基于模糊神经网络对抗生成的城市固废焚烧过程二噁英排放预警

Dioxin emission risk warning model in MSWI process based on adversarial generative FNN

中文关键词  城市固废焚烧  二噁英  模糊神经网络  生成对抗网络  虚拟样本  预警模型

英文关键词  municipal solid waste incineration  dioxin  fuzzy neural network  generative adversarial network  virtual samples  warning model

中文摘要

      城市固废焚烧(MSWI)过程产生的二噁英(DXN)类剧毒污染物是全世界范围内备受关注的环保指标,进 行DXN排放浓度预警是缓解焚烧建厂“邻避效应”和实现城市精准污染防控等难题的关键之一.受限于产生机理上 的全流程相关、记忆效应等特性以及检测技术上的高难度和离线化验上的高成本等原因,DXN建模数据面临着维 数高、不确定性强和样本稀疏等问题.对此,本文提出基于模糊神经网络(FNN)对抗生成的DXN排放预警方法.首 先, 采用基于随机森林(RF)的自适应特征选择算法降低输入变量维数;接着,基于FNN的生成对抗网络(GAN)迭代 产生用于预警建模的候选虚拟样本,以缓解不确定性和稀疏性问题;然后,通过多约束选择机制进行虚拟样本筛选 以提高样本质量;最后,构建基于真实与虚拟混合样本的DXN排放预警模型.基于北京某MSWI电厂的实际DXN数 据验证了所提方法的有效性.

英文摘要

      Dioxin (DXN) emission in municipal solid waste incineration (MSWI) process is the key environmental protection index strictly restricted in the world. The risk warning of DXN emission is one of the primary problems to alleviate the “not in my backyard” in incineration plant construction and to realize accurate pollution control in the city. However, due to the correlation of the whole process and the memory effect in terms of the generation mechanism of DXN, the difficulty of online detection technology, and the high cost of offline testing, its modeling samples have the characteristics of high dimension, strong uncertainty, and small quantity. To solve the above problem, the method of DXN emission risk warning model in the MSWIprocess based on adversarial generative fuzzy neural network (FNN) is proposed. Firstly, an adaptive feature selection algorithm based on random forest (RF) is used for input feature reduction. Then, a generative adversarial network (GAN) based on FNN is used to generate candidate virtual samples for DXN risk warning modeling to alleviate the problems of uncertainty and small samples. In addition, the virtual samples are screened through the multi-constraint selection mechanism to improve the sample quality. Finally, the risk warning model of DXN emission based on mixed samples is constructed. The effectiveness of the proposed method is verified based on actual DXN data of an MSWIpower plant in Beijing.