井下人员定位系统作为矿山“六大安全避险系统”之一，为地下矿山安全生产提供了良好保障。然而，井下作业空间局限、巷道狭长幽闭、煤矿环境复杂、存在较多干扰，单一的定位方法无法满足高精度的井下定位需求。本文在剖析现有定位技术的基础上，结合井下环境特点，开展行人航位推算技术与超宽带技术在井下人员定位领域的研究，并提出一种组合定位策略，提高井下人员定位系统的精度和稳定性。主要内容如下。

（1）针对传统PDR累计误差大和多运动状态下定位精度不高的问题，提出一种基于行人运动状态识别辅助的PDR定位算法框架。通过搭建神经网络模型进行运动状态识别，并利用该结果辅助进行步频检测和步长估算算法的优化。针对传统峰值检测算法使用单一固定阈值导致步频计算偏差过大的问题，引入滑动窗口思想结合运动状态识别结果，提出一种自适应步频检测算法，误检率下降了12.5%。针对多运动状态下步长估算模型不精确的问题，结合运动状态识别结果进行个性化步长公式设计，改进后的步长估算效果较常用的Weinberg算法精度提高了1.52%；针对定位过程中航向角误差较大的问题，提出了基于地图匹配的航向角修正算法，结合井下巷道环境设定地图标志位对PDR定位的航向误差进行抑制。实验结果表明，加入地图标志位后的航向估计平均误差降低了69.47%。通过对三种算法的优化，提高了PDR技术的定位精度。

（2）针对井下LOS/NLOS环境混合交错、不易区分的问题，提出一种基于BO-LGBM的NLOS识别方法。与传统机器学习和神经网络方法相比，识别精度提高了4.7%。针对UWB定位过程中存在的LOS误差和NLOS误差，根据井下巷道环境特点分别建立了多项式拟合和指数型拟合的测距修正模型，通过对比两种模型在不同环境下的修正效果来选择最优拟合模型，实验结果表明：多项式拟合模型适用于LOS环境下的测距修正，而指数函数模型更适用于NLOS环境下的测距误差修正。两种环境下修正后的测距误差分别降低了74%和68.3%。使用修正后的测距值进行两种不同坐标解算算法的对比，实验结果表明，梯度下降法解算的UWB定位坐标精度优于最小二乘法。

（3）针对单一定位技术在井下复杂环境中精度低和稳定性差的问题，对两种单一定位技术的误差特点进行分析，设计了一种非视距评判依据，利用基站与标签间距离的累积分布函数来评估当前位置受NLOS影响程度，并建立了基于自适应扩展卡尔曼滤波的PDR与UWB组合定位模型。分别在井下LOS和NLOS场景下验证该模型性能，结果表明，自适应EKF定位方案优于相单一PDR、UWB定位方案、EKF组合定位方案及PDR/WIFI组合定位方案。视距环境下，定位精度分别提升了68.6%、51%、34%和34.7%；非视距环境下，PDR定位与真实轨迹偏差较大，UWB定位存在定位信息缺失现象，自适应EKF定位方案可以输出完整且贴合真实曲线的定位轨迹，更适用于井下高精度人员定位系统，验证了本文所提算法的稳定性和有效性。

As one of the "six safety systems", the underground personnel positioning system provides a good guarantee for the safe production of underground mines. However, the limited underground working space, narrow and long tunnel claustrophobia, complex coal mine environment, there are many disturbances, a single positioning method can not meet the requirements of high-precision underground positioning. Based on the analysis of existing positioning technology and combining the characteristics of underground environment, this paper carries out the research of pedestrian dead estimation technology and UWB technology in the field of underground personnel positioning, and puts forward a combination positioning strategy to improve the accuracy and stability of underground personnel positioning system. The main contents are as follows.

(1) Aiming at the problems of large cumulative error of traditional PDR and low positioning accuracy in multi-motion state, a PDR positioning algorithm framework based on pedestrian motion state detection is proposed. The neural network is used to identify the motion state, and the results are used to help optimise the step frequency detection and step size estimation algorithms. To solve the problem that the traditional peak detection method using a single threshold causes excessive deviation in the step frequency calculation, an adaptive step frequency detection algorithm is proposed by introducing the idea of sliding window. The false detection rate is reduced by 12.5%. Aiming at the inaccuracy of the step estimation model under multiple motion states, the personalised step formula is designed by combining the results of motion state recognition. The accuracy of the improved step estimation is increased by 1.52% compared to the commonly used Weinberg algorithm. Considering the large heading error in the positioning process, a heading correction algorithm based on map matching is proposed, and the map marker is set in combination with the underground road environment to suppress the heading error of PDR positioning. The experimental results show that the average heading estimation error is reduced by 69.47% after adding the map marker. The positioning accuracy of PDR is improved by optimising the three techniques.

(2) A BO-LGBM-based NLOS identification method is proposed to solve the problem of mixed and indistinguishable LOS/NLOS environments in boreholes. Compared to traditional machine learning and neural network methods, the detection accuracy is improved by 4.7%. Aiming at the LOS and NLOS errors existing in the UWB positioning process, polynomial fitting and exponential fitting range correction models were established, respectively, according to the environmental characteristics of the underground roadway. The optimal fitting model was selected by comparing the correction effects of the two models in different environments. The experimental results show that The polynomial fitting model is suitable for range correction in the LOS environment, while the exponential function model is more suitable for range error correction in the NLOS environment. The corrected range error was reduced by 74% and 68.3% respectively in the two environments. The results show that the accuracy of the UWB coordinates calculated by the gradient descent method is better than that of the least squares method.

(3) To solve the problem of low precision and instability of single-point positioning technology in complex underground environments, the error characteristics of two single-point positioning technologies were analyzed, and a non-line-of-sight (NLOS) judgment criterion was designed. The current position was evaluated by the cumulative distribution function of the distance between the base station and the tag to assess the degree of NLOS influence. An indoor positioning model based on adaptive extended Kalman filter (EKF) combining PDR and UWB was established. The model performance was verified in LOS and NLOS scenarios underground, the results show that in the line-of-sight environment, the positioning accuracy was improved by 68.6%, 51%, 34%, and 34.7%; in the non-line-of-sight environment, the PDR positioning had a large deviation from the actual trajectory, and the UWB positioning had a problem of missing positioning information. The adaptive EKF positioning scheme can output a complete and consistent positioning trajectory, which is more suitable for the high-precision personnel positioning system in underground environments, verifying the stability and effectiveness of the proposed algorithm.

[1]康红普,谢和平,任世华,等.全球产业链与能源供应链重构背景下我国煤炭行业发展策略研究[J].中国工程科学,2022,24(6):26‒37.

[2]丁国峰,吕振福,曹进成等.我国大型煤炭基地开发利用现状分析[J].能源与环保,2020,42(11):107-110+120.

[3]王显政.我国煤矿安全生产50年回顾与展望[J].煤矿安全,2020,51(10):1-4.

[4]谢军,庄建楼,康成斌.基于北斗系统的物联网技术与应用[J].南京航空航天大学学报,2021,53(03):329-337.

[5]吕志强.煤矿井下通信系统的现状及趋势[J].中国煤炭,2014,40(05):89-92.

[6]孙继平.煤矿信息化自动化新技术与发展[J].煤炭科学技术,2016,44(01):19-23+83.

[7]邢云冰,龙广玉,胡春雨,等.基于SVM的类别增量人体活动识别方法[J].计算机科学,2022,49(05):78-83.

[8]何坚,郭泽龙,刘乐园,等.基于滑动窗口和卷积神经网络的可穿戴人体活动识别技术[J].电子与信息学报,2022,44(01):168-177.

[9]孙彦玺,陈继斌,武东辉.基于卷积神经网络-双向长短期记忆网络的人体活动识别方法[J].科学技术与工程,2022,22(04):1517-1525.

[10]黄彬,常俊,武浩,等.基于注意力机制的双向LSTM人体活动识别[J].传感器与微系统,2023,42(09):121-124.

[11]KANG W,HAN Y. Smart PDR:Smartphone-Based Pedestrian Dead Reckoning for ndoor Localization[J]. Ieee Sensors Journal,2015,15(5):2906-2916.

[12]BEAUREGARD S,HAAS H. Pedestrian dead reckoning:a basis for personal positioning [M]. Proccedings of the 3rd Workshop on Positioning. Navigation and Communication, 2006:27-35.

[13]毕京学,汪云甲,曹鸿基,等.一种波峰波谷检测的智能手机计步算法[J].中国惯性技术学报,2020,28(03):287-292.

[14]程国良,李飞,张言哲.一种基于自适应波峰检测的MEMS计步算法[J].中国惯性技术学报,2015,(3):315-321.

[15]朱庄生,张雨龙,李驰.基于MEMS惯性测量单元的多源信息自适应步数检测方法[J].中国惯性技术学报,2017,25(3):299-303.

[16]毕京学,汪云甲,齐红霞,等.基于随机森林回归的智能手机用步长估计模型[J].中国惯性技术学报,2020,28(02):147-152.

[17]欧毅,张泽欣,李星海,等.基于多节点姿态信息融合的行人步长估计算法[J].重庆理工大学学报(自然科学),2021,35(03):112-118.

[18]路永乐,惠嘉威,杨杰,等.基于自适应步长约束的行人导航系统误差修正算法[J].中国惯性技术学报,2023,31(02):126-131+140.

[19]郭倩倩,崔丽珍,杨勇,等.基于LSTM个性化步长估计的井下人员精准定位PDR算法[J].工矿自动化,2022,48(01):33-39.

[20]朱平,余学祥,韩雨辰,等.一种基于深度学习的高精度行人步长估算方法[J].测绘科学,2023,48(06):19-26+130.

[21]郭倩倩,崔丽珍,杨勇,等.煤矿井下基于降噪自编码器的航向估计算法研究[J].传感技术学报,2021,34(06):797-803.

[22]彭雅慧,潘树国,高旺,等.基于直线检测和数字地图匹配的车辆航向角估计[J].电子测量与仪器学报,2022,36(03):194-201.

[23]林新华,赵陶然,林威,等.基于序贯自适应误差状态Kalman滤波的MARG姿态估计算法[J].中国惯性技术学报,2023,31(12):1175-1180.

[24]Kermajani H R,Gomez C. Modeling the network convergence time in RPL in error- prone,IEEE802.15.4 chain topology multihop networks[C]//2014 11th International Symposium on Wireless Communications Systems(IsWCS).IEEE,2014:365-369.

[25]刘辉.基于PDR和UWB的室内定位实验研究[D].武汉:中国矿业大学,2022.

[26]张瑞峰,张忠娟,吕辰刚.基于质心-Taylor的UWB室内定位算法研究[J].重庆邮电大学学报(自然科学版),2011,23(06):717-721.

[27]卫星,田傲然,陆阳.基于动态阈值检测的UWB测距算法[J].合肥工业大学学报(自然科学版),2018,41(09):1178-1183+1266.

[28]鲁旭涛,薛凯亮,李静,等.基于神经网络预测的室内测距算法研究[J].国外电子测量技术,2021,40(07):70-76.

[29]李晓辉,杜洋帆,石潇竹,等.基于信道模糊关联识别的NLOS测距误差补偿算法[J].北京邮电大学学报,2020,43(01):21-27.

[30]戢静红,张振宇,邓平.一种基于随机森林的LOS/NLOS基站识别方法[J].电讯技术,2023,63(10):1596-1602.

[31]Wang T,Hu K K,LiZH,et al. A semi-supervised learning approach for UWB ranging error mitigation[J]. IEEE Wireless Communications Letters,2021,10(3):688-691.

[32]朱代先,孙小婷,郭明香,等.矿山井下超宽带无线定位算法研究[J].西安科技大学学报,2020,40(02):356-362.

[33]张少辉,亓玉浩,翟方文,等.基于冗余距离筛选的UWB定位优化方法[J].清华大学学报(自然科学版),2022,62(05):934-942.

[34]蒋锐,虞跃,徐友云,等.基于CHAN的改进卡尔曼滤波室内定位算法[J].通信学报,2023,44(02):136-147.

[35]李雪强,李建胜,王安成,等.基于抗差估计的VIO/UWB自适应组合定位算法[J].中国惯性技术学报,2023,31(11):1083-1091.

[36]田广亮,张丽杰.SVR辅助改进鲁棒卡尔曼滤波的UWB/INS组合定位方法[J].电子测量技术,2022,45(03):79-84.

[37]郁露,唐超礼,黄友锐,等.基于UWB和IMU的煤矿机器人紧组合定位方法研究[J].工矿自动化,2022,48(12):79-85.

[38]潘树国,贾丰硕,魏建雄,等.基于UWB单基站系统/INS紧组合的智能车定位方法[J].中国惯性技术学报,2023,31(12):1196-1202.

[39]李倩,蒋正华,孙炎,等.基于因子图的INS/UWB室内行人紧组合定位技术[J].仪器仪表学报,2022,43(05):32-45.

[40]郭倩倩.煤矿井下基于PDR/Wi-Fi多源融合定位算法研究[D].内蒙古,内蒙古科技大学,2022.

[41]艾青,杨俊杰,蒋伟,等.改进PDR与RSSI融合的室内定位方法[J].传感器与微系统,2023,42(12):75-78+82.

[42]邵小强,原泽文,杨永德,等.基于GRU-DRSN的双通道人体活动识别[J].科学技术与工程,2024,24(02):676-683.

[43]Wang F,Jiang M Q,Qian C,et al. Residual attention network for image classification[C] //Proceedings of the IEEE Conferenceon Computer Vision and Pattern Recognition (CVPR). Hawaii:IEEE,2017:3156-3164.

[44]Zhao Minghang,Zhong Shisheng,Fu Xuyun,et al．Deep residual shrinkage networks for fault diagnosis[J]． IEEE Transactions on Industrial Informatics,2019,16(7)：4681-4690．

[45]Hochreiter S,Schmidhuber J. Long short-term memory[J]. Neural Computation,1997, 9(8):1735-1780.

[46]Geert L,Thijs K,Babak E B,et al. A survey on deep learning in medical image analysis[J]. Medical Image Analysis,2017,42:60-80.

[47]陈聪,候磊,李乐乐,等.基于GRU改进RNN神经网络的飞机燃油流量预测[J].科学技术与工程,2021,21(27):11663-11673.

[48]杨韫韬,聂勇伟,张青,等.基于RNN和注意力机制的双向人体姿态补全方法[J/OL].计算机辅助设计与图形学学报,2022,34(11):1772-1783.

[49]Woo S,Park J,Lee J Y,et al. CBAM:convolutional block attention module[C].Proceedings of the European Conference on Computer Vision. Springer:ECCV,2018:3-19.

[50]赵立新,白银光,何春燕,等.基于双维度注意力机制的西红柿成熟度检测方法[J].科学技术与工程,2023,23(11):4571-4578.

[51]李晓鹏,王玫,周陬,等.一种基于蓝牙峰值纠正的改进PDR定位方法[J].计算机应用与软件,2022,39(06):40-45+60.

[52]Weinberg. Using the ADXL202 in pedometer and personal navigation applications[J]. Analog Devices AN-602 application note,2002,2(2):1-6.

[53]朱启举,樊振辉,梅春波,等.基于景象匹配的低精度MEMS航向修正与组合算法[J].中国惯性技术学报,2023,31(09):870-875.

[54]Wu W,Dasgupta S,Ramirez E,et al. Classification accuracies of physical activities using smartphone motion sensors[J]. Journal of Medical Internet Research,2012,14(5):130.

[55]王晓玲,赵岳生,罗宇彤.基于机器学习的人体活动识别[J].物联网技术,2023,13(12):3-6+11.

[56]刘斌,刘宏建,金笑天等.基于智能手机传感器的人体活动识别[J].计算机工程与应用,2016,52(04):188-193.

[57]李新科,刘欣雨,李勇明,等.基于惯导信息和卷积神经网络的人体活动识别[J].生物医学工程学杂志,2020,37(04):596-601.

[58]Mutegeki,R.,Han. A CNN-LSTM Approach to Human Activity Recognition. In Proceedings of the 2020 International Conference on Artificial Intelligence in Information and Communication(ICAIIC),Fukuoka,Japan,19–21February 2020,362–366.

[59]Nidhi D,Shiva N S,Vijay B S. Multi-input CNN-GRU based human activity recognition using wearable sensors[J]. Computing,2021,103:1461-1478.

[60]王培杰.基于TWR测距技术的UWB室内定位系统优化[D].西安:西安理工大学,2023.

[61]张培玟.基于RSSI测距的室内定位算法研究[D].西安:长安大学,2023.

[62]毛永毅,张颖.非视距传播环境下的AOA定位跟踪算法[J].计算机应用,2011, 31(02):317-319.

[63]董宇,逯暄,彭甫镕,等.基于TOA算法的UWB室内定位的定标方法[J].测控技术,2023, 42(11):47-52.

[64]李康乐.基于TOA的煤矿井下人员定位算法研究[D].西安:西安科技大学,2019.

[65]陆冰琳.室内环境下基于UWB的TDOA/AOA三维混合定位算法[J].电子元器件与信息技术,2022,6(01):159-163.

[66]张振宇.基于机器学习的LOS/NLOS传播路径识别技术研究[D].四川：西南交通大学, 2022.

[67]KE G,MENG Q,FINLEY T,et al. Lightgbm:A highly efficient gradient boosting decision tree[J]. Advances in neural information processing systems,2017,30.

[68]Machado M R,Karray S,Sousa I T D. LightGBM: an Effective Decision Tree Gradient Boosting Method to Predict Customer Loyalty in the Finance Industry[C]//International Conference on Computer Science and Education.2019.

[69]张哲,王勇.基于字词特征融合与BO-LightGBM的自动漏洞评估方法[J/OL].微电子学与计算机,2023,(07):27-35.

[70]孙伟,孙沛伦.一种基于机器学习分类器的UWB异常信号识别方法[J].测绘科学,2023,48(05):1-8+34.

[71]马陇飞,萧汉敏,陶敬伟,等.基于梯度提升决策树算法的岩性智能分类方法[J].油气地质与采收率,2022,29(01):21-29.

[72]袁驰,黄波.UWB定位中基于XGBoost的NLOS识别方法[J].移动通信,2023,47 (05):83-89.

[73]张红莉,李月琴,韩磊,等.基于LGBM和深度神经网络的HRRP目标识别方法[J].探测与控制学报,2022,44(02):97-103+114.

[74]苏佳,张晶晶,易卿武,等.基于CNN-CAM的NLOS/LOS识别方法研究[J].无线电工程,1-11[2024-06-04].

[75]蒋弘威,綦俊峰,刘雨聪,等.EKF的航姿参考系统自适应滤波算法[J].电子世界,2021,(19):41-42.

[76]李朝君,宋维,陈妍,等.基于CDF的核电厂可靠性指标分配方法研究[J].核科学与工程,2019,39(05):765-770.

[77]沈辉.基于UWB/PDR的井下定位技术研究[D].武汉,中国矿业大学,2021.