Scholarlink

<p>Metallurgical-grade silicon is a vital industrial material, but the substantial emission of Sulfur dioxide (SO<sub>2</sub>) from its smelting process hinders the sustainable development of the silicon production industry. Conventional end-of-pipe desulfurization is expensive and poses a risk of secondary pollution, highlighting the need for source-level SO<sub>2</sub> control. This study utilizes production data to examine the influence of carbonaceous reducing agent ratios and impurity element content on SO<sub>2</sub> generation. A deep neural network (DNN) prediction model is developed, and the contributions of input factors to SO<sub>2</sub> emissions are quantified via Sobol sensitivity analysis and Shapley additive explanations values to identify key influencing variables. The results show that the DNN model achieves high predictive accuracy for SO<sub>2</sub> concentration (R<sup>2</sup> ≈ 0.99). Sensitivity analysis identifies the proportions of carbonaceous reducing agents, sulfur content, and basic oxide concentrations (e.g., CaO) in ash as the main factors influencing SO<sub>2</sub> emissions. Furthermore, under sufficient fixed carbon conditions, moderately increasing pellet and coal input while reducing raw material moisture content mitigates SO<sub>2</sub> generation. Conversely, excessive use of highly reactive carbon sources such as charcoal exacerbates emissions. The established model and methodology provide a scientific foundation for optimizing raw material formulations and implementing source emission reduction in silicon smelting process.</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

Analysis of Multisource Driving Mechanisms of Sulfur Dioxide Emissions in Metallurgical-grade Silicon Smelting and Deep Neural Network Modeling

  • 期刊论文

  • 发表日期：

    2026年2月13日

查看全文
  • Lin Wang,
  • Zhengjie Chen,
  • Yiyou Hu,
  • Yangdong Ou,
  • Xiaoqing Ren,
  • Zhongyi Zhang,
  • Wenhui Ma

摘要

Metallurgical-grade silicon is a vital industrial material, but the substantial emission of Sulfur dioxide (SO2) from its smelting process hinders the sustainable development of the silicon production industry. Conventional end-of-pipe desulfurization is expensive and poses a risk of secondary pollution, highlighting the need for source-level SO2 control. This study utilizes production data to examine the influence of carbonaceous reducing agent ratios and impurity element content on SO2 generation. A deep neural network (DNN) prediction model is developed, and the contributions of input factors to SO2 emissions are quantified via Sobol sensitivity analysis and Shapley additive explanations values to identify key influencing variables. The results show that the DNN model achieves high predictive accuracy for SO2 concentration (R2 ≈ 0.99). Sensitivity analysis identifies the proportions of carbonaceous reducing agents, sulfur content, and basic oxide concentrations (e.g., CaO) in ash as the main factors influencing SO2 emissions. Furthermore, under sufficient fixed carbon conditions, moderately increasing pellet and coal input while reducing raw material moisture content mitigates SO2 generation. Conversely, excessive use of highly reactive carbon sources such as charcoal exacerbates emissions. The established model and methodology provide a scientific foundation for optimizing raw material formulations and implementing source emission reduction in silicon smelting process.