摘要
城市集体供热管网属于市政工程管网的重要组成部分,其安全稳定运行与城市经济生产和居民日常生活息息相关,因此对供热管网进行准确实时的状态监测至关重要。近年来,基于深度学习的方法已经被广泛应用于状态监测领域,如对抗自编码器(adversarial auto-encoder,AAE)。然而,从信息论的角度看,在AAE模型训练过程中样本与特征表示之间的互信息存在衰减现象,从而直接影响到该网络模型的故障检测性能。为此,提出了一种基于逐层互信息对抗自编码器(layer-by-layer mutual information adversarial auto-encoder,LM-AAE)的故障检测方法,该方法通过显性引入低维特征空间与前面每一层神经网络的互信息,以最大化正常输入样本与特征表示之间的相关性,有效克服了AAE模型训练过程中的互信息衰减问题。最后,将LM-AAE模型、VAE模型和传统AAE模型分别用于连续搅拌釜式加热器实验,结果表明LM-AAE模型在保证较小故障漏报率的同时具有最小的故障误报率。证明了引入逐层互信息策略可以使模型在故障检测任务中更具优越性。
Being an important part of the municipal engineering network,safe and stable operation of urban central heating pipe network is closely related to the city's economic production and residents'daily life,so that it is crucial to conduct accurate and real-time condition monitoring of the heating pipe network.In recent years,deep learning-based methods have been widely used in the field of condition monitoring,such as adversarial auto-encoder(AAE).However,from the perspective of information theory,there is a decay of mutual information between samples and feature representations during the training process of AAE model,which directly affects the fault detection perfor-mance of the model.A fault detection method based on layer-by-layer mutual information(LM)AAE is proposed.By explicitly introducing mutual information between the low-dimensional feature space and each previous layer of the neural network and the correlation is maximum between normally input samples and feature representations,the mutual information decay problem is effectively overcome during AAE model training.Finally,LM-AAE model,VAE model and traditional AAE model are used for continuous stirred kettle-type heater experiments respectively.The results show that LM-AAE model has both the smallest fault false alarm rate and a smaller fault leakage rate.It is demonstrated that the introduction of a layer-by-layer mutual information strategy can make the model more su-perior in fault detection.
作者
刘自鹏
李灵
刘述
李磊
熊凌云
刘雅儒
Liu Zipeng;Li Ling;Liu Shu;Li Lei;Xiong Lingyun;Liu Yaru(School of Electrical and Information Engineering,Changsha University of Science and Technology,Changsha 410114,China;Hangzhou Zhiyuan Research Institute,Hangzhou 310013,China;Chengnan College,Changsha University of Science and Technology,Changsha 410114,China;School of Transportation Engineering,Changsha University of Science and Technology,Changs ha 410114,China)
出处
《市政技术》
2024年第5期220-227,共8页
Journal of Municipal Technology
基金
湖南省自然科学基金项目(2022JJ40510)
湖南省教育厅科学研究项目(22B0329,21B0311)
长沙理工大学校级研究生科研创新项目(CXCLY2022083)。
关键词
供热管网
故障检测
无监督学习
对抗自编码器
逐层互信息
heating pipe network
fault detection
unsupervised learning
adversarial auto-encoder(AAE)
layer-by-layer mutual information
