Research On Wireless Network Fingerprint Matching Positioning Algorithm Based On Machine Learning

With the rapid development of electronic technology and Internet of Things,Location Based Service has became a new wave of research in the new age.GPS and Beidou navigation and positioning technology can provide continuous,high-precision outdoor positioning services,including vehicle tracking and navigation,and pedestrian navigation.However,in an indoor environment,because satellite signals are affected by non-line-of-sight propagation,indoor positioning accuracy cannot meet positioning requirements,so there is a lot of room for development of indoor positioning technology.Because i Beacon nodes have the characteristics of low power consumption,easy deployment,and fast signal transmission frequency.At the same time,smart phones,tablets and other devices support this function.Therefore,this paper focuses on the i Beacon positioning technology.By consulting the relevant references of indoor positioning,this paper analyzes and introduces common indoor positioning technologies and algorithms in detail,fully studies the commonly used fingerprint matching and positioning algorithms based on machine learning,applies principal component analysis,affinity propagation clustering,BP neural network and weighted K Nearest Neighbors algorithm for fingerprint matching and positioning.The main research work of this paper is as follows:(1)The indoor positioning system based on i Beacon is implemented.Android application is responsible for collecting i Beacon information,data encapsulation,and data transmission.The background server analyzes and processes the client data,executes the positioning algorithm,and feeds back the results to the client for display.(2)Aiming at the problem that the i Beacon fluctuates and has noisy signals within a certain range,the i Beacon signal value is collected at a fixed reference point for a fixed number of times,and the signal distribution is found to conform to the Gaussian distribution.A scheme to obtain the real RSSI signal by merging Gaussian filtering and Kalman filtering is determined.(3)In view of the large workload of wireless fingerprint information collection in the stage of offline fingerprint library establishment,this paper uses Kriging interpolation method to interpolate a certain number of reference points to establish a perfect fingerprint library.(4)Based on the low positioning accuracy and large amount of calculation of the traditional machine learning fingerprint matching algorithm,a combination algorithm based on BP neural network and WKNN position algorithm is proposed.Firstly,the principal component analysis is performed on the offline fingerprint database to obtain its transformation matrix.Then the fingerprint database is subjected to affinity propagation clustering algorithm and neural network training.In the positioning phase,PCA conversion is performed on the signal vector,the sub-region to which the signal vector belongs is determined,and the WKNN algorithm is used to predict the position.Then the BP neural network coordinate prediction is performed on the signal vector.Finally,the coordinates of the above two are weighted and averaged to obtain the final position coordinates.The experimental results show that the use of Kriging interpolation method to interpolate the signal in the unknown area can effectively reduce the signal collection work of the offline fingerprint library.At the same time,the principal component analysis of the offline fingerprint library can appropriately improve the positioning accuracy and affinity propagation clustering algorithm can reduce the calculation amount of fingerprint matching and improve the efficiency of fingerprint matching.The positioning accuracy of combined positioning algorithm is better than that of single positioning algorithm.