Data-Driven Control for Local Stabilization of Neural Networks Subject to Input Saturation: A Memory-Type...

Area of Science:

• Control Systems Engineering
• Artificial Intelligence
• Computational Neuroscience

Background:

• Discrete-time neural networks (DNNs) are crucial in various applications but face challenges with input saturation.
• Stabilizing DNNs often requires precise system models, which are difficult to obtain in practice.
• Event-triggered control mechanisms aim to reduce communication load but can lead to superfluous triggers.

Purpose of the Study:

• To develop a data-driven control method for local asymptotic stabilization of DNNs under input saturation.
• To design a memory-type event-triggered mechanism (MEM) that mitigates communication load.
• To establish a stabilization criterion and determine an estimated region of attraction (ERA) using only accessible data.

Main Methods:

• A memory-type event-triggered mechanism (MEM) was designed to reduce communication load.
• A memory-dependent Lyapunov function (MLF) was constructed to incorporate the MEM's memory term.
• A data-based stabilization criterion was developed using MEM, MLF, and data-based system representations.
• A hybrid optimization scheme combining linear objective minimization and particle swarm optimization (PSO) was employed to maximize the ERA.

Main Results:

• A data-based stabilization criterion was developed, relying solely on accessible data, eliminating the need for full system matrix knowledge.
• An estimated region of attraction (ERA) was determined, and its size was maximized using a hybrid optimization approach.
• The feedback gain and trigger matrix were co-designed to ensure closed-loop system stability.
• Numerical simulations validated the effectiveness of the optimization algorithm and the advantages of the MEM.

Conclusions:

• The proposed data-driven control method effectively stabilizes discrete-time neural networks under input saturation using accessible data.
• The memory-type event-triggered mechanism significantly reduces communication load while maintaining system stability.
• The approach is practical for real-world applications where precise system modeling is challenging.