基于特征约简与选择性集成算法的城市固废焚烧过程二噁英排放浓度软测量
Soft measuring approach of dioxin emission concentration in municipal solid waste incineration process based on feature reduction and selective ensemble algorithm
摘要点击 143  全文点击 32  投稿时间:2019-10-21  修订日期:2020-08-24
查看全文  查看/发表评论  下载PDF阅读器
DOI编号  10.7641/CTA.2020.90874
  2021,38(1):110-120
中文关键词  城市固废焚烧(MSWI)  二噁英(DXN)  变量投影重要性(VIP)  特征约简  选择性集成算法
英文关键词  municipal solid waste incineration (MSWI)  Dioxin (DXN)  variable projection importance (VIP)  feature reduction  selective ensemble
基金项目  国家自然科学基金项目(62073006, 61803191, 61973226), 国家重点研发计划项目(2018YFC1900800–5), 矿冶过程自动控制技术国家(北京市)重 点实验室项目(BGRIMM–KZSKL–2020–02).
作者单位E-mail
汤健 北京工业大学信息学部 tjian001@126.com 
乔俊飞 北京工业大学信息学部  
徐喆 北京工业大学信息学部  
郭子豪 北京工业大学信息学部  
中文摘要
      城市固废焚烧(MSWI)排放的污染物二噁英(DXN)对生态环境与人类健康具有极大危害, 其浓度的实时检 测对MSWI过程运行优化和城市污染控制至关重要. 具有痕量特性的DXN排放浓度不能实时检测, 机理模型难以构 建, 并且其与过程变量间的映射关系复杂. 针对上述问题, 本文提出了一种基于特征约简和选择性集成算法的DXN 排放浓度软测量方法. 首先, 对在线采集的MSWI过程变量和离线化验的DXN排放浓度数据进行预处理, 获得具有 小样本高维特性的建模样本; 接着, 基于变量投影重要性(VIP)值和特征约简比率值确定模型输入特征; 最后, 基于 操纵训练样本的集成构造策略构建自适应确定核参数的选择性集成模型. 采用国外文献和国内工业MSWI过程 的DXN排放浓度数据仿真验证了所提方法的有效性.
英文摘要
      Municipal solid waste incineration (MSWI) process produces a type of highly toxic and persistent pollutant, i.e., Dioxins (DXN), which has tremendous realistic and potential hazards to the ecological environment and human health. It is very important for optimizing operation of MSWI process and controlling urban pollution in terms of realization of continuous real-time measurement of DXN emission concentration. The generation mechanism of DXN is very complex. Thus, there is a complex non-linear mapping relationship between DXN and input/output variables of MSWI process. Aim at these problems, a soft measuring method of DXN emission concentration based on feature selection and selective ensemble strategy is proposed. Firstly, the process variables of easy-to-measure are matched to obtain the modeling sample with characteristics of small sample and high dimension. Then, the variable projection importance (VIP) value based on linear projection to latent structure algorithm and input feature selection ratio based on expert experience are used to select the input features. At last, by using ensemble construction strategy based on manipulating training sample with characteristic of adaptive select kernel parameter is constructed. The proposed method is simulated and validated by using the data of DXN emission concentration in reference and actual MSWI process.