基于自适应K值选择的K近邻算法研究  

K Nearest Neighbor Algorithm Based on Adaptive K Value Selection

作　　者：闫中亚 汪云甲[1,2] 刘克强[1,2] 王行风[1,2] 

机构地区：[1]中国矿业大学国土环境与灾害监测国家测绘地理信息局重点实验室,江苏徐州221116 [2]中国矿业大学环境与测绘学院,江苏徐州221116

出　　处：《测绘地理信息》2016年第6期25-29,共5页Journal of Geomatics

基　　金：国家高技术研究发展计划(863计划)资助项目(2013AA12A201);面向地理国情服务的资源开发地表沉降监测资助项目(201412016);2014江苏省普通高校研究生科研创新计划资助项目(KYLX_1394);江苏高校优势学科建设工程"测绘科学与技术学科"资助项目(SZBF2011-6-B35)~~

摘　　要：针对室内定位中基于位置指纹的K近邻法采用固态K值无法得到最优定位结果的问题,提出自适应K值选择的K近邻法。算法利用相邻定位点短时间间隔内空间位置变化引起的信号强度变化规律推测运动趋势,并与不同K值的定位结果构建的空间矢量进行匹配,从而自适应地从K近邻法的不同K值中选取最优的K值。同时依据室内AP的几何布局特征划分多个矢量域内,并对定位结果进行区域改正。试验结果表明,该算法能够很好地抑制较大误差的出现,提高定位的实时性、定位精度和稳定性。The K nearest neighbor algorithm of location- based fingerprint with solid K value will not get the best result. Hence, a new K nearest neighbor algorithm through a daptive K value selection is proposed to resolve this issue. Algorithm uses the discipline that the change of signal strength during short intervals to speculate movement trend. And it matches with the space vector that constructed by the positioning results of different K. Thereby, the best K value can be adaptively chose from the different K values of K nea-rest neighbor algorithm. At the same time, based on the geometric layout features of indoor＇s AP , testing ground is divided into multiple vector filed, and correct positioning results by vector field. Experimental results show that the algorithm can suppress the occurrence of a large error, and iraprove real-time positioning, positioning accuracy and stability.

关 键 词：室内定位 K近邻算法 自适应K值选择 矢量域 空间关系 

分 类 号：P228.42[天文地球—大地测量学与测量工程]