一种BP神经网络的室内定位WiFi标定方法
|
摘要
针对不同型号智能手机间WiFi软硬件异构导致的对同一AP信号源的观测量存在偏差,最终影响定位精度的问题,提出一种使用BP神经网络的WiFi标定方法:使用离群点检测算法剔除不同手机RSSI数据对中的离群点,获得相对纯净的数据对输入到BP神经网络进行训练;并对网络各层的权值和偏向值进行反复更新,使得输出值逼近真实值;当输出层误差的平方和小于阈值时则训练完成,保存各层的权值和偏向值就可得到较为稳定的网络标定模型,利用该模型可对不同型号手机的观测量进行校正。实验结果表明,该标定方法的定位精度比标定前可以提高39.72%,有效降低手机软硬件异构对定位精度的影响。
Aiming at the problem that the observational deviation of the same AP signal source caused by the isomerism of WiFi software and hardware between different types of smart phones ultimately affects the positioning accuracy,the paper proposed a WiFi calibration method using BP neural network:outlier detection algorithm was used to eliminate the outliers in different mobile phones' RSSI data pairs,relatively pure data were obtained to train in BP neural network,and the weights and bias values of each layer of the network were updated repeatedly so that the output values were approximated to the true ones;the training was completed when the sum of squared errors of the output layer was less than the threshold value,then the weights and bias of each layer were saved to get a stable network calibration model by which the measurements of different types of mobile phones were calibrated.Experimental result showed that the positioning accuracy of the proposed method could be improved 39.72%compared with that of non-calibrated positioning,illustrating that the method would help efficiently reduce the impact of the isomerism of smart phones' software and hardware on the positioning accuracy.
Aiming at the problem that the observational deviation of the same AP signal source caused by the isomerism of WiFi software and hardware between different types of smart phones ultimately affects the positioning accuracy,the paper proposed a WiFi calibration method using BP neural network:outlier detection algorithm was used to eliminate the outliers in different mobile phones' RSSI data pairs,relatively pure data were obtained to train in BP neural network,and the weights and bias values of each layer of the network were updated repeatedly so that the output values were approximated to the true ones;the training was completed when the sum of squared errors of the output layer was less than the threshold value,then the weights and bias of each layer were saved to get a stable network calibration model by which the measurements of different types of mobile phones were calibrated.Experimental result showed that the positioning accuracy of the proposed method could be improved 39.72%compared with that of non-calibrated positioning,illustrating that the method would help efficiently reduce the impact of the isomerism of smart phones' software and hardware on the positioning accuracy.
引文
[1]裴凌,刘东辉,钱久超.室内定位技术与应用综述[J].导航定位与授时,2017,4(3):1-10.
[2]阮陵,张翎,许越,等.室内定位:分类、方法与应用综述[J].地理信息世界,2015,22(2):8-14,30.
[3]雷雨伟,修春娣,杨威,等.WiFi与MEMS-IMU融合定位方法在室内定位中的应用[J].导航定位学报,2016,4(4):81-87.
[4] BLUMROSEN G,HOD B,ANKER T,et al.Enhanced calibration technique for rssi-based ranging in body area networks[J].Ad Hoc Networks,2013,11(1):555-569.
[5] SHIMOSAKA M,SAISHO O.Efficient calibration for RSSI-based indoor localization by bayesian experimental design on multi-task classification[EB/OL].[2018-03-18].http://delivery.acm.org/10.1145/2980000/2971710/p244-shimosaka.pdf?ip=65.49.14.55&id=2971710&acc=OA&key=4D4702B0C3E38B35%2E4D4702B0C3E38B35%2E4D4702B0C3E38B35%2E40F8DA9B0674C863&__acm__=1537947776_1b755117d550c800fa54bca72e39fb69.
[6] DONG B,BURGESS T,NEUNER H B.Unsupervised database free RSSI calibration indoor navigation[C]//The Institute of Electrical and Electronic Engineers(IEEE).Proceedings of 2017European Navigation Conference(ENC).Geneva:IEEE,2017:198-205.
[7]程杰.基于欧式距离平均值的离群点自动检测算法[J].电脑知识与技术,2015,11(7):240-242.
[8]刘晓晨,张静.基于改进BP神经网络的室内无线定位方法[J].计算机应用与软件,2016,33(6):114-117.
[9]张凌.室内环境监测中的BP神经网络算法的改进及其仿真分析[D].南京:南京邮电大学,2017.
[10]焦李成,杨淑媛,刘芳,等.神经网络七十年:回顾与展望[J].计算机学报,2016,39(8):1697-1716.
[11]张宏,马岩,李勇,等.基于遗传BP神经网络的核桃破裂功预测模型[J].农业工程学报,2014,30(18):78-84.
[2]阮陵,张翎,许越,等.室内定位:分类、方法与应用综述[J].地理信息世界,2015,22(2):8-14,30.
[3]雷雨伟,修春娣,杨威,等.WiFi与MEMS-IMU融合定位方法在室内定位中的应用[J].导航定位学报,2016,4(4):81-87.
[4] BLUMROSEN G,HOD B,ANKER T,et al.Enhanced calibration technique for rssi-based ranging in body area networks[J].Ad Hoc Networks,2013,11(1):555-569.
[5] SHIMOSAKA M,SAISHO O.Efficient calibration for RSSI-based indoor localization by bayesian experimental design on multi-task classification[EB/OL].[2018-03-18].http://delivery.acm.org/10.1145/2980000/2971710/p244-shimosaka.pdf?ip=65.49.14.55&id=2971710&acc=OA&key=4D4702B0C3E38B35%2E4D4702B0C3E38B35%2E4D4702B0C3E38B35%2E40F8DA9B0674C863&__acm__=1537947776_1b755117d550c800fa54bca72e39fb69.
[6] DONG B,BURGESS T,NEUNER H B.Unsupervised database free RSSI calibration indoor navigation[C]//The Institute of Electrical and Electronic Engineers(IEEE).Proceedings of 2017European Navigation Conference(ENC).Geneva:IEEE,2017:198-205.
[7]程杰.基于欧式距离平均值的离群点自动检测算法[J].电脑知识与技术,2015,11(7):240-242.
[8]刘晓晨,张静.基于改进BP神经网络的室内无线定位方法[J].计算机应用与软件,2016,33(6):114-117.
[9]张凌.室内环境监测中的BP神经网络算法的改进及其仿真分析[D].南京:南京邮电大学,2017.
[10]焦李成,杨淑媛,刘芳,等.神经网络七十年:回顾与展望[J].计算机学报,2016,39(8):1697-1716.
[11]张宏,马岩,李勇,等.基于遗传BP神经网络的核桃破裂功预测模型[J].农业工程学报,2014,30(18):78-84.