【目的】为了解决煤矿井下人员精确定位中的超宽带信号在井下容易受到非视距(NLOS)的干扰严重影响定位精度的问题。【方法】提出一种基于自训练抑制NLOS的井下人员定位方法,该方法设计了一个新型的通用数据融合框架。首先,PDR结合地图信息去除不可行的位置,采用基于多粒度网格滤波器联合估计位置和航向生成弱标签。其次,通过多传感器数据融合对弱标签迭代改进,生成训练样本,实现自主收集训练数据。最后将地图、惯性传感器和超宽带测量的数据采用贝叶斯估计进行数据融合去推断位置。【结果】通过在井下环境中模拟测试,结果表明,对于复杂的井下场景,NLOS条件下的均方根误差由原来的1.02下降到0.32m,测距误差改善了69%,定位误差小于0.3m的定位结果可以从49%提高到89%,证明了所提出的井下人员定位方法的有效性。

【Purposes】The research of precise location of underground personnel in coal mine is  of great significance to protect their life safety. The ultra-wideband signal is susceptible to non-line- of-sight (NLOS) interference, which will seriously affects the positioning accuracy. 【Methods】In or⁃ der to solve the problem that the existing supervised learning methods for NLOS identification and  suppression require long time, labor intensive feature, and high cost became of the needs to obtain  training data and label allocation, a method for underground personnel positioning based on self- training and suppression of NLOS is proposed,and a new general data fusion framework is designed.  First, PDR and map information are combined to remove infeasible positions, and multi-granularity  mesh filters are used to estimate the position and heading, and the map information is fully utilized to  generate weak labels. Second,through multi-sensor data fusion, the weak label is iteratively im⁃ proved, and training samples are generated to realize autonomous collection of training data. Finally,  the data of map, inertial sensor, and ultra-wideband measurement are fused by Bayesian estimation to  infer the location. 【Findings】Through the simulation tests in the downhole environment, the results  show that for complex downhole scenes, the root-mean-square error of NLOS decreases from the  original 1.02 to 0.32 m, the ranging error is improved by 69%, and the positioning result with the po⁃ sitioning error less than 0.3 m can be increased from 49% to 89%. Thus, the effectiveness of the pro⁃ posed method for locating underground personnel is proved.