Road disease identification model training method and system based on federated learning and multi-center data collaboration

By coordinating the collaborative execution of nodes and terminal devices, dynamically adjusting the group participation probability, personalized hyperparameter assignment, and adaptive scaling random coding compression processing, the problems of model bias and large communication overhead in multi-center road defect identification are solved, and efficient and secure model training is achieved.

CN121413808BActive Publication Date: 2026-04-10安徽交控工程集团有限公司
View PDF 2 Cites 0 Cited by

Patent Information

Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
安徽交控工程集团有限公司
Filing Date
2025-11-04
Publication Date
2026-04-10

AI Technical Summary

Technical Problem

Existing technologies for multi-center road defect identification suffer from problems such as uneven data distribution, heterogeneous computing and communication capabilities leading to model bias, low training efficiency, and high communication overhead, and lack effective federated learning solutions.

Method used

The method of coordinating nodes and multiple terminal devices to perform collaborative execution includes initializing a global model, dynamically adjusting the probability of group participation, personalized assignment of hyperparameters, parameter compression based on distribution characteristics, and adaptive scaling random encoding compression processing, which optimizes resource efficiency and ensures privacy and security.

Benefits of technology

It improves model accuracy and fairness, optimizes system resource efficiency, enhances privacy protection capabilities, and addresses the challenges posed by data distribution differences and heterogeneous devices.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN121413808B_ABST
    Figure CN121413808B_ABST
Patent Text Reader

Abstract

The application discloses a road disease identification model training method and system based on federal learning and multi-center data cooperation. The method is cooperatively executed by a coordination node and a plurality of terminal devices, and comprises the following steps: initializing a global model and distributing; performing multi-round iterative training, each round of iteration comprising: the coordination node performs distribution-aware node selection based on data distribution characteristics, and dynamically assigns personalized hyperparameters to the selected devices; the terminal device trains the model using local data, and uploads the model update after compression based on the distribution characteristics; the coordination node reconstructs the update, performs weighted fusion based on the correlation weight between devices to update the global model, and finally performs adaptive scaling and coding compression on the global model update and distributes it through a dynamic caching mechanism. The application can effectively utilize scattered multi-center data, improve the training efficiency and recognition accuracy of the road disease identification model under the premise of ensuring data privacy.
Need to check novelty before this filing date? Find Prior Art

Citation Information

Patent Citations

  • Federal learning model training method applied to road condition prediction and road condition prediction method

    CN115422820A

  • Asynchronous federated learning gradient dynamic compression method in Internet of Vehicles

    CN117440342A