Partial-Neurons-Based Passivity-Guaranteed State Estimation for Neural Networks With Randomly Occurring Time Delays

Area of Science:

• Control Systems Engineering
• Artificial Neural Networks
• Stochastic Systems

Background:

• State estimation is crucial for analyzing and controlling complex systems.
• Discrete-time artificial neural networks with time delays present significant challenges for state estimation.
• Partial availability of measurement outputs complicates traditional estimation techniques.

Purpose of the Study:

• To develop a passivity-guaranteed state estimation (SE) method for discrete-time artificial neural networks with randomly occurring time delays.
• To address SE problems where measurements are only available from a fraction of neurons.
• To ensure asymptotic stability in the mean square and a guaranteed passivity performance level for the estimation error dynamics.

Main Methods:

• Utilizing the Lyapunov-Krasovskii functional method.
• Employing stochastic analysis techniques.
• Characterizing random time delays using a Bernoulli-distributed random variable.

Main Results:

• Sufficient criteria for the existence of state estimators are derived.
• The estimation error dynamics are guaranteed to achieve asymptotic stability in the mean square.
• A guaranteed passivity performance level is achieved.
• The parameterization of the estimator gain is obtained via convex optimization.

Conclusions:

• The proposed method effectively addresses the state estimation problem for artificial neural networks with random time delays and partial measurements.
• The derived criteria ensure both stability and passivity performance.
• Numerical simulations validate the theoretical results and the effectiveness of the proposed state estimator.