• 中国计算机学会会刊
  • 中国科技核心期刊
  • 中文核心期刊

计算机工程与科学

• 论文 • 上一篇    下一篇

基于ELM和连续过程神经网络的抽油机工况诊断

刘志刚1,许少华2,李盼池1   

  1. (1.东北石油大学计算机与信息技术学院,黑龙江 大庆 163318;2.山东科技大学信息科学与工程学院,山东 青岛 266590)
  • 收稿日期:2015-12-28 修回日期:2016-03-17 出版日期:2017-10-25 发布日期:2017-10-25
  • 基金资助:

    国家自然科学基金(61170132);黑龙江省教育厅基金(11551015)

Working condition diagnosis of oil pump machine based on
continuous process neural network and extreme learning machine

LIU Zhi-gang1,XU Shao-hua2,LI Pan-chi1   

  1. (1.School of Computer and Information Technology,Northeast Petroleum University,Daqing 163318;
    2.College of Information Science and Engineering,Shandong University of Science and Technology,Qingdao
    266590,China)
  • Received:2015-12-28 Revised:2016-03-17 Online:2017-10-25 Published:2017-10-25

摘要:

普通神经网络进行抽油机工况诊断时存在诊断精度偏低的问题,提出选用连续过程神经元网络作为诊断模型,特征输入选取
能直接反映示功图几何形态特征的位移和载荷两种连续信号。为提高模型学习速度,提出过程神经网络的极限学习算法,将
训练转换为最小二乘问题,根据样本输入计算隐层输出矩阵,使用SVD法求解Moore-Penrose广义逆,最后计算隐层输出权值
。通过诊断实验,模型学习速度提升5倍左右,与普通神经网络进行对比,诊断精度提高8个百分点左右,验证了方法的有效
性。

关键词: 工况诊断, 过程神经元网络, 极限学习, Moore-Penrose广义逆, 网络训练

Abstract:

To tackle the shortcoming of common neural networks of low diagnostic accuracy in the working condition
diagnosis of the oil pump machine, we propose a method that selects continuous process neural networks as the
diagnostic model. Because the geometric shape characteristics of the indicator diagram are directly reflected
by the continuous signals of move and load, we select them as the feature input of the diagnostic model. In
order to increase the learning speed, we introduce an extreme learning algorithm to process neural networks.
The model training process is converted to a least square problem. The output matrix of the hidden layer is
calculated according to the samples, the Moore-Penrose generalized inverse is solved by the SVD algorithm and
then the output weights of the hidden layer are calculated. Diagnostic experiments show that the learning
speed is increased by about 5 times, the diagnostic accuracy is improved by about 8 percentage in comparison
with common neural networks, and the validity of the method is verified.

 

Key words: working condition diagnosis, process neural network, extreme learning, moore-Penrose generalized inverse, network