Reinforcement Learning Based Topology Control for UAV Networks.

Area of Science:

• Computer Science
• Electrical Engineering
• Network Engineering

Background:

• Unmanned Aerial Vehicle (UAV) swarms offer rapid, low-cost wireless network deployment in areas lacking infrastructure.
• Scalability challenges arise in large-scale drone networks concerning multi-hop connectivity and control.
• Efficient network topology control is crucial for managing complex, dynamic drone network conditions.

Purpose of the Study:

• To propose an adaptive topology control system for drone networks.
• To optimize UAV connectivity by considering interference and energy consumption.
• To enhance the scalability and efficiency of drone-based wireless networks.

Main Methods:

• Developed a topology control system analyzing UAV relative positions.
• Implemented a Deep Deterministic Policy Gradient (DDPG) reinforcement learning approach for adaptive connectivity optimization.
• Dynamically adjusted learning parameters to minimize convergence time during UAV deployment.

Main Results:

• The proposed system effectively optimizes network topology by selecting neighbors and mapping data flows.
• Connectivity optimization adaptively responds to changing network conditions like user density and UAV power constraints.
• Simulation experiments and theoretical analysis validated the system's performance across various topologies.

Conclusions:

• The DDPG-based reinforcement learning approach provides an effective solution for dynamic topology control in drone networks.
• The system enhances network performance by balancing interference and energy efficiency.
• This research contributes to the development of scalable and robust drone-based wireless communication systems.