引用本文:许超凡,汤健,夏恒,徐喆,乔俊飞.联合多窗口检测的MSWI过程二噁英排放预测模型[J].控制理论与应用,2024,41(11):2093~2102.[点击复制]
XU Chao-fan,TANG Jian,XIA Heng,XU Zhe,Qiao Jun-fei.Dioxin emission prediction for municipal solid waste incineration process with multi-window combination detection[J].Control Theory and Technology,2024,41(11):2093~2102.[点击复制]
联合多窗口检测的MSWI过程二噁英排放预测模型
Dioxin emission prediction for municipal solid waste incineration process with multi-window combination detection
摘要点击 127  全文点击 28  投稿时间：2022-03-07  修订日期：2022-06-04
查看全文  查看/发表评论  下载PDF阅读器
DOI编号  10.7641/CTA.2023.20161
  2024,41(11):2093-2102
中文关键词  城市固废焚烧  二噁英排放预测  多窗口检测  概念漂移  模型更新
英文关键词  municipal solid waste incineration (MSWI)  dioxin (DXN) emission prediction  multi-window detection  concept drift  model update
基金项目  北京市自然科学基金资助项目(4212032, 4192009), 国家自然科学基金项目(62073006, 62021003), 科技创新2030–“新一代人工智能”重大项目 (2021ZD0112301, 2021ZD0112302)资助.
作者单位邮编
许超凡 北京工业大学信息学部 100024
汤健* 北京工业大学 信息学部 100124
夏恒 北京工业大学信息学部 
徐喆 北京工业大学信息学部 
乔俊飞 北京工业大学信息学部 
中文摘要
      二噁英(DXN)是城市固废焚烧过程(MSWI)排放的难以实时检测的剧毒污染物. MSWI过程的时变特性导致软测量模型的在线预测性能降低. 针对上述问题, 本文提出一种联合多窗口检测的DXN排放在线预测方法. 首先, 联合数据标准化、在线预测窗口实现新样本的DXN排放预测, 再联合离群样本检测、特征空间检测和输出空间检测窗口实现漂移样本的识别. 然后, 对上述漂移样本进行去冗处理并判断其数量是否满足预设定的阈值, 若满足则构建新模型, 否则继续采用历史模型. 最后, 采用MSWI过程数据验证了所提方法的有效性.
英文摘要
      Dioxin (DXN) is a highly toxic pollutant emitted from municipal solid waste incineration (MSWI), which is difficult to be detected in real time. The time-varying characteristics of the MSWI process leads to a decrease in the online prediction performance of the soft-sensor model. Aiming at the above problems, this paper proposes a method of DXN emission prediction with multi-window combination detection. Firstly, combined the data standardization window and the online prediction window enables DXN emission prediction for new samples, and the outlier detection window, the feature detection space window and the output space detection window are combined to realize the identification of drifting samples. Then, the detected drift samples are combined and removed duplicates, and whether the number of drift sample sets reaches the pre-set threshold is judged. If it reaches, the new models are constructed, otherwise the historical model is to be used continuously. Finally, the real industrial process data are used to verify the effectiveness of the proposed method.