基于梯度提升回归树的井下定位算法

为了提高井下定位系统的定位精度，提出了基于梯度提升回归树(gradient boost regression tree, GBRT)的井下定位算法。本文首先介绍了GBRT算法的实现过程，然后利用射线追踪算法模拟井下多径信号叠加后的接收信号强度(received signal strength, RSS)数据集，最后对比了GBRT、K最近邻(k-nearest neighbor, KNN)、随机森林(random forest, RF)、支持向量机(support vector machine, SVM)和神经网络多层感知器(multi-layer perceptron regressor, MLPR)算法的定位结果并对GBRT的定位结果进行5点平均滤波。实验结果表明，在100个点组成的行人轨迹定位中，GBRT算法的定位结果的均方误差为0.381米，明显优于其他四种算法，平滑滤波后的定位轨迹更加贴合真实轨迹。因此，本算法可以有效提高定位精度，可以满足井下定位系统的精度要求。

A Coal Mine Underground Positioning Algorithm Based on Gradient Boost Regression Tree

In order to improve the positioning accuracy of the downhole positioning system, this paper proposes a downhole positioning algorithm based on the gradient boosting regression tree (GBRT). Firstly, this article introduced the implementation of the GBRT algorithm. Then, the ray tracing algorithm was used to simulate the superimposed RSS (received signal strength) data set of downhole multipath signals. Finally, the positioning results of GBRT, KNN (k-nearest neighbor), RF (random forest), SVM (support vector machine) and MLPR (multi-layer perceptron regressor) algorithms were compared and the 5-point average filtering of GBRT positioning results was performed. The experimental results show that in the pedestrian trajectory of 100 points, the mean square error of the positioning result of GBRT algorithm is 0.381 meters, which is obviously better than the other four algorithms. The filtered positioning trajectory is more in line with the real trajectory. Therefore, this algorithm can effectively improve the positioning accuracy and can meet the accuracy requirements of the downhole positioning system.