Improved Indoor positioning algorithm using KPCA and ELM

Abstract

RSS (Received Signal Strength) values which are used in indoor position estimation based on fingerprinting are affected by noise. The RSS value received by the fixed line-of-sight condition points obeys the Gauss distribution and does not match the Gauss distribution. WIFI signal transmission attenuation is also a nonlinear attenuation. This paper presents a joint KPCA-ELM locating algorithm, the use of KPCA (Kernel Principal Component Analysis) of the nonlinear characteristics allow the original RSS data being replaced and dimension reduction, constructing new features. ELM (Extreme Learning Machine) is a fast and efficient single-layer feedforward neural network algorithm for training with new characteristics. KPCA-ELM positioning algorithm can effectively reduce the influence of noise on RSS value and improve the accuracy. The experimental results show that KPCA-ELM algorithm can effectively improve the accuracy of indoor positioning.