A Multi-Mode Switching Variational Bayesian Adaptive Kalman Filter Algorithm for the SINS/PNS/GMNS Navigation System of Pelagic Ships

Jie Zhang, Shanpeng Wang, Wenshuo Li, Zhenbing Qiu2022 · DOI

TL;DR

This paper introduces a novel multi-mode switching variational Bayesian adaptive Kalman filter (MMS-VBAKF) to enhance the reliability and accuracy of ship navigation systems (SINS/PNS/GMNS) in challenging pelagic environments. By adaptively estimating measurement noise and switching between operating modes based on interference levels, the algorithm significantly improves navigation performance under uncertain conditions.

What Problem Does This Paper Address?

Navigating pelagic ships (ships operating in the open ocean) presents significant challenges due to complex and unpredictable environmental conditions. Traditional navigation methods like Inertial Navigation Systems (INS) suffer from accumulating errors over time. Satellite navigation (like GPS) is susceptible to interference. Integrated systems combining Strap-down Inertial Navigation System (SINS), Polarization Navigation System (PNS), and Geomagnetic Navigation System (GMNS) offer a promising solution, but their performance is heavily impacted by environmental factors. Specifically, cloudy or hazy weather can degrade PNS accuracy, while the presence of magnetic anomalies can distort GMNS data. Existing methods often assume Gaussian noise, which is unrealistic in these harsh environments. This paper addresses the need for a robust navigation system that can adapt to varying levels of interference from PNS and GMNS sensors, ensuring reliable navigation even when one or both systems are compromised.

Methods

The core of this paper is the development of the Multi-Mode Switching Variational Bayesian Adaptive Kalman Filter (MMS-VBAKF). This algorithm builds upon the Variational Bayesian Adaptive Kalman Filter (VBAKF) by incorporating an interference evaluation and multi-mode switching mechanism.

Here's a breakdown of the key methodological components:

  • SINS/PNS/GMNS Integrated Navigation Model: The paper establishes a standard state-space model for the integrated navigation system, combining SINS, PNS, and GMNS.
  • Variational Bayesian Adaptive Kalman Filter (VBAKF): This forms the foundation for estimating system states and, crucially, adaptively estimating unknown and time-varying measurement noise statistics. It uses variational inference and mean-field theory to approximate the posterior distribution of parameters, allowing for real-time noise estimation.
  • Interference Evaluation Mechanism: This is a novel component designed to assess the quality of PNS and GMNS data.
  • For PNS: The Degree of Polarization (DoP) is used to quantify interference. A threshold-based system categorizes interference into three levels: slight (Case SI), tolerance (Case TI), and excessive (Case EI).
  • For GMNS: A parameter derived from the geomagnetic vector's deviation from the World Magnetic Model (WMM) is used to evaluate interference, also categorized into three levels: slight (Case SI), tolerance (Case TI), and excessive (Case EI).
  • Multi-Mode Switching: Based on the interference evaluation, the system dynamically switches its operating mode:
  • Case SI (Slight Interference): The VBAKF algorithm runs normally, adaptively estimating noise.
  • Case TI (Interference Tolerance): The system's measurement noise covariance is adaptively weighted, balancing current measurements with prior information.
  • Case EI (Excessive Interference): The affected sensor system (PNS or GMNS) is isolated, and the measurement model is reconstructed to rely on the remaining functional sensor(s). This ensures continued navigation even with a complete failure of one sensor type.

Datasets

Not discussed in the paper. The evaluation is performed using simulation experiments.

Key Insights and Ideas

  • Adaptive Noise Estimation is Crucial: The paper strongly emphasizes that real-world navigation sensors, especially in challenging environments, do not produce noise with constant statistical properties. The VBAKF's ability to estimate this time-varying noise in real-time is key to improving accuracy.
  • Interference is Multi-faceted: The work highlights that interference isn't just a binary "good" or "bad" but exists on a spectrum. The three-tiered interference classification (SI, TI, EI) allows for a more nuanced and effective response.
  • Mode Switching is Essential for Robustness: The ability to gracefully degrade performance or switch to alternative sensors when one system fails (Case EI) is critical for maintaining continuous and reliable navigation in pelagic environments. This is a significant improvement over systems that might fail entirely.
  • PNS and GMNS Complementarity: The paper leverages the strengths of both PNS (sky-based) and GMNS (Earth's magnetic field-based) to create a more resilient system, recognizing that environmental factors can affect each differently.

Key Results

The paper demonstrates the effectiveness of the MMS-VBAKF algorithm through simulation experiments under three scenarios:

  1. Random Unknown Noise: Compared to a standard Kalman Filter (KF), the MMS-VBAKF showed significant improvements in accuracy. For example, the RMSE for pitch angle estimation was reduced by 26.9%, roll by 77.0%, and heading by 78.2%.
  2. Periodic Sinusoidal Noise: In the presence of periodic noise (simulating ship sway), the MMS-VBAKF again outperformed the KF, exhibiting better anti-sway characteristics. RMSE improvements were 29.2% for pitch, 78.8% for roll, and 64.1% for heading.
  3. Partial System Failure (Interference Cases): This is where the multi-mode switching truly shines.
  4. When the polarization sensor was shielded (Case 2, 140-160s), the standard KF experienced large attitude errors (>10°). The MMS-VBAKF, by isolating the PNS, maintained good accuracy.
  5. When strong magnetic interference occurred (Case 4, 600-630s), the standard KF showed heading errors exceeding 20°. The MMS-VBAKF, by isolating the GMNS, again maintained robust performance.
  6. The paper reports that for the multi-mode VBAKF, the RMSE for roll angle and heading angle was significantly improved by 40.8% and 49.8% respectively, compared to a multi-mode KF, in these interference scenarios.

Limitations and Future Work

  • Simulation-Based: The results are based on simulations. While comprehensive, real-world testing would be the next logical step to validate performance under actual oceanic conditions.
  • Specific Interference Types: The paper focuses on interference affecting PNS and GMNS. Future work could explore interference affecting SINS or other potential sensor modalities.
  • Computational Complexity: While VBAKF is adaptive, variational inference can sometimes be computationally intensive. The paper doesn't explicitly detail the computational overhead compared to simpler filters, which might be a consideration for real-time implementation on resource-constrained platforms.
  • Parameter Tuning: The thresholds for interference detection (Thol, Thmag) are mentioned as needing to be determined by actual accuracy requirements. Optimal tuning of these parameters for various operational scenarios could be an area for further research.

Summary updated

BibTeX
@misc{zhang2022multi,
  author = {Jie Zhang and Shanpeng Wang and Wenshuo Li and Zhenbing Qiu},
  title = {{A Multi-Mode Switching Variational Bayesian Adaptive Kalman Filter Algorithm for the SINS/PNS/GMNS Navigation System of Pelagic Ships}},
  year = {2022},
  doi = {10.3390/s22093372},
}
Figure