A traffic state prediction method fusing graph convolutional network and large language model
By integrating graph convolutional networks with large language models, this approach addresses the challenges of insufficient modeling and cross-modal adaptation in long-term traffic state prediction, achieving high-precision long-term traffic state prediction and medium- to long-term traffic decision support.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- CHINA UNIV OF MINING & TECH
- Filing Date
- 2026-05-11
- Publication Date
- 2026-07-17
AI Technical Summary
Existing traffic condition prediction methods are insufficient in long-term time-series dependency modeling, have low prediction accuracy, are difficult to adapt large models across modes, and have high fine-tuning costs, making it difficult to meet the needs of medium- and long-term traffic decision-making.
We employ a method that integrates graph convolutional networks and large language models. By combining hybrid graph convolutional spatial feature extraction, TCN temporal feature extraction, discrete-time feature encoding, and adaptive residual fusion, along with learnable soft prompts and LoRA low-rank adaptation techniques, we achieve long-term traffic state prediction.
It improves the accuracy and stability of long-term traffic forecasting, achieves a comprehensive and accurate representation of road network spatial correlation, reduces the cross-modal adaptation cost of large models, and supports medium- and long-term traffic control decisions.
Smart Images
Figure CN122416733A_ABST