首页|基于混合数据增强的MSWI过程燃烧状态识别

基于混合数据增强的MSWI过程燃烧状态识别

扫码查看
原文链接
  • 万方数据
  • 维普
国内城市固废焚烧(Municipal solid waste incineration,MSWI)过程通常依靠运行专家观察炉内火焰识别燃烧状态后再结合自身经验修正控制策略以维持稳定燃烧,存在智能化水平低、识别结果具有主观性与随意性等问题.由于MSWI过程的火焰图像具有强污染、多噪声等特性,并且存在异常工况数据较为稀缺等问题,导致传统目标识别方法难以适用.对此,提出一种基于混合数据增强的MSWI过程燃烧状态识别方法.首先,结合领域专家经验与焚烧炉排结构对燃烧状态进行标定;接着,设计由粗调和精调两级组成的深度卷积生成对抗网络(Deep convolutional generative adversarial network,DCGAN)以获取多工况火焰图像;然后,采用弗雷歇距离(Fréchet inception distance,FID)对生成式样本进行自适应选择;最后,通过非生成式数据增强对样本进行再次扩充,获得混合增强数据构建卷积神经网络以识别燃烧状态.基于某MSWI电厂实际运行数据实验,表明该方法有效地提高了识别网络的泛化性与鲁棒性,具有良好的识别精度.
Combustion States Recognition Method of MSWI Process Based on Mixed Data Enhancement
The municipal solid waste incineration(MSWI)process usually relies on operating experts to observe the flame inside furnace for recognizing the combustion states.Then,by combining the experts'own experience to modify the control strategy to maintain the stable combustion.Thus,this manual mode has disadvantages of low intelligence and the subjectivity and randomness recognition results.The traditional methods are difficult to apply to the MSWI process,which has the characteristics of strong pollution,multiple noise,and scarcity of samples un-der abnormal conditions.To solve the above problems,a combustion states recognition method of MSWI process based on mixed data enhancement is proposed.Firstly,combustion states are labeled by combining the experience of domain experts and the design structure of furnace grate.Next,a deep convolutional generative adversarial net-work(DCGAN)consisting of two levels of coarse and fine-tuning was designed to acquire multi-situation flame im-ages.Then,the Fréchet inception distance(FID)is used to adaptively select generated samples.Finally,the sample features are enriched at the second time by using non-generative data enhancement strategy,and a convolutional neural network is constructed based on the mixed enhanced data to recognize the combustion state.Experiments based on actual operating data of a MSWI plant show that this method effectively improves the generalization and robustness of the recognition network and has good recognition accuracy.

Municipal solid waste incineration(MSWI)deep convolutional generative adversarial network(DCGAN)combustion states recognitionnon-generation data enhancementmixed data enhancement

郭海涛、汤健、丁海旭、乔俊飞

展开 >

北京工业大学信息学部 北京 100124

智慧环保北京实验室 北京 100124

城市固废焚烧 深度卷积生成对抗网络 燃烧状态识别 非生成式数据增强 混合数据增强

国家自然科学基金国家自然科学基金北京市自然科学基金北京市自然科学基金国家重点研发计划矿冶过程自动控制技术国家(北京市)重点实验室项目

6207300662021003421203241920092018YFC1900800-5BGRIMM-KZSKL-2020-02

2024

10.16383/j.aas.c210843

自动化学报
中国自动化学会 中国科学院自动化研究所

自动化学报

CSTPCD北大核心
影响因子：1.762
ISSN：0254-4156
年,卷(期)：2024.50(3)
  • 2
  • 48