A Performance-Based Hybrid PSO-GA Framework for Optimal Process Noise Covariance Tuning in INS/GNSS Integration | ||
| مهندسی مکانیک مدرس | ||
| Articles in Press, Accepted Manuscript, Available Online from 18 December 2025 | ||
| Document Type: مقاله پژوهشی | ||
| DOI: 10.48311/mme.2025.117363.82879 | ||
| Authors | ||
| محمد حسن نجارنصب1; حسن سالاریه* 2 | ||
| 1دانشکده مهندسی مکانیک، دانشگاه صنعتی شریف، تهران، ایران | ||
| 2دانشیار، دانشکده مکانیک، دانشگاه شریف، تهران، ایران | ||
| Abstract | ||
| Accurate state estimation in INS/GNSS integrated navigation is critical, and its most common implementation, the Extended Kalman Filter (EKF), is highly dependent on the tuning of its process noise covariance matrix (Q). Conventional tuning methods, such as those based on stationary Allan-variance (AV) analysis, often fail to capture sensor behavior under real-world dynamic conditions, leading to filter inconsistency, sub-optimal accuracy, and poor dynamic response, such as overshoot upon GNSS re-acquisition. This paper proposes a novel, data-driven offline framework to optimize Q. Our approach parameterizes the continuous-time covariance (Qc) using four physically meaningful scalars and tunes them using a hybrid Particle Swarm Optimization and Genetic Algorithm (PSO-GA). The optimization minimizes a performance-based objective function: the mean trajectory-wide Root Mean Square Error (RMSE) evaluated over a comprehensive set of sliding, pure-INS windows. Experimental validation on a real-world dataset demonstrates that the proposed tuning framework significantly outperforms conventional Allan-variance based tuning. Specifically, it reduces the mean terminal position and attitude errors by 52% and 84%, respectively, while simultaneously tightening the estimated error bounds by 59% and 82%, indicating superior filter consistency and robustness. Critically, in a simulated GNSS outage-and-recovery scenario, the proposed filter exhibited rapid, stable convergence without any overshoot, a significant improvement over the AV-tuned filter which suffered from severe overshoots. By directly linking the Q matrix parameters to observed navigation performance, this work provides a practical and robust methodology for EKF tuning. | ||
| Keywords | ||
| INS/GNSS integration; Extended Kalman Filter (EKF); Process noise covariance tuning; Hybrid PSO-GA; low-cost IMU | ||
|
Statistics Article View: 99 |
||
| Number of Journals | 45 |
| Number of Issues | 2,171 |
| Number of Articles | 24,674 |
| Article View | 24,436,091 |
| PDF Download | 17,551,308 |