Temporal-Spatial Traffic Flow Prediction Model Based on Prompt Learning
Traffic flow prediction is one of the most important and attractive topics in geographical information science (GIS), traffic management, and logistics. Traffic flows exhibit significant complexity and dynamics, requiring a thorough understanding of their spatiotemporal evolution patterns for accura...
Saved in:
| Main Authors: | , , , , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
MDPI AG
2024-12-01
|
| Series: | ISPRS International Journal of Geo-Information |
| Subjects: | |
| Online Access: | https://www.mdpi.com/2220-9964/14/1/11 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | Traffic flow prediction is one of the most important and attractive topics in geographical information science (GIS), traffic management, and logistics. Traffic flows exhibit significant complexity and dynamics, requiring a thorough understanding of their spatiotemporal evolution patterns for accurate prediction and analysis. Existing studies utilizing deep learning for traffic flow prediction often suffer from distribution shift issues, leading to poor generalization capabilities when dealing with data that has different spatiotemporal distributions. Based on this, we propose a traffic flow prediction model based on prompt learning, leveraging graph convolutional networks to focus on the spatiotemporal dependencies of traffic flows. The model utilizes spatiotemporal context learning capabilities to capture the periodic states of traffic flows, enhancing the extraction of spatiotemporal features by integrating spatiotemporal information. Experimental results show that the spatiotemporal traffic flow prediction model equipped with a spatiotemporal prompt learning module outperforms several mainstream benchmark models in terms of predictive performance. The model presents efficient learning performance that reaches optimal state in a short period of time, reduces the impact of distribution shifts, and can be adapted to spatiotemporal traffic flow data under varying spatiotemporal contexts. |
|---|---|
| ISSN: | 2220-9964 |