Space-Air-Ground Integrated Mobile Crowdsensing for Partially Observable Data Collection by Multi-Scale Convolutional...

Area of Science:

• Computer Science
• Artificial Intelligence
• Network Engineering

Background:

• Mobile crowdsensing (MCS) is a growing paradigm for large-scale data collection.
• Unmanned Aerial Vehicles (UAVs) are key components in Space-Air-Ground Integrated Networks (SAGIN).
• Existing research lacks exploration of multi-task MCS problems involving UAV swarms and satellites.

Purpose of the Study:

• To address the challenges of energy-efficient data collection and recharging in SAGIN using UAV swarms.
• To propose a novel deep reinforcement learning (DRL) approach for the Space-Air-Ground integrated Mobile CrowdSensing (SAG-MCS) problem.
• To enhance multi-agent cooperation and performance under partial observability.

Main Methods:

• Development of the Multi-Scale Soft Deep Recurrent Graph Network (ms-SDRGN) utilizing a multi-scale convolutional encoder.
• Implementation of a graph attention mechanism for inter-UAV communication modeling and information aggregation.
• Application of a gated recurrent unit for long-term performance and a maximum-entropy method for stochastic policy learning.

Main Results:

• The proposed ms-SDRGN method significantly outperforms three state-of-the-art DRL baselines in the SAG-MCS problem.
• ms-SDRGN achieved a 29.0% improvement in reward and a 3.8% increase in CFE score compared to the best baseline.
• The study demonstrates the scalability and robustness of ms-SDRGN in diverse DRL environments.

Conclusions:

• The ms-SDRGN approach provides an effective solution for complex multi-agent MCS tasks in SAGIN.
• The heuristic reward function successfully encourages global cooperation among UAV agents.
• The findings highlight the potential of DRL for optimizing resource management in integrated space-air-ground networks.