Abstract
Wireless sensor network node positioning faces challenges when dealing with limited data. Traditional methods rely heavily on either data-driven approaches or artificially designed content, making achieving accurate localization with sparse data difficult. This paper addresses the challenge of precise node localization in wireless sensor networks by introducing a novel approach that combines human cognition with machine learning techniques. Specifically, it aims to improve localization accuracy in scenarios with limited data by integrating Voronoi diagram division and a hybrid regression model. The proposed method involves a two-stage process: offline preparation and online testing. In the offline stage, the Voronoi diagram division is utilized to segment the localization space, reducing the need for manual intervention. The hybrid regression model, termed HWR-SKR, combines Support Vector Regression (SVR) and K-nearest neighbors Regression (KNR) to leverage the strengths of both algorithms. Training and testing are conducted based on RSS (Received Signal Strength) data, incorporating anchor node coordinates and Voronoi cell vertex coordinates. Experimental results demonstrate the effectiveness of the proposed approach in achieving precise node localization with limited data. The HWR-SKR hybrid regression model outperforms individual SVR and KNR models, offering improved accuracy and robustness in real-time positioning tasks. By integrating human cognition through Voronoi diagram division and machine learning techniques via the HWR-SKR hybrid regression model, this study presents a promising solution for accurate node localization in wireless sensor networks, particularly in scenarios with sparse data. The approach offers a practical method for improving localization performance, contributing to advancements in sensor network applications requiring precise spatial awareness.
