@article{ref1,
title="Traffic state prediction for urban networks: a spatial-temporal transformer network model",
journal="Journal of transportation engineering, Part A: Systems",
year="2023",
author="Ji, Xinkai and Mao, Peipei and Han, Yu",
volume="149",
number="11",
pages="e04023105-e04023105",
abstract="Traffic state prediction plays an important role in traffic management, e.g., it can provide travelers with accurate routing information to achieve a better travel experience. In this paper, we propose a spatial-temporal transformer network (STTN) model on the traffic state prediction for ubran networks. The STTN model integrates four modules: road embedding (RE); basic information embedding (BIE); temporal transformer (TT); and spatial-temporal transformer (STT). Specifically, the road topology information and other basic road information are embedded in the RE and BIE modules, respectively. The TT module, which is developed based on the Transformer encoder, captures the variation of the sequential historical traffic flow data. The STT module fuses a TT, which captures the spatial correlations and temporal dynamics of network traffic state, and the attention mechanism, which adjusts the importance of different historical data. The performance of the proposed STTN model is demonstrated using real traffic data collected from crowd-sourced vehicles. The proposed model achieves better prediction accuracy in terms of f1-score and weighted f1-score compared with those of other baseline models. The ablation study shows that some modules in the proposed STTN have a significant impact on improving short-term prediction ability.<p /> <p>Language: en</p>",
language="en",
issn="2473-2907",
doi="10.1061/JTEPBS.TEENG-7860",
url="http://dx.doi.org/10.1061/JTEPBS.TEENG-7860"
}