基于主动学习机制GAN的MSWI过程二噁英排放风险预警模型  

Dioxin Emission Risk Warning Model in MSWI Process Based on GAN With Active Learning Mechanism

作　　者：汤健 崔璨麟 夏恒 王丹丹 乔俊飞 TANG Jian;CUI Canlin;XIA Heng;WANG Dandan;QIAO Junfei(Faculty of Information Technology,Beijing University of Technology,Beijing 100124,China;Beijing Laboratory of Smart Environmental Protection,Beijing 100124,China)

机构地区：[1]北京工业大学信息学部,北京100124 [2]智慧环保北京实验室,北京100124

出　　处：《北京工业大学学报》2023年第5期507-522,共16页Journal of Beijing University of Technology

基　　金：国家自然科学基金资助项目(62073006);北京市自然科学基金资助项目(4212032)。

摘　　要：针对构建城市固废焚烧(municipal solid waste incineration,MSWI)过程剧毒污染物二噁英(dioxin,DXN)排放风险预警模型的样本极为稀少的问题,提出一种基于主动学习机制生成对抗网络(generative adversarial network,GAN)的DXN排放风险预警建模方法.首先,以DXN风险等级作为条件信息使得GAN生成候选虚拟样本;然后,利用基于最大均值差异和多视角可视化分布信息的主动学习机制进行虚拟样本的初筛和评估,以获得期望虚拟样本;最后,基于混合样本构建DXN排放风险预警模型.通过基准数据集和MSWI过程数据集验证了所提方法的有效性.基于主动学习机制GAN的DXN排放风险预警建模方法可以有效解决样本稀少的问题,提高模型精度.To solve the problem that the samples to construct a risk warning model of dioxin(DXN)emission in municipal solid waste incineration(MSWI)process are extremely scarce,a modeling method of DXN emission risk warning based on generative adversarial network(GAN)with active learning mechanism was proposed.First,the risk level of DXN was added as condition information to GAN,so that the generator generated candidate virtual samples with specified requirement.Then,the active learning mechanism based on maximum mean discrepancy and multi-view visual distribution information was used to evaluate and screen the virtual samples that met the experts'expectations.Finally,the DXN emission risk warning model was constructed based on the mixed samples composed of virtual samples and real samples.The validity and rationality of the proposed method were verified by using benchmark and MSWI process data sets.The proposed modeling method of DXN emission risk warning based on GAN with active learning mechanism can effectively solve the problem of scarce samples and improve the accuracy of the model.

关 键 词：城市固废焚烧(municipal solid waste incineration MSWI) 二噁英(dioxin DXN)排放风险预警 生成对抗网络(generative adversarial network GAN) 虚拟样本生成(virtual sample generation VSG) 最大均值差异 主动学习 

分 类 号：U461[机械工程—车辆工程] TP308[交通运输工程—载运工具运用工程]