A novel dynamic outlier-robust Kalman filter with Moving Horizon Estimation

Area of Science:

• Control Systems Engineering
• Signal Processing
• Data Science

Background:

• Dynamic outliers present a significant challenge to the performance of standard Kalman filters (KF).
• Accurate state estimation is crucial in various dynamic systems, but is often compromised by unexpected data deviations.

Purpose of the Study:

• To develop an innovative filtering approach that enhances robustness against dynamic outliers.
• To improve the accuracy and adaptability of state estimation in the presence of measurement anomalies.

Main Methods:

• Developed a method to identify state and measurement dynamic outliers by analyzing measurement information.
• Modeled noise using a Gaussian-Student's t mixture distribution (GSTM) with parameters inferred via variational Bayesian (VB) methods.
• Integrated the GSTM noise model into the Moving Horizon Estimation (MHE) framework.
• Optimized estimation accuracy by determining the optimal window size through simulation experiments.

Main Results:

• The proposed filter significantly improves the system's capacity to adapt to dynamic changes.
• The integrated GSTM noise model enhances the flexibility and accuracy of noise modeling within the MHE framework.
• Simulation results confirm the filter's superior robustness in resisting dynamic outliers compared to existing filters.

Conclusions:

• The proposed filter effectively addresses the challenge of dynamic outliers in KF applications.
• The combination of advanced outlier detection, flexible noise modeling, and MHE provides enhanced state estimation performance.
• This approach offers a more reliable solution for systems requiring accurate state estimation under noisy and anomalous conditions.