Wireless Sensor-Based Indoor Localization With Kernel Learning Strengthened Optimization

The accuracy of an indoor localization system based on wireless sensor network becomes increasingly essential for the deployment of the Internet of Things (IoT). However, the performance of an indoor localization system is often affected by the complex non-Gaussian noise that is common in indoor environments. To overcome the above challenge, this article proposes a novel fractional-order derivative q-Laplace kernel mean p-power error (FrqLaKMPE) algorithm and provides a detailed convergence analysis. The proposed algorithm utilizes the robustness of the q-Laplace kernel learning strategy to mitigate the effects of complex non-Gaussian noise, while the fractional-order derivative effectively captures higher order statistical information, accurately reflecting data distribution and system characteristics. Furthermore, FrqLaKMPE demonstrates a faster rate of convergence and provides an interpretable solution. To validate the effectiveness of the proposed FrqLaKMPE, three representative indoor localization scenarios obtained by wireless sensors are used. Meanwhile, the comparison with prevalent positioning methods [adaptive filters (AFs), kernel AFs (KAFs), trilateration, Naive-Bayes, and K-nearest neighbor (KNN)] is also implemented. The experimental results indicate that FrqLaKMPE improves positioning accuracy by at least 5.7%, demonstrating good robustness and marking a significant advancement in this field.