Research on Traffic Flow Forecasting Based on Spatial-Temporal Network

• Published in: 2025 8th International Conference on Advanced Algorithms and Control Engineering (ICAACE) pp. 1532 - 1535
• Main Authors: Chen, Xun, Liang, Jinsu, Deng, Linyi, Xie, Tianyi
• Format: Conference Proceeding
• Language: English
• Published: IEEE 21.03.2025
• Subjects: Adaptation models; Computational modeling; Data mining; Data models; Feature extraction; Forecasting; GAT; Predictive models; Roads; spatial and temporal information; traffic flow forecasting; Transformer; Transformers; Vectors
• DOI: 10.1109/ICAACE65325.2025.11020230

Traffic flow forecasting is an important research subject related to social livelihood and economic development. How to improve the accuracy of traffic flow forecasting has been widely concerned by people. This paper proposes a traffic flow prediction model based on Transformer. The model uses GAT, which can dynamically aggregate spatial features, to model the spatial dependence of traffic flow. The self-attention mechanism in Transformer is used to adaptively capture long-term dependencies from traffic flow data to model time dependencies of traffic flow. Spatial-temporal coding is embedded in the feature vector of input data. The traffic flow prediction model is tested on six real-world traffic flow datasets, and compared with some classic baseline models, the traffic flow prediction model proposed in this paper has achieved good prediction performance.