Abstract: A feedforward neural network named Kernel Extreme Learning Machine is applied to the problem of indoor fingerprint positioning based on the nolinear location fingerprinting. The fast learning speed and tightness of the network can reduce algorithm offline learning time and improve its robustness. By changing the environment of offline data and adopting different Wi-Fi access points, localization performance of KELM is analysed. Under the same condition, experimental results suggest that KELM has stronger capability than ELM, SVM and kNN etc.

Key words: indoor positioning, location fingerprinting, Kernel Extreme Learning Machine（KELM）, Wi-Fi

摘要： 考虑到位置指纹的非线性特性，提出基于核极限学习机（KELM）的位置指纹定位方法。KELM以其快速学习的特点，同时拥有紧密的网络结构，有效解决传统定位算法离线学习时间长和鲁棒性差的问题。通过改变离线数据收集环境，采用不同Wi-Fi接入点作信号源来分析KELM算法的定位性能，实验结果表明，同等条件下与基本ELM、SVM和kNN等位置指纹定位方法相比，KELM表现出更好的定位能力。

关键词: 室内定位, 位置指纹, 核极限学习机, Wi-Fi