Scholarlink

<p>Accurate traffic flow prediction is crucial for the efficient operation of intelligent transport systems. However, existing models usually treat spatio-temporal features separately, which makes it difficult to effectively simulate the intrinsic coupling relationship of complex traffic flows. To solve this problem, we propose the HST-GAN model to unify the spatio-temporal relationships in traffic data from a multi-frequency perspective through spatio-temporal fusion feature extraction. Firstly, HST-GAN dynamically adjusts the graph structure through the event-aware mechanism to accurately capture the local impacts of high-frequency emergencies on traffic flow; meanwhile, the global spatio-temporal feature module in the model extracts low-frequency periodic information from historical data. With the designed dynamic gating module, the effective fusion of high-frequency spatio-temporal features and low-frequency spatio-temporal features is achieved to model complex traffic patterns more comprehensively. In addition, we constructed a discriminator to predict by generating confrontation, which not only improves the prediction accuracy and noise resistance of the model under uncertain data, but also effectively solves the problem of difficult learning of multi-frequency features in traffic prediction. The experimental results show that the prediction accuracies of HST-GAN on six real traffic datasets are better than the existing mainstream methods.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Traffic flow prediction based on generative adversarial network with hybrid spatio temporal features learning

  • 期刊论文

  • 发表日期：

    2025年9月3日

查看全文
  • Wei Sun,
  • Kaiqi Deng,
  • Junbo Gao

摘要

Accurate traffic flow prediction is crucial for the efficient operation of intelligent transport systems. However, existing models usually treat spatio-temporal features separately, which makes it difficult to effectively simulate the intrinsic coupling relationship of complex traffic flows. To solve this problem, we propose the HST-GAN model to unify the spatio-temporal relationships in traffic data from a multi-frequency perspective through spatio-temporal fusion feature extraction. Firstly, HST-GAN dynamically adjusts the graph structure through the event-aware mechanism to accurately capture the local impacts of high-frequency emergencies on traffic flow; meanwhile, the global spatio-temporal feature module in the model extracts low-frequency periodic information from historical data. With the designed dynamic gating module, the effective fusion of high-frequency spatio-temporal features and low-frequency spatio-temporal features is achieved to model complex traffic patterns more comprehensively. In addition, we constructed a discriminator to predict by generating confrontation, which not only improves the prediction accuracy and noise resistance of the model under uncertain data, but also effectively solves the problem of difficult learning of multi-frequency features in traffic prediction. The experimental results show that the prediction accuracies of HST-GAN on six real traffic datasets are better than the existing mainstream methods.