Multi-Sensor Optimal Data Fusion Based on the Adaptive Fading Unscented Kalman Filter

Area of Science:

• Control Engineering
• Signal Processing
• Estimation Theory

Background:

• Multi-sensor systems often face challenges with nonlinear dynamics and stochastic noise.
• Process-modeling errors can significantly degrade the performance of traditional data fusion techniques.
• Accurate state estimation is crucial for applications like integrated navigation.

Purpose of the Study:

• To develop an optimal data fusion methodology for multi-sensor nonlinear stochastic systems.
• To enhance the adaptability and robustness of state estimation against process-modeling errors.
• To achieve globally optimal state estimation using a two-level fusion structure.

Main Methods:

• Implementation of an adaptive fading unscented Kalman filter (AFUKF) with Mahalanobis distance for local filters.
• Development of a two-level fusion structure with local AFUKFs and a top-level unscented transformation-based fusion.
• Application of the principle of linear minimum variance for optimal fusion of local estimations.

Main Results:

• The proposed methodology effectively mitigates the influence of process-modeling errors on data fusion.
• Improved adaptability and robustness in state estimation for multi-sensor nonlinear stochastic systems were achieved.
• Globally optimal fusion results were obtained, validated by simulations and experiments.

Conclusions:

• The novel data fusion methodology provides superior performance for multi-sensor nonlinear stochastic systems.
• The approach demonstrates significant improvements in adaptability, robustness, and accuracy.
• The method is validated for integrated navigation systems, including Inertial Navigation System/Global Navigation Satellite System/Celestial Navigation System (INS/GNSS/CNS).