Research On WiFi Fingerprint Positioning System In Indoor Environment

At present,with the rapid development of wireless communication and artificial intelligence technology,mobile intelligent terminals have increasingly become an indispensable part of people's daily life,making indoor location-based services more and more important.At the same time,the large-scale deployment of Access Points(AP)in indoor scenes has brought a broad development prospect for indoor positioning based on WiFi technology,thus attracting a large number of researchers at home and abroad to carry out research on WiFi location fingerprint positioning technology.However,there are many factors that interfere with WiFi signal propagation in indoor scenes,such as shadows and fading,which make the Received Signal Strength(RSS)appear to a certain degree of uncertainty,restricting the stability and accuracy of WiFi location fingerprint positioning.This also brings some challenges to the research of fingerprint positioning technology.In this thesis,the characteristics of RSS signal and its mapping relationship with physical location in a complex indoor environment are deeply analyzed,and on this basis,the problems that need to be faced and solved in the offline stage and online stage of WiFi location fingerprint positioning are summarized.Aiming at these problems,this thesis mainly does the following work:(1)In view of the large fluctuation range of the RSS value of the receiver caused by factors such as building shielding and pedestrian movement in indoor scenes,as well as the limited coverage ability of some AP signals and the weak ability of their fingerprint features to distinguish the physical location,this thesis proposes an AP selection algorithm based on optimal stability ability and position discrimination ability.Among the deployed AP nodes in the indoor environment,the most favorable AP for positioning was selected for fingerprint database construction.This algorithm firstly measures the stability of AP by using the variance and occurrence frequency of the receiving RSS signals,selects the APs with the strongest stability and the least environmental impact as the optimal stable AP subset,and then uses RelieF feature selection algorithm to calculate the weight coefficient of each AP feature of the RSS sample fingerprint.The APs with the largest weight coefficient,that is,the APs with the strongest position discrimination ability can generate the stable ability and the AP subset with the best position discrimination ability can be used to construct the offline fingerprint database.The experimental results show that,compared with the most typical AP selection algorithm,this algorithm can significantly improve the positioning accuracy and reduce the operation consumption of the system.(2)The initial clustering center of the traditional K-means clustering algorithm is randomly selected,which not only leads to outliers in the clustering results,but also may fall into the local optimal solution.To solve this problem,a K-means clustering algorithm based on density is proposed.The algorithm considers the distribution in the sample space,calculates the density of each sample point,and selects K samples with high density distribution in the sample set as the initial clustering center in turn.The experimental results show that the optimized K-means clustering algorithm has higher contour coefficient and smaller sum of error squares,and the clustering effect is more excellent.(3)A fingerprint location model based on improved RBF neural network is proposed.The model is designed to optimize the selection and initial parameters of the sample center in the network hidden layer for location prediction in the online stage.Firstly,K-means clustering algorithm based on density is used to determine the sample center and number of the hidden layer,and then the Beetle antennae search algorithm is introduced to continuously search and update the initial parameters of the network,so as to find the initial parameter value that makes the network performance reach the best,and improve the positioning accuracy of the model.The experimental results show that the optimized RBF neural network model has faster convergence speed,stronger generalization ability,and higher accuracy of position estimation in the online stage.