Indoor Positioning Algorithm Based on Nonlinear PLS Integrated With RVM

Indoor positioning based on received signal strength indicator of WLAN has received more and more attention because of low cost and easy implementation. However, traditional localization algorithms often fail to achieve better positioning results because of multi-path effect and shadow effect. In order to solve the problem of multi-collinearity and more noise in WLAN indoor location data, this paper presents a novel nonlinear partial least square (PLS) method to address the problem of low precision in WLAN location. The proposed method integrates an inner relevant vector machine (RVM) function with an external linear PLS framework. First, the localization area is divided into a number of small areas by K-means algorithm. Then, PLS is applied to extract the features of the fingerprint database to reduce the number of the variable dimensions and eliminate the correlations. The obtained score matrices are used as the input and output of RVM. Finally, the coordinates of test points are regressed and predicted by the RVM-PLS algorithm. Simulation and experiments in real scenario prove the effectiveness of the proposed method. Compared with SVM-PLS, RBF-PLS, SVM-PCA, EBQPLS, PLS, SVM, RBF, RVM, and WKNN algorithm, the experimental results show that the proposed algorithm has higher positioning accuracy.