Contaminative Data-Driven Koopman Resilient Distributed Filtering for Unknown Stochastic Nonlinear Systems

Area of Science:

• Control Systems
• Nonlinear Dynamics
• Data Science

Background:

• Koopman operators are limited in handling unknown stochastic dynamics.
• Noisy measurements and process/measurement noise degrade filtering performance.

Purpose of the Study:

• To develop a Koopman-enhanced distributed filtering method for unknown stochastic nonlinear systems with contaminated datasets.
• To address limitations of conventional Koopman operators in handling stochastic dynamics and noise.

Main Methods:

• Utilized delay-coordinate embedding to reconstruct observations from noisy measurements.
• Developed robust subspace dynamic mode decomposition (SDMD) for reliable Koopman operator identification.
• Designed a distributed filtering scheme with an adaptive event-triggered mechanism.

Main Results:

• The proposed method effectively reconstructs system dynamics from noisy data.
• Robust Koopman operator identification suppresses noise effects.
• Adaptive event-triggered mechanism balances network burden and estimation accuracy.

Conclusions:

• The Koopman-enhanced distributed filtering approach is effective and robust for unknown stochastic nonlinear systems.
• The method demonstrates improved performance in the presence of contaminated datasets and noise.