China Electric Power University(Natural Science Edition):1-9[2023-06-07]. http://kns.cnki.net/kcms/detail/13.1212.TM.20230607.0918.002.html.
[17]	孙欣,王思敏,谢敬东,等. 考虑多维影响因素的改进Transformer−PSO短期电价预测方法[J/OL]. 上海交通大学学报: 1-16[2023-06-01]. https://doi.org/10.16183/j.cnki.jsjtu.2023.065. SUN Xin,WANG Simin,XIE Jingdong. et al. Improved Transformer-PSO short-term electricity price forecasting method considering multidimensional influencing factors[J/OL]. Journal of Shanghai Jiaotong University: 1-16[2023-06-01]. https://doi.org/10.16183/j.cnki.jsjtu.2023.065.
[18]	耿鑫月,胡昌华,郑建飞,等. 双时间尺度下基于Transformer的锂电池剩余寿命预测[J]. 空间控制技术与应用,2023,49(4):119-126. doi: 10.3969/j.issn.1674-1579.2023.04.013 GENG Xinyue,HU Changhua,ZHENG Jianfei,et al. Remaining useful life prediction of lithium batteries based on transformer under the dual time scales[J]. Aerospace Control and Application,2023,49(4):119-126. doi: 10.3969/j.issn.1674-1579.2023.04.013
[19]	王译文,黎建军,曲再鹏. 基于Transformer的短期血糖预测方法研究[J]. 中国计量大学学报,2023,34(3):372-378. doi: 10.3969/j.issn.2096-2835.2023.03.007 WANG Yiwen,LI Jianjun,QU Zaipeng. Research on short-term blood glucose prediction methods based on Transformer[J]. Journal of China University of Metrology,2023,34(3):372-378. doi: 10.3969/j.issn.2096-2835.2023.03.007
[20]	田晟,胡啸. 基于Transformer模型的车辆轨迹预测[J/OL]. 广西师范大学学报(自然科学版): 1-12[2023-06-07]. https://doi.org/10.16088/j.issn.1001-6600.2023061203. TIAN Sheng,HU Xiao. Vehicle trajectory prediction based on transformer model [J/OL]. Journal of Guangxi Normal University(Natural Science Edition): 1-12[2023-06-07]. https://doi.org/10.16088/j.issn.1001-6600.2023061203.
[21]	郭海涛,张志强. 利用标志气体对煤层火灾进行早期监测[J]. 黑龙江科技信息,2007(22):60. GUO Haitao,ZHANG Zhiqiang. Early monitoring of coal seam fire by using marker gas[J]. Heilongjiang Science and Technology Information,2007(22):60.