Remote Estimator Design for Time-Delay Neural Networks Using Communication State Information.

Area of Science:

• Control Systems Engineering
• Artificial Intelligence
• Networked Systems

Background:

• Neural networks are susceptible to performance degradation due to distributed delays and imperfect measurements.
• Packet dropouts in transmission channels introduce significant uncertainty in networked control systems.

Purpose of the Study:

• To design a robust estimator for neural networks considering neuron-dependent nonlinearities and channel packet dropouts.
• To develop a channel-state-dependent estimator that balances estimator complexity and performance.
• To guarantee stochastic stability and strict dissipativity for the augmented system.

Main Methods:

• Utilized neuron-dependent nonlinearity to model uncertain measurements.
• Employed Markov chains and an augmented chain to model packet dropouts.
• Defined a channel state variable for designing a channel-state-dependent estimator.
• Derived estimator gains using linear matrix inequality methods.

Main Results:

• Established sufficient conditions for stochastic stability and strict $(Q, S, R)-\gamma -$ dissipativity.
• Demonstrated the effectiveness of the developed estimator through a practical example.
• The channel-state-dependent approach allows for a trade-off between estimator performance and computational load.

Conclusions:

• The proposed estimator design effectively addresses challenges posed by distributed delays, uncertain nonlinearities, and packet dropouts in neural networks.
• The developed methods provide a robust framework for analyzing and designing estimators in complex networked systems.
• The results confirm the practical applicability and superior performance of the channel-state-dependent estimator.