Abnormal event-oriented multi-modal primitive learning traffic prediction method, system and device, and storage medium

By using a multimodal primitive learning method, combining multi-source data and prior information about future events, dynamic edge weight components and time-varying adjacency matrices are generated. This solves the problems of prediction accuracy and spatial correlation offset in existing traffic prediction models under abnormal scenarios, and enables reliable prediction of traffic conditions in complex environments.

CN122090622AActive Publication Date: 2026-05-26NANJING HYDRAULIC RES INST
7 Cites 1 Cited by

Patent Information

Application Number
CN202610549428.4
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-04-23
Publication Date
2026-05-26
Estimated Expiration
2046-04-23

AI Technical Summary

Technical Problem

Existing traffic prediction models struggle to effectively model the long-term impact of external disturbances on traffic evolution trajectories when faced with sudden and abnormal scenarios. Furthermore, they suffer from decreased prediction accuracy and negative migration issues when samples are scarce. In particular, they are unable to flexibly represent drastic shifts in the local correlation relationships of the road network in complex and ever-changing external environments.

Method used

By acquiring multi-source heterogeneous correlation data and prior information about future events, dynamic edge weight components are generated using a multimodal primitive learning method. Topological mask constraints are then applied using a static adjacency matrix to construct a time-varying adjacency matrix. Graph propagation operations are performed and the matrix is ​​aggregated with a multimodal latent representation set. Conditional decoding prediction is then performed using prior information about future events to output the future traffic state.

Benefits of technology

It improves the model's ability to capture spatial correlations and its prediction accuracy under abnormal interference, and enables reliable prediction of traffic conditions in complex scenarios.

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

Abstract

The invention discloses an abnormal event-oriented multi-modal primitive learning traffic prediction method, system and device, and a storage medium. The method comprises the following steps: acquiring multi-source heterogeneous associated data, future event prior information and a static adjacency matrix of a target road network; extracting a multi-modal implicit representation set, wherein the implicit representation set at least comprises traffic implicit representations; generating a dynamic edge weight component based on traffic hidden representation, applying topological mask constraint by using the static adjacency matrix, and determining a time-varying adjacency matrix; using the time-varying adjacency matrix to execute graph propagation operation to obtain enhanced spatial features, and aggregating the enhanced spatial features with other hidden representations to obtain fusion node representations; and converting future event prior information into prospective condition representation, combining the prospective condition representation with the fusion node representation to carry out condition decoding prediction, and outputting a future traffic state prediction result. According to the invention, the space correlation capture capability and prediction precision of the model under abnormal interference are improved, and reliable pre-judgment of the traffic state in a complex scene is realized.
Need to check novelty before this filing date? Find Prior Art