Scholarlink

<p>With air pollution representing an escalating global public health crisis, reliable air quality forecasting is critical for guiding data-driven environmental policymaking. However, existing efforts demonstrate limitations in multivariate, long-range predictive accuracy, constraining more proactive interventions. This research applies state-of-the-art Transformer-based Informer model, leveraging self-attention mechanisms for effective sequence modeling, to advance air quality forecasting. Using an extensive dataset encompassing eight key pollutants in Delhi, India, the Informer model achieves remarkable performance across evaluation metrics for pollutants forecasting. Impressively, this high accuracy holds even for difficult 24-h ahead multivariate predictions across the intricate 8-pollutant system, substantially elevating current forecasting capabilities. The consistent mean square error below 0.04 highlights the predictive advancements derived from applying advanced deep learning architectures. With integration of meteorological and emissions data, the paper contributes to precisely informed, anticipatory environmental assessment which can be useful in policymaking to protect ecological balance and public health in the face of the urgent air pollution crisis worldwide.</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

Advancing multi-pollutant air quality forecasting using Transformer-based Informer architecture

  • 期刊论文

  • 发表日期：

    2025年2月24日

查看全文
  • Vahid Khatibi,
  • Parsa Nikpour

摘要

With air pollution representing an escalating global public health crisis, reliable air quality forecasting is critical for guiding data-driven environmental policymaking. However, existing efforts demonstrate limitations in multivariate, long-range predictive accuracy, constraining more proactive interventions. This research applies state-of-the-art Transformer-based Informer model, leveraging self-attention mechanisms for effective sequence modeling, to advance air quality forecasting. Using an extensive dataset encompassing eight key pollutants in Delhi, India, the Informer model achieves remarkable performance across evaluation metrics for pollutants forecasting. Impressively, this high accuracy holds even for difficult 24-h ahead multivariate predictions across the intricate 8-pollutant system, substantially elevating current forecasting capabilities. The consistent mean square error below 0.04 highlights the predictive advancements derived from applying advanced deep learning architectures. With integration of meteorological and emissions data, the paper contributes to precisely informed, anticipatory environmental assessment which can be useful in policymaking to protect ecological balance and public health in the face of the urgent air pollution crisis worldwide.