Scholarlink

<p>As environmental regulations grow increasingly stringent and renewable energy integration expands, the rapid load variations in thermal power units have intensified dynamic fluctuations in sulfur dioxide (SO₂) emissions. Furthermore, the significant time-lag effects across different levels of equipment prevent the Distributed Control System (DCS) from responding promptly, often resulting in exceedances of pollutant emission limits. Therefore, accurate long-term prediction of SO₂ emission concentration at the desulfurization outlet has become particularly crucial. To address this issue, this paper proposes a TCN-Informer hybrid model based on a stacked integration algorithm, which combines the strengths of Temporal Convolutional Network (TCN) and Informer to achieve precise medium- to long-term forecasting of SO₂ emission concentrations. Comparative experiments were conducted across multiple time horizons, and the results demonstrate that the TCN-Informer model achieved 13 best performances over five prediction horizons in terms of five evaluation metrics: coefficient of determination (R<sup>2</sup>), mean absolute error (MAE), mean squared error (MSE), root mean squared error (RMSE), and mean absolute percentage error (MAPE). Experimental results indicate that for medium- to long-term predictions (24-step and 48-step), the model achieves MAE values of 0.825 and 0.914, MAPE values of 5.841% and 9.441%, and R<sup>2</sup> values of 0.9859 and 0.9781, respectively. This demonstrates its strong capability in predicting medium- and long-term SO₂ emission trends, providing technical support for achieving ultra-low SO₂ emissions in coal-fired power plants.</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

Sulfur Dioxide Emission Prediction of Coal-Fired Power Plants based on the stacked TCN-Informer Network

  • 期刊论文

  • 发表日期：

    2025年11月21日

查看全文
  • Haidong Kou,
  • Xiaojian Wang,
  • Jialu Chen,
  • Yelidana Hayireti,
  • Hao Zhou

摘要

As environmental regulations grow increasingly stringent and renewable energy integration expands, the rapid load variations in thermal power units have intensified dynamic fluctuations in sulfur dioxide (SO₂) emissions. Furthermore, the significant time-lag effects across different levels of equipment prevent the Distributed Control System (DCS) from responding promptly, often resulting in exceedances of pollutant emission limits. Therefore, accurate long-term prediction of SO₂ emission concentration at the desulfurization outlet has become particularly crucial. To address this issue, this paper proposes a TCN-Informer hybrid model based on a stacked integration algorithm, which combines the strengths of Temporal Convolutional Network (TCN) and Informer to achieve precise medium- to long-term forecasting of SO₂ emission concentrations. Comparative experiments were conducted across multiple time horizons, and the results demonstrate that the TCN-Informer model achieved 13 best performances over five prediction horizons in terms of five evaluation metrics: coefficient of determination (R2), mean absolute error (MAE), mean squared error (MSE), root mean squared error (RMSE), and mean absolute percentage error (MAPE). Experimental results indicate that for medium- to long-term predictions (24-step and 48-step), the model achieves MAE values of 0.825 and 0.914, MAPE values of 5.841% and 9.441%, and R2 values of 0.9859 and 0.9781, respectively. This demonstrates its strong capability in predicting medium- and long-term SO₂ emission trends, providing technical support for achieving ultra-low SO₂ emissions in coal-fired power plants.