An intelligent traffic flow prediction method and system based on multi-objective automatic federated learning
The intelligent traffic flow prediction method based on multi-objective automatic federated learning automatically searches for the optimal combination of lightweight network modules and hyperparameters, solving the problem of time-consuming and labor-intensive traditional model design and the trade-off between prediction accuracy and hardware cost, thus achieving efficient and safe traffic flow prediction.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- DONGHUA UNIV
- Filing Date
- 2026-06-03
- Publication Date
- 2026-07-17
AI Technical Summary
Existing federated learning-based traffic flow prediction schemes struggle to automate and optimize traffic flow prediction in multiple dimensions under complex spatiotemporal evolution patterns. Furthermore, traditional convolutional architectures face a trade-off between prediction accuracy and hardware overhead when deployed at the edge, resulting in time-consuming and labor-intensive model design that is difficult to adapt to complex and ever-changing traffic flow scenarios.
An intelligent traffic flow prediction method based on multi-objective automatic federated learning is adopted. Through an offline multi-objective automatic federated learning optimization module and an online real-time prediction module, the NSGA-II algorithm is used to automatically search for the optimal combination of lightweight network modules and hyperparameters. By combining five lightweight neural network modules, the model can be automatically optimized and trained efficiently.
While protecting data privacy and security, it achieves high-precision traffic flow prediction, while significantly reducing model complexity and communication costs, and improving system operating efficiency.
Smart Images
Figure CN122416735A_ABST