Air-ground collaborative incremental federated learning method based on unmanned aerial vehicle position optimization

By employing an air-ground collaborative incremental federated learning method that optimizes drone location, important samples are selected and user frequency and drone location are optimized. This solves the problem of resource waste in dynamic environments and achieves efficient, energy-saving, and intelligent training results.

CN117933317BActive Publication Date: 2026-06-23NANJING UNIV OF AERONAUTICS & ASTRONAUTICS
View PDF 0 Cites 0 Cited by

Patent Information

Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
NANJING UNIV OF AERONAUTICS & ASTRONAUTICS
Filing Date
2024-01-24
Publication Date
2026-06-23

AI Technical Summary

Technical Problem

Existing federated learning algorithms are insufficient in selecting the importance of samples in dynamic environments, resulting in wasted computing resources and slow convergence speed for users with limited resources. Drone model aggregation servers need to dynamically optimize deployment strategies to cope with real-time sample updates.

Method used

We adopt an air-ground collaborative incremental federated learning method based on UAV location optimization. Through a model-agnostic meta-learning framework, we filter samples collected by users in real time to generate local personalized models. We derive the upper bound of the gradient norm of the filtered samples and combine user CPU frequency and UAV location optimization to form an optimization objective of energy consumption and training loss. This is decomposed into the optimization problems of sample selection, user frequency and UAV location. We iteratively optimize to minimize the trade-off between energy consumption and training loss.

Benefits of technology

It effectively saves energy consumption of resource-constrained equipment, improves training effectiveness, and achieves a highly efficient, energy-saving, and intelligent training process.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN117933317B_ABST
    Figure CN117933317B_ABST
Patent Text Reader

Abstract

The application provides an air-ground cooperative incremental federated learning method based on unmanned aerial vehicle position optimization, considers that unmanned aerial vehicles equipped with aggregation servers provide auxiliary communication and calculation services for ground users far away from base stations or under base station failure conditions, the unmanned aerial vehicles assist user aggregation models as air base stations to update obsolete models through incremental federated learning to cope with new samples continuously, and through joint optimization of the unmanned aerial vehicle position, user frequency and sample screening, the trade-off problem of minimizing energy consumption and training loss under the conditions of meeting the time delay and basic restrictions of decision variables is realized. The user in the application is a resource-limited edge device in the Internet of Things, and the method provided by the application has the characteristics of high efficiency, energy saving and intelligence, greatly saves the energy consumption of the resource-limited edge device in the Internet of Things system, and improves the training effect.
Need to check novelty before this filing date? Find Prior Art