A calibration method of triaxial magnetometers based on an improved PSO algorithm

Purpose This study aims to develop an accurate and robust calibration method for triaxial magnetometers by introducing a reinforcement learning-driven adaptive particle swarm optimization (RLAPSO) framework. Design/methodology/approach A RLAPSO algorithm is proposed for triaxial magnetometer error calibration. Within the conventional particle swarm optimization (PSO) framework, a Sobol low-discrepancy sequence-based initialization strategy is employed to improve population diversity, an outlier detection-driven inertia weight adjustment mechanism is introduced to dynamically balance exploration and exploitation and a reinforcement learning-based learning factor regulation strategy is integrated to enable adaptive parameter tuning. These mechanisms collaboratively enhance global search capability, convergence speed and solution stability. Findings Simulation experimental results demonstrate that RLAPSO achieves stable and accurate calibration under different noise conditions and significantly outperforms several comparative algorithms in terms of convergence speed and parameter estimation accuracy. Furthermore, the algorithm maintains consistent optimization efficiency and solution reliability across different population sizes and search space settings, further validating its applicability and robustness for high-precision calibration tasks in complex interference environments. Originality/value This work proposes a reinforcement learning-driven adaptive PSO framework that integrates low-discrepancy initialization and dynamic parameter regulation, providing an effective and robust solution for high-precision triaxial magnetometer calibration.