聚类阈值结合动态K值的蓝牙室内定位算法

Bluetooth indoor localization algorithm based on clustering threshold and EWKNN

针对固定阈值的动态K近邻算法定位时未能有效剔除距离较远参考点的问题,该文提出了基于聚类阈值结合动态K值的算法:①在离线阶段建立聚类指纹库,并在每个聚类子块中设定聚类阈值;②在线阶段根据待测点信号确定其所处的聚类子块和对应的阈值,由信号强度的欧氏距离和聚类阈值选取K个信号距离最小的参考点;③以信号距离倒数为权重计算坐标加权平均,作为定位结果。实验分析得出,与固定阈值的动态K值算法的平均定位误差为2.64m;聚类阈值结合动态K值算法的平均定位误差为1.12m,降低了57.6%,并且点位误差在1m和2m内的可信度分别为42.3%和77.8%。聚类阈值结合动态K值算法能够较好地剔除距离较远的参考点,可以有效提高蓝牙指纹定位的精度。

Aiming at the problem that the fixed K-nearest neighbor algorithm does not effectively eliminate the reference point far away from the fixed threshold,this paper proposes an algorithm based on the clustering threshold combined with the dynamic K value:First,the cluster fingerprint database is established in the offline phase,and The clustering threshold is set in the cluster sub-blocks;then,the online cluster determines the cluster sub-block and the corresponding threshold according to the signal to be measured,and selects K signals from the Euclidean distance of the signal strength and the clustering threshold.The minimum reference point is the distance;finally,the coordinate weighted average is calculated as the weight of the signal distance reciprocal as the positioning result.The experimental analysis shows that the average positioning error of the fixed threshold dynamic Kvalue algorithm is 2.64 m;the average positioning error of the clustering threshold combined with the dynamic Kvalue algorithm is 1.12 m,which is reduced by 57.6%,and the point error is within 1 mand 2 m.The credibility is 42.3%and 77.8%,respectively.The clustering threshold combined with the dynamic K-value algorithm can better eliminate the reference points far away,which can effectively improve the accuracy of Bluetooth fingerprint positioning.