基于邻域门控注意力机制的Wi-Fi指纹室内定位算法研究

打开文本图片集

中图分类号：TP393 文献标识码：A文章编号：1006-8228（2026）02-17-07

Abstract：Inresponsetotheproblemsofhigh-dimensionalredundancy，susceptibilitytosuddenfluctuationsandlocaloclusion interference，anddificultyinalaningglobalandlocaldetailsithasinglebranchstructureintradioalWiifingeprintid positioningmethods，anLFG-ConvNetmodelisproposedanddesigned.ThismodelusesLDA（LinearDiscriminantAnalysis）to performdimensionalityreductiononhigh-dimensionalRSSIfeatures，efectivelyremovingredundantinformationandsuppresing noiseinterference；itintroducesneighborhood-focusedatentionanddynamicgatingatentionmechanismstoenhancetheextraction ofeffectivelocalfeaturesandsuppressielevant interferenc，therebyimprovingheobustessofthodeltlocalocusionand suddennoise；themodeladoptsanetworkstructurewithparalelfusionoftheatentionbranchandtheconvolutionbranchto achievethecollborativeutlizationofglobalfeaturesandlocaldetails.TheexperimentalresultsbasedonEOW-3regionalWiFi fingerprintdatafromtheUniversityofVictoriainCanadashowthattheLFG-ConvNetmodelperformsexcelentlyinboth positioning accuracy and stability： the average positioning error （MDE） is 1.8296m ，the 80% positioning error （P8o） and 90% （20 positioning error （P90） are 2.9035m and 4.3658m ，respectively，and the cumulative distribution function （CDF） curve of the overall positioning errorissignificantlyshiftedtotheleftfurtherverifying theeffctivenessandsuperiorityofthemodel.

Keywords：Wi-Fi Fingerprint Positioning;Atention Mechanism; Indoor Positioning;Deep Learning

0引言

室内定位技术在精确定位实现过程中面临诸多挑战。（剩余8768字）