Alliance-based Internet of Vehicles federal learning method, system and device

By employing an adaptive alliance and lightweight gradient computation-based federated learning method for vehicle-to-everything (V2X) networks, the problems of dynamic changes in vehicle data distribution and resource constraints are addressed. This approach enables high-precision personalized modeling and privacy protection, adapts to dynamic V2X environments, and improves model prediction accuracy and system robustness.

CN121935758AActive Publication Date: 2026-04-28QUFU NORMAL UNIV
18 Cites 0 Cited by

Patent Information

Application Number
CN202610406347.9
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-03-31
Publication Date
2026-04-28
Estimated Expiration
2046-03-31

AI Technical Summary

Technical Problem

Existing federated learning methods for vehicle networks struggle to accurately characterize the distribution features of data categories when the distribution of vehicle data is dynamically changing and unknown, resulting in poor clustering performance. Furthermore, the computational process for resource-constrained vehicle nodes is complex, increasing communication overhead and failing to meet the requirements for privacy protection and high reliability.

Method used

We adopt a federated learning method for vehicle networking based on alliances. By initializing the global model, we perform local model parameter updates and gradient calculations to form an adaptive alliance. We use auxiliary datasets to estimate the class distribution, realize lightweight gradient calculation and weighted aggregation, and dynamically reorganize the alliance to adapt to changes in data distribution.

Benefits of technology

It enables high-precision personalized modeling in resource-constrained environments, protects data privacy, adapts to dynamic vehicle networking environments, reduces computational and communication burdens, and improves model prediction accuracy and system robustness.

✦ Generated by Eureka AI based on patent content.
Patent Text Reader

Abstract

The invention belongs to the technical field of Internet of Vehicles federated learning, and particularly relates to an alliance-based Internet of Vehicles federated learning method, system and device. The method comprises the following steps: firstly, converting convergence local model parameters uploaded by each vehicle into proportion vectors representing local data category distribution by using a small-scale balance auxiliary data set at a base station side; the distribution similarity among the vehicles is measured based on the divergence, the vehicles with similar distribution are clustered to form a plurality of alliances in a self-adaptive manner through a preset divergence threshold, and the number of the alliances does not need to be preset; and finally, vehicles in each alliance cooperatively train an exclusive alliance model. According to the method, the problem of model generalization under data isomerism is effectively solved, the prediction precision and convergence speed of each vehicle end model are remarkably improved while the privacy of the original data is guaranteed, and the method is suitable for an Internet of Vehicles environment.
Need to check novelty before this filing date? Find Prior Art