Search

• Fine-grained modeling of temporal heterogeneity and spatial heterogeneity. • Trend spatial attention module models spatial heterogeneity. • Pyramidal attention models temporal heterogeneity and long-term dependence. • Trend construction module introduces local and global trend blocks. • TGAN achieves state-of-the-art performance on multiple datasets. Traffic prediction is an important part of urban computing. Accurate traffic prediction assists the public in planning travel routes and relevant departments in traffic management, thus improving the efficiency of people's travel. Existing approaches usually use graph neural networks or attention mechanisms to capture the spatial–temporal correlation of traffic data, neglecting to model the spatial heterogeneity and temporal heterogeneity in traffic data at a fine-grained level, which leads to biased prediction results. To address the above challenges, we propose a Trend Graph Attention Network (TGAN) to perform traffic prediction tasks. Specifically, we designed a trend spatial attention module, which constructs the spatial graph structure in the form of a trend-to-trend. Its main idea is to transfer information between nodes with similar attributes to solve the problem of spatial heterogeneity. For modeling the long-term temporal dependence, we introduce a trend construction module to build local and global trend blocks and perform aggregation operations between time steps and trend blocks so that each time step shares local and global fields. Lastly, we perform direct interaction between future and historical data to generate multi-step prediction results at once. Experimental results on five datasets for two types of traffic prediction tasks show that TGAN outperforms the state-of-the-art baseline. [ABSTRACT FROM AUTHOR]

Traffic prediction is an important part of smart city management. Accurate traffic prediction can be deployed in urban applications such as congestion alerting and route planning, thus providing sustainable services to the public or relevant departments. Although some improvements have been made in existing traffic prediction methods, there are challenges due to the following: (1) Time series has multi-scale nature, that is, from different scale time ranges, traffic flow changes show different trends; (2) Spatial heterogeneity, meaning that traffic conditions in similar functional areas are usually similar. This task remains difficult. To address the above challenges, we propose a new spatial-temporal prediction method, namely Multi-Scale Spatial-Temporal Aware Transformer (MSSTAT), which is a Transformer architecture with multi-scale characteristics. Specifically, compared to the input of encoder, the input of different decoder layers has different scale information, MSSTAT synchronizes model the connection between time steps and scale information by a kind of Parallel Cross Multi-Head Attention, which gives each time step several times the perceived field while also being able to weaken the impact brought by anomaly point. In addition, to add connections between regions with similar functions, we map the traffic data of each node as a probability distribution and then measure the similarity between the nodes by the Wasserstein Distance, which leads to our proposed spatial-temporal aware adjacency matrix. Experimental results on four traffic flow datasets show that MSSTAT outperforms the state-of-the-art baseline. [ABSTRACT FROM AUTHOR]