Particle filtering of dynamical networks: Highlighting observability issues

Area of Science:

• Complex Systems
• Network Science
• Dynamical Systems

Background:

• Measuring every node in high-dimensional networks is impractical.
• Optimal sensor placement is key for reliable system state reconstruction.
• Observability is a critical challenge in network analysis.

Purpose of the Study:

• To propose a particle filtering (PF) framework for assessing network observability.
• To investigate the impact of network dynamics and topology on observability.
• To compare PF with existing nonlinear observability metrics.

Main Methods:

• Developed a particle filtering (PF) framework to evaluate network observability.
• Applied the PF framework to benchmark networks: Kuramoto and Rössler oscillators.
• Analyzed the influence of heterogeneous nodal dynamics and sensor node selection.

Main Results:

• Conjectured that sensor nodes with dynamical affinity to neighbors enhance observability in heterogeneous networks.
• Demonstrated that the choice of measured variable affects PF performance.
• Showcased PF effectiveness against established nonlinear observability metrics.

Conclusions:

• Particle filtering provides a robust method for assessing network observability.
• Dynamical and topological properties significantly influence network state reconstruction.
• The proposed PF framework offers insights into optimal sensor placement strategies.