Proportional-Integral Observer Design for Multidelayed Sensor-Saturated Recurrent Neural Networks: A Dynamic...

Area of Science:

• Control Theory
• Neural Networks
• Signal Processing

Background:

• Recurrent Neural Networks (RNNs) are susceptible to performance degradation due to time delays and sensor saturation.
• Existing observer designs often lack efficiency in resource utilization for complex network systems.

Purpose of the Study:

• To design a Proportional-Integral Observer (PIO) for discrete-time multidelayed RNNs with sensor saturation.
• To develop a dynamic event-triggered protocol for efficient data transmission.
• To ensure the boundedness of estimation errors in the mean-square sense.

Main Methods:

• Lyapunov-like approach for error boundedness analysis.
• Matrix inequality techniques for observer parameterization.
• Dynamic event-triggered control for adaptive data transmission.

Main Results:

• A general framework for analyzing estimation error boundedness was established.
• The ultimate bound of the error dynamics was determined.
• Explicit expressions for the PIO were derived using matrix inequalities.
• The proposed PIO demonstrated effectiveness and superiority in simulations.

Conclusions:

• The developed PIO effectively addresses challenges in discrete-time multidelayed RNNs with sensor saturation.
• The dynamic event-triggered protocol enhances resource utilization in observer-based systems.
• The proposed method offers a robust and efficient solution for state estimation in complex neural network architectures.