Fully-Distributed Neural-Network-Based Approaches for Monotonic Game With Finite-Time Disturbance Rejection

Area of Science:

• Control Theory
• Game Theory
• Artificial Intelligence

Background:

• Variational generalized Nash equilibrium (vGNE) seeking is crucial for multi-agent systems with coupling constraints.
• Existing methods often require parameter predesign and lack robustness against disturbances.
• Dynamical players introduce complexities in achieving equilibrium in general monotonic games.

Purpose of the Study:

• To develop a distributed neural network for seeking vGNE in general monotonic games with multiple coupling constraints.
• To address challenges posed by high-order dynamics and parameter predesign requirements.
• To enhance the robustness and full distribution of the neural network against external disturbances.

Main Methods:

• Designed a distributed vGNE-seeking neural network (vGSNN) using a high-pass filter to convert high-order dynamics to second-order.
• Proposed an adaptive weight controller to remove the need for fixed parameter predesign, enabling full distribution.
• Incorporated a sliding-mode controller for finite-time disturbance rejection and maintained full distribution.

Main Results:

• The proposed vGSNN successfully transforms high-order dynamics into manageable second-order dynamics.
• Adaptive weights eliminated the need for parameter predesign, achieving full distribution of the network.
• The sliding-mode controller ensured finite-time disturbance rejection, enhancing robustness.
• Effectiveness was validated through an uncrewed aerial vehicle (UAV) swarm game simulation.

Conclusions:

• The developed vGSNN offers an effective, distributed solution for vGNE seeking in complex monotonic games.
• The adaptive and sliding-mode control strategies enhance robustness and reduce design complexity.
• The approach is validated in a practical scenario, demonstrating its applicability to real-world multi-agent systems.