Comparative Study of Kalman Filtering and Particle Filtering in Sensor Localization Systems

Abstract

In many data-driven applications in engineering and scientific fields, accurately locating sensor nodes is crucial for the reliability and performance of wireless sensor networks. This paper systematically compares the performance of two excellent state estimation algorithms and studies their application in sensor localization. This study used real datasets and synthetic datasets to evaluate various metrics. These metrics include convergence speed, root mean square error, mean absolute error, robustness to noise and node failures, computational cost, and scalability. Experimental results show that both methods achieved sub-meter accuracy in low-noise environments. In high-noise, multipath, and irregular sensor data conditions, particle filters are generally more effective than Kalman filters. Compared to the Kalman filter, particle filters can reduce localization errors by 15% to 30% in high-noise environments. Even with more than 40% of the sensor nodes failing, they can still operate normally. The aforementioned advantages come with higher computational costs and memory usage, making them unsuitable for all-weather system design. This study conducted an in-depth analysis of the advantages and disadvantages of each algorithm and demonstrated the application scenarios of complex sensor networks. This study provides a reference point for selecting suitable decentralized estimation methods for industrial and scientific applications.