摘要
本文利用深度学习在模式识别和特征提取方面的优势,提出了基于堆栈自编码和Softmax算法的多联机制冷剂充注量故障诊断策略。针对堆栈自编码和Softmax的故障诊断方法,本文主要从网络的层数、隐含层节点数、学习率大小、迭代次数以及Batchsize(批次样本数)大小这些超参数的选择探索与故障诊断模型性能的关系。此外,在堆栈自编码的基础上,本文还采用了传统自编码的变种(降噪自编码和稀疏自编码)来对故障诊断模型进行优化。结果表明:堆栈降噪自编码及堆栈稀疏自编码与Softmax的故障诊断模型能获得更好的诊断性能,在一定参数条件下诊断准确率均能达到96%以上。
The advantages of deep learning in pattern recognition and feature extraction are used,and a variable refrigerant flow refrigerant charge fault diagnosis strategy based on stack auto-encoding and Softmax algorithm is proposed in this paper.Aiming at the stack auto-encoding and Softmax fault diagnosis methods,the selection of hyperparameters and the fault diagnosis model from the number of layers of the network,the number of hidden layer nodes,the size of the learning rate,the number of iterations,and the size of Batchsize(the number of samples in batches)are mainly explored.In addition,on the basis of stack auto-encoding,a variant of traditional autoencoding-denoising auto-encoding and sparse auto-encoding are used to optimize the fault diagnosis model.The results show that the stack noise reduction self-encoding and stack sparse self-encoding and Softmax fault diagnosis models can achieve better diagnostic performance,and the diagnostic accuracy rate can reach more than 96%under certain parameters.
作者
苟伟
王凌云
李正飞
陈焕新
刘志龙
陈建业
程亨达
张鉴心
GOU Wei;WANG Lingyun;LI Zhengfei;CHEN Huanxin;LIU Zhilong;CHEN Jianye;CHENG Hengda;ZHANG Jianxin(School of Energy and Power Engineering,Huazhong University of Science and Technology,Wuhan 43007,Hubei,China;State Key Laboratory of Compressor Technology(Anhui Laboratory of Compressor Technology),Hefei 230031,Anhui,China)
出处
《制冷技术》
2022年第4期27-33,共7页
Chinese Journal of Refrigeration Technology
基金
国家自然科学基金(No.51876070)
压缩机技术国家重点实验室开放基金项目(No.SKL-YSJ201912)。
关键词
多联机系统
故障诊断
堆栈自编码
堆栈降噪自编码
堆栈稀疏自编码
Variable refrigerant flow system
Fault diagnosis
Stack auto-encoding
Stack noise reduction autoencoding
Stack sparse auto-encoding