UKF-based remote state estimation for discrete artificial neural networks with communication bandwidth constraints

Area of Science:

• Control Engineering
• Artificial Intelligence
• Networked Systems

Background:

• Remote state estimation is crucial for discrete neural networks operating under communication constraints.
• Limited transmission bandwidth necessitates strategies for handling partial measurement data.
• Nonlinear dynamics in neural networks pose challenges for traditional estimation methods.

Purpose of the Study:

• To design a robust remote state estimator for discrete neural networks with communication bandwidth limitations.
• To address the challenge of transmitting only partial measurement components due to bandwidth constraints.
• To ensure the stability and performance of the state estimation process.

Main Methods:

• Development of a Unscented Kalman Filter (UKF)-based state estimator.
• Incorporation of techniques to handle nonlinear activation functions in neural networks.
• Analysis of estimator stability under communication constraints.

Main Results:

• The proposed UKF-based estimator effectively handles nonlinearities and bandwidth limitations.
• Sufficient conditions for exponential boundedness of the state estimation error in mean square were established.
• A numerical example demonstrated the practical effectiveness of the developed method.

Conclusions:

• The developed remote state estimator provides a viable solution for discrete neural networks with communication constraints.
• The stability analysis confirms the reliability of the estimation under partial data transmission.
• The UKF-based approach offers a robust method for state estimation in networked neural systems.