Enabling mobility-aware federated stochastic gradient descent for decentralized swarm intelligence in internet of Vehicle-Things based UAV networks

Abstract

This study investigates the optimization of energy efficiency, transmission performance, and system reliability in Internet of Vehicle-Things (IoVT) Based UAV communication and machine learning environments. The purpose is to enhance operational efficiency and system resilience under varying conditions. The key problems addressed include high energy consumption, slow convergence rates, and increased error rates under noise interference. Using simulation-based methods, various parameters such as hovering cost, batch size, learning rates, noise power, and cooperation costs were analyzed. Results show a 33% reduction in loss with increased batch size, a 50% decrease in convergence rounds with higher learning rates, and a 25% increase in error rate due to noise interference. Energy consumption increased linearly with hovering cost, while transmission power rose from 5 dBm to 10 dBm as UAV count increased from 2 to 10. The IoVT-based UAV Aultopilot system (CPU) utilization varied directly with frequency, reaching 100% at 3 GHz. The findings contribute to formulating adaptive policies for UAV communication and learning systems by recommending dynamic adjustments of transmission power, batch size, and learning rate based on environmental factors and system demands. Future research should focus on real-world implementations and policy adjustments for dynamic scenarios to ensure sustainable, energy-efficient system performance.