Neural network optimal control for tripartite UAV confrontation systems based on fuzzy differential game

Area of Science:

• Control Systems Engineering
• Artificial Intelligence
• Game Theory

Background:

• Unmanned aerial vehicle (UAV) systems are increasingly complex, involving multiple autonomous agents with competing objectives.
• Traditional control strategies struggle with the dynamic and adversarial nature of multi-agent UAV confrontations.
• Differential game theory and fuzzy logic offer potential frameworks for optimizing decision-making in such systems.

Purpose of the Study:

• To propose a novel optimal control strategy for tripartite UAV confrontation systems (attackers, defenders, targets).
• To address the complexity of direct solution for the tripartite differential game model.
• To ensure the stability and convergence of the proposed control strategy.

Main Methods:

• Construction of a tripartite UAV mutual confrontation model and a nonlinear differential control system.
• Application of fuzzy evaluation and differential game theory to decompose the game into attacker-defender and attacker-target sub-games.
• Utilizing an evaluation neural network with adaptive dynamic programming to approximate the optimal value function.

Main Results:

• Derivation of separate optimal control strategies for attackers, defenders, and targets.
• Successful approximation of the optimal value function using a neural network.
• Proof of convergence for neural network weights and stability of the control system via Lyapunov theory.

Conclusions:

• The proposed fuzzy differential game-based neural network optimal control strategy is effective for tripartite UAV confrontation.
• The method provides a robust framework for decentralized control in complex multi-agent adversarial scenarios.
• Simulation results validate the effectiveness and stability of the designed control strategy.