首页|井下动态环境基于DAE的XGBoost自适应定位算法研究

井下动态环境基于DAE的XGBoost自适应定位算法研究

扫码查看
原文链接
  • 万方数据
  • 维普
针对煤矿井下高动态环境导致WiFi定位模型的精度降低的问题,提出极端梯度提升(XGBoost)的指纹定位算法,利用其高维数据特征的学习能力完成定位.与传统的梯度提升树(GBDT)算法相比,在完成更好定位效果的同时,速度也大大提升.同时针对WiFi数据的波动性和XGBoost算法面对动态环境模型漂移问题,分别提出融合降噪自编码器(DAE)和自适应机制的D-XGBoost算法和Z-XGBoost算法.实验结果表明:XGBoost算法的定位精度比GBDT算法提高了,效率提高了5 倍多.融合DAE的D-XG-Boost算法的定位准确率比XGBoost算法提高了17%;融合了自适应机制的Z-XGBoost算法有效降低了模型漂移造成的误差.所提改进算法更好地改善了WiFi定位模型精度降低和模型漂移问题.
Research on DAE-based XGBoost self-adaptive localization algorithm for dynamic downhole environment
Aiming at the problems that highly dynamic environment in underground coal mines causes precision of WiFi localization model to decrease,extreme gradient boosting(XGBoost)fingerprint localization algorithm is proposed to accomplish localization using its learning ability of high-dimensional data features.Compared with the traditional gradient boosting tree(GBDT)algorithm,the speed is greatly improved while accomplishing better localization results.Meanwhile,aiming at the problem of volatility of WiFi data and the drift of XGBoost algorithm facing dynamic environment model,the D-XGBoost algorithm and Z-XGBoost algorithm,which integrate denoising auto-encoder(DAE)and self-adaptive mechanism,are proposed respectively.Experimental results show that the localization precision of XGBoost algorithm is improved,compared with the GBDT algorithm and the efficiency is increased by more than five times.The localization accuracy of the D-XGBoost algorithm fusing DAE is improved by 17%compared with the XGBoost algorithm;the Z-XGBoost algorithm fusing the self-adaptive mechanism effectively reduces the error caused by model drift.The proposed improved algorithm better improves the problem of WiFi precision degradation of localization model and model drift.

extreme gradient boosting(XGBoost)downhole fingerprint localizationmodel driftnoise-reduction self-encodererror compensation

洪金祥、崔丽珍、窦占树

展开 >

内蒙古科技大学信息工程学院,内蒙古包头 014010

极端梯度提升 井下指纹定位 模型漂移 降噪自编码器 误差补偿

内蒙古自治区科技计划资助项目内蒙古自治区科技计划资助项目内蒙古自然科学基金资助项目国家自然科学基金资助项目

2019GG3282022YFSH00512020MS0602762261042

2024

10.13873/J.1000-9787(2024)10-0023-04

传感器与微系统
中国电子科技集团公司第四十九研究所

传感器与微系统

CSTPCD北大核心
影响因子：0.61
ISSN：1000-9787
年,卷(期)：2024.43(10)