化工学报 ›› 2018, Vol. 69 ›› Issue (12): 5155-5163.DOI: 10.11949/j.issn.0438-1157.20180704

• 过程系统工程 • 上一篇    下一篇

基于稀疏局部嵌入深度卷积网络的冷水机组故障诊断方法

刘旭婷1,2, 李益国1,2, 孙栓柱3, 刘西陲1,2, 沈炯1,2   

  1. 1. 东南大学能源与环境学院, 江苏 南京 210096;
    2. 东南大学能源热转换及其过程测控教育部重点实验室, 江苏 南京 210096;
    3. 江苏方天电力技术有限公司, 江苏 南京 211102
  • 收稿日期:2018-07-02 修回日期:2018-09-12 出版日期:2018-12-05 发布日期:2018-12-05
  • 通讯作者: 李益国
  • 基金资助:

    国家自然科学基金项目(51476027)。

Fault diagnosis of chillers using sparsely local embedding deep convolutional neural network

LIU Xuting1,2, LI Yiguo1,2, SUN Shuanzhu3, LIU Xichui1,2, SHEN Jiong1,2   

  1. 1. School of Energy and Environment, Southeast University, Nanjing 210096, Jiangsu, China;
    2. Key Laboratory of Energy Thermal Conversion and Control of Ministry of Education, Southeast University, Nanjing 210096, Jiangsu, China;
    3. Jiangsu Frontier Electric Technology Co. Ltd., Nanjing 211102, Jiangsu, China
  • Received:2018-07-02 Revised:2018-09-12 Online:2018-12-05 Published:2018-12-05
  • Supported by:

    supported by the National Natural Science Foundation of China (51476027).

摘要:

针对于冷水机组提出一种基于稀疏局部嵌入深度卷积网络(sparsely local embedding network,SLENet)的故障诊断方法。采用稀疏局部嵌入方法代替卷积核,对输入数据进行特征选择,避免了复杂的训练和调参过程。另外采用空间金字塔最大池化作为网络的输出层,减少了网络的输出维数和分类器的计算量。针对美国采暖、制冷与空调工程师学会提供的冷水机组的典型故障数据进行分类,结果表明,该方法相比深度卷积网络(CNN)和支持向量机(SVM)方法具有更高的故障诊断精度。

关键词: 算法, 神经网络, 安全, 故障诊断, 稀疏局部嵌入, 深度卷积网络, 空间金字塔最大池化

Abstract:

A fault diagnosis method for chillers based on a sparsely local embedding deep convolutional neural network (sparsely local embedding network, SLENet) is proposed. The sparsely local embedding filters are adopted for feature selection in the first two layers of SLENet, as a result complicated training and adjusting process are avoided. In addition, the spatial pyramid max pooling layer is employed to construct the output layer of SLENet to reduce the output dimensions of the proposed network, and then alleviate the computationally burden of classification. The experimental data from ASHRAE RP-1043 are used to validate the fault diagnosis method. The results demonstrate that the method achieves high diagnostic accuracy compared to convolutional neural network (CNN) and support vector machine (SVM) methods.

Key words: algorithm, neural networks, safety, fault diagnosis, sparsely local embedding, deep convolutional neural network, spatial pyramid max pooling

中图分类号: