Track prediction method based on distributed training and differential privacy

Through a distributed training architecture among vehicles, roadside units, and central servers, combined with dual attention and adaptive differential privacy algorithms, the adaptability and privacy protection issues of distributed trajectory prediction models in complex traffic environments are solved, achieving efficient trajectory prediction and data security.

CN120681168APending Publication Date: 2025-09-23SOUTHEAST UNIV
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202510789111.3
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2025-06-13
Publication Date
2025-09-23

AI Technical Summary

Technical Problem

In the existing technology, the distributed trajectory prediction model has poor adaptability in actual application scenarios, does not fully consider the trajectory prediction scenarios of single vehicles and multiple vehicles, and lacks privacy protection measures, resulting in data privacy leakage and high communication overhead.

Method used

A personalized federated aggregation algorithm based on dual attention and an adaptive differential privacy algorithm are adopted to collaboratively train the trajectory prediction model among vehicles, roadside units, and central servers through a distributed training architecture. Noise addition is used to protect data privacy, and model parameter transmission is optimized according to scenario characteristics.

Benefits of technology

It improves the adaptability and accuracy of the model in complex traffic environments, reduces communication overhead, ensures data privacy and security, and improves global model performance.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN120681168A_ABST
    Figure CN120681168A_ABST
Patent Text Reader

Abstract

The invention discloses a trajectory prediction method based on distributed training and differential privacy, and the method comprises the steps: firstly, building a trajectory prediction basic model which achieves the function of predicting the driving trajectory of a target vehicle in different scenes based on a single-vehicle trajectory prediction model and a multi-vehicle trajectory prediction model; then building a distributed model training architecture to realize model cooperative training among the vehicle, the road side unit and the central server; the road side units serve as clients in federated learning, a local track prediction model is trained based on the shared track data, model parameters are sent to a central server, the central server calls a personalized federated aggregation algorithm based on double attention to aggregate the model parameters of all the road side units, and the model parameters of all the road side units are obtained. Obtaining a global trajectory prediction model; and finally, a privacy protection algorithm is used in the transmission process of the trajectory data and the model parameters, and a privacy security guarantee is provided for the transmission process of the data and the parameters.
Need to check novelty before this filing date? Find Prior Art