摘要
文章针对当前室内5G多变量信号接收强度指纹变化的不确定性导致定位误差较大等问题,提出一种改进的基于集成学习模型的5G指纹室内定位方法。离线阶段,对5G辅同步信号中的四个接收测量值进行预处理,建立指纹数据库,将K最邻近、朴素贝叶斯、多层感知机算法作为弱分类器进行训练;在线阶段,使用各个弱分类器的约登指数分别来加权对应算法的预测概率,形成强分类器,使用加权概率质心法解算出定位坐标。结果表明,本方法与K最邻近、朴素贝叶斯、多层感知机算法相比,在视距场景下,平均定位误差分别降低了25.94%、32.53%、35.62%,非视距场景下,平均定位误差分别降低了28.79%、11.32%、25%。
In view of the uncertainty of the current indoor 5G multivariate signal received signal strength fingerprint changes,resulting in large positioning errors and other problems,an improved 5G fingerprint indoor positioning method based on an integrated learning model is proposed.In the offline phase,it preprocesses the four received measurement values in the 5G secondary synchronization signal,establishes a fingerprint database,and trains K-Nearest Neighbors,Naive Bayes,multi-layer perceptron algorithm as the basic weak classifier;In the online phase,the Youden index of each weak classifier is used to weight the predicted probability of the corresponding algorithm to form a strong classifier.The weighted probability centroid method is used to calculate the positioning coordinates.The result shows that in line of sight scenarios,compared with K-Nearest Neighbors,Naive Bayes and multi-layer perceptron algorithms,this method reduces the average positioning error by 25.94%,32.53%,and 35.62%.In non line of sight scenarios,the average positioning error is reduced by 28.79%,11.32%,and 25%.
作者
李亚捷
蒋振伟
赵昆
曾毅
Li Yajie;Jiang Zhenwei;Zhao Kun;Zeng Yi(Shanghai Key Laboratory of Multidimensional Information Processing,East China Normal University,Shanghai 200241,China;China United Network Communications Co.,Ltd.,Shanghai Branch,Shanghai 200082,China)
出处
《信息通信技术》
2024年第5期71-76,84,共7页
Information and communications Technologies
基金
上海市科委资助项目(22DZ2229004)。
