Research On Fingerprinting Positioning Technologybased On Machine Learning

With the development of wireless communication technology and popularity of intelligent terminals, the era of mobile Internet has arrived.In this process, Location Based Service (LBS) is constantly enriching people's lives, changing people's living habits, and showing great commercial value. In order to promote the development of related industries and improve the quality of location services, it is of great value to do research both on the indoor and outdoor positioning technology.The Cellular network is widely covered, and it has a huge amount of users. Therefore, positioning technology based on Cellular network becomes a good general positioning methodology.In order to solve the problems encountered in the positioning scheme,the distribution of the crowdsourcing data in cellular network is analyzed and modeled in this thesis. Then, based on the fingerprinting matching algorithm, several generalized localization algorithms are proposed. And a variety of machine learning models are adopted in this thesis to improve the accuracy and efficiency of the algorithm.In this thesis, the traditional localization methods used in current cellular networks are described, and the characteristics of each algorithm are analyzed. The algorithms commonly used in fingerprinting localization technology are specially introduced. Then, several kinds of machine learning algorithms applied by the scholars in the positioning system in recent years are introduced. The advantages and application scenarios of each algorithm are analyzed. Then, the combination of base stations (CBS) received in MR is considered as an important factor in location progress, and an algorithm based on maximal similarity in CBS is designed. Compared with the weighted K-nearest neighbor algorithm,the proposed algorithm gets higher accuracy. In order to improve the efficiency of location algorithm, a multi-partitioned decision model based on logistic regression is proposed, which also reduces the accuracy of positioning. Then, by analyzing the actual distribution of crowdsourcing data, this thesis proposed the function of minimal error expectation,which has strong anti-noise ability and shows great performance in simulation experiment. In the last chapter, a hybrid model and a trajectory-information based model are designed. For all MR in the continuous trajectory, a node with minimum outlier in the trajectory is selected as the anchor node to help locate the MR. The result of Experiment shows that the hybrid model is effective to improve the positioning accuracy, while the trajectory-information based model achieves higher positioning accuracy.