Asynchronous Fault Detection for Memristive Neural Networks With Dwell-Time-Based Communication Protocol

Area of Science:

• Control Systems Engineering
• Networked Systems Security
• Computational Neuroscience

Background:

• Discrete-time memristive neural networks are susceptible to network-induced phenomena.
• Stochastic communication protocols (SCP) and denial-of-service (DoS) attacks disrupt data transmission.
• Asynchronous mode mismatches between sensors and filters complicate fault detection.

Purpose of the Study:

• To design an asynchronous fault detection filter for discrete-time memristive neural networks.
• To address challenges posed by stochastic communication protocols and denial-of-service attacks.
• To ensure the stochastic stability of the neural network system under these conditions.

Main Methods:

• A dwell-time-based stochastic communication protocol is employed to coordinate sensor-filter packet transmission.
• Denial-of-service attacks are modeled using a Bernoulli distribution to capture their random occurrence.
• A hidden Markov model (HMM) is utilized to represent the asynchronous nature of the fault detection filter.

Main Results:

• Sufficient conditions for the stochastic stability of the memristive neural networks are derived using Lyapunov theory.
• A compensation strategy is developed to model both the SCP and DoS attacks.
• The proposed method effectively handles mode mismatches between data transmission and the filter.

Conclusions:

• The developed theoretical framework provides a robust solution for asynchronous fault detection in memristive neural networks.
• The effectiveness of the proposed method is validated through a numerical example.
• This work contributes to enhancing the security and reliability of networked neural network systems.