WiFi Indoor Localization Based On Machine Learining

Internet applications such as online map,uber car,take out,i.e. have high demands to positioning accuracy. GPS satellite positioning system is not able to achieve positioning of indoor terminals,so it's hightly promoted to research on the indoor positioning system based on WiFi in recent years. There are many ways to implement WiFi positioning system,and fingerprint database two-step localization method is commonly used.This method can be divided into two steps: offine fingerprint database instruction and online matching process. The method has high positioning accuracy, but the computation process is complex. Machine learning algorithms are widely used in data processing and data mining,this paper will plug machine learning algorithm into WiFi positioning system fingerprint database management, which can reduce the computational complexity and improve the positioning accuracy.First, this paper introduces the development situation of several kinds of indoor positioning technology and analyzes their advantages and disadvantages respectively. The realization of WiFi indoor positioning technology: fingerprint database two-step localization method is mainly introduced. In this paper, MATLAB is used to simulate positioning area,in which we deploy reference points to collect RSSI fingerprints and match with the pending site to estimate its location with optimized matching learning algorithm.Second, in order to reduce the computational burden of the positioning stage, clustering method in machine learning is introduced to manage fingerprint database. While applying position match process after the management of fingerprint database, the system will first look for the cluster center which lies nearest to the pending site and then traverse all other reference points that belongs to this cluster. Kernel fuzzy c-means method can divide fingerprint data into k clusters by minimizing the objective function, but the cluster amount k and the choice of kernel function can be optimized instead of setting randomly.First, this paper uses sample density method to determine the value of optimal cluster amount k, second, the optimal parameters of the kernel function is obtained using approximate ideal Kernel matrix method. Then,with Iris data set, the optimized k value and kernel parameters is verified that can effectively improve the clustering performance.Next, on-site experiment is conducted in the second school building in BUPT and the WiFi positioning system is constructed. This paper analyzes the influence of the amount of access points as well as the distribution of reference points on positioning accuracy. Fingerprint database is organized by the intelligence-optimized fuzzy C-means clustering algorithm, and then calculate the root-mean-square error of each positioning experiment, test results shown that our proposed algorithms effectively reduce the computation load of positioning process of and achieve better performance than that of the existing KFCM algorithms with 23.55% improvement in localization accuracy.