Scholarlink

<p>Accurate prediction of ozone concentration is essential for environmental management and public health. This study introduces a novel spatio-temporal deep learning model, ST-OzoneNet. The model captures the spatial dependence between monitoring sites by constructing multiple graph structures and introducing a multi-graph attention module. At the same time, adaptive noise complete ensemble empirical mode decomposition (CEEMDAN) and variational mode decomposition (VMD) are used to extract multi-scale features, and bi-directional long short-term memory network (BiLSTM) and multi-head attention mechanism are combined to capture short-term fluctuations and long-term trends of ozone. Experimental results demonstrate that ST-OzoneNet outperforms the baseline model. In the 72-hour prediction task, the root mean square error (RMSE) of the model remains at 14.02, and the coefficient of determination (R<InlineEquation ID="IEq1"> <EquationSource Format="TEX">\(^{2}\)</EquationSource> </InlineEquation>) reaches 0.91, showing good long-term prediction stability. In addition, R<InlineEquation ID="IEq2"> <EquationSource Format="TEX">\(^{2}\)</EquationSource> </InlineEquation> is greater than 0.95 when using the data of Shanghai and Chongqing to test the model, which confirms its good generalization ability. This study provides an efficient and reliable technical solution for ozone concentration prediction, which has important application value.</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

ST-OzoneNet: A spatio-temporal deep learning model for accurate prediction of ozone concentration

  • 期刊论文

  • 发表日期：

    2025年10月4日

查看全文
  • Chaoli Tang,
  • Diandian Zhen,
  • Lingqian Zhang,
  • Fengmei Zhao,
  • Yuanyuan Wei

摘要

Accurate prediction of ozone concentration is essential for environmental management and public health. This study introduces a novel spatio-temporal deep learning model, ST-OzoneNet. The model captures the spatial dependence between monitoring sites by constructing multiple graph structures and introducing a multi-graph attention module. At the same time, adaptive noise complete ensemble empirical mode decomposition (CEEMDAN) and variational mode decomposition (VMD) are used to extract multi-scale features, and bi-directional long short-term memory network (BiLSTM) and multi-head attention mechanism are combined to capture short-term fluctuations and long-term trends of ozone. Experimental results demonstrate that ST-OzoneNet outperforms the baseline model. In the 72-hour prediction task, the root mean square error (RMSE) of the model remains at 14.02, and the coefficient of determination (R \(^{2}\) ) reaches 0.91, showing good long-term prediction stability. In addition, R \(^{2}\) is greater than 0.95 when using the data of Shanghai and Chongqing to test the model, which confirms its good generalization ability. This study provides an efficient and reliable technical solution for ozone concentration prediction, which has important application value.