Indoor Localization Technology Of WLAN Location Fingerprint Based On Machine Learning

With the rapid development of communication technology,location-based service plays an important role in people's life and work,which makes location technology,especially indoor localization,receive more and more attention.With the popularity of WLAN(wireless local area network)and the extensive use of intelligent devices,WLAN Indoor Positioning Technology Based on location fingerprint,which has strong practicability,low cost and high positioning accuracy,has attracted more and more attention.However,there are two key problems that restrict the development of location fingerprint indoor location.First,the pre-built fingerprint database cannot update the changed fingerprint data.Because the fingerprint database needs to be collected and constructed in the experimental environment in advance,and when the location fingerprint data changes due to factors such as the movement of indoor objects and the flow of people,it will increase the positioning error to continue to use the original fingerprint database;secondly,the location accuracy and real-time of the target to be tested are not high,which is difficult to meet the needs of production and life.In view of the above problems,a better indoor localization technology model is proposed.Thespecific work of this paper is as follows:(1)In this paper,a clustering method based on the fusion of self-organizing map clustering and Kalman filtering is proposed to build a dynamic fingerprint database.The main tasks are as follows: 1)updating fingerprint data.For the repeated fingerprints in the fingerprint data,data fusion and update are carried out according to the data characteristics of the last time;2)dynamic clustering of fingerprint data.Based on the existing training results,unsupervised learning is carried out for the updated fingerprint data,and the categories of fingerprint data are updated.(2)The indoor localization technology based on the improved limit gradient lifting algorithm is proposed.The improved algorithm model solves the problem of class imbalance caused by training massive data sets.When the number of class samples is too large or too small,sample feature extraction will lead to biased learning of large sample class,ignoring the characteristics of small sample class.Firstly,the loss contribution mechanism is proposed to simulate the loss of samples and the loss contribution weight mechanism to calculate the probability of accurate sample classification;finally,according to the loss contribution weight,the feature weight of small samples is increased to improve the learning and attention of small samples and improve the classification effect of the algorithm.In this paper,a clustering framework of dynamic fingerprint database and a high-precision indoor localization framework are proposed.The two frameworksare improved by clustering from fingerprint database and location estimation of points to be measured.At the same time,each frame carries out algorithm verification,and the two positioning frames are jointly applied in a group of experiments to verify the positioning effect.The results show that the positioning accuracy of the combined positioning frame within 2m reaches 93%.