第32卷第3期测绘工程Vol.32No.32023年5月EngineeringofSurveyingandMappingMay2023DOI:10.19349/j.cnki.issn1006-7949.2023.03.003基于BP神经网络的WiFi/地磁定位方法杨朝永,赵冬青,贾晓雪,张乐添,赖路广,程振豪(信息工程大学,郑州450001)摘要:针对WiFi在室内定位中信号波动性较大、地磁指纹存在误匹配等问题,提出一种基于BP神经网络的WiFi/地磁融合定位方法。该方法通过Z-score标准化消除WiFi、地磁数据不同量纲级的影响,同时,选取Tanh函数替代Sigmoid函数作为BP神经网络的激活函数,改善深度学习中梯度消失、梯度爆炸等问题。离线阶段,将处理后的WiFi、地磁数据作为输入层对改进的神经网络进行学习训练,在线阶段,将训练好的BP神经网络用于智能手机的定位。实验结果表明,文中提出的定位方式较单一传感器的定位方式整体定位精度提升约为14%。关键词:室内定位;BP神经网络;WiFi;地磁中图分类号:P209;P228文献标识码:A文章编号:1006-7949(2023)03-0014-05AnlocalizationmethodbasedonBPneuralnetworkcombiningWiFiandgeomagnetismYANGChaoyong,ZHAODongqing,JIAXiaoxue,ZHANGLetian,LAILuguang,CHENGZhenhao(InformationEngineeringUniversity,Zhengzhou450001,China)Abstract:AimingattheproblemsofWiFisignalfluctuationinindoorpositioningandthemismatchofgeomagneticfingerprints,thispaperproposesaWiFi/geomagneticfusionpositioningmethodbasedonBPneuralnetwork.ThismethodeliminatestheinfluenceofdifferentdimensionlevelsofWiFiandgeomagneticdatathroughZ-scorestandardization.Atthesametime,theTanhfunctionisselectedtoreplacetheSigmoidfunctionastheactivationfunctionoftheBPneuralnetworktoimprovetheproblemsofgradientdisappearanceandgradientexplosionindeeplearning.Intheofflinestage,theprocessedWiFiandgeomagneticdataareusedastheinputlayertolearnandtraintheimprovedneuralnetwork.Intheonlinestage,thetrainedBPneuralnetworkisusedforsmartphonepositioning.Theexperimentalresultsshowthatthepositioningmethodproposedinthispaperimprovestheoverallpositioningaccuracybyabout14%comparedwiththepositioningmethodbasedonasinglesensor,andhasbetterpositioningperformance.Keywords:indoorpositioning;BPneuralnetwork;WiFi;geomagnetism收稿日期:2022-04-26基金项目:国家自然科学基金资助项目(41774037;4210403...
