Fault detection based on integrated kernel entropy component analysis algorithm

Clc Number:

Fund Project:

  • Article
  • |
  • Figures
  • |
  • Metrics
  • |
  • Reference
  • |
  • Related
  • |
  • Cited by
  • |
  • Materials
  • |
  • Comments

    Aiming at the problem of kernel parameter selection in kernel entropy component analysis (KECA), an fault detection algorithm based on ensemble kernel entropy component analysis (EKECA) is proposed. Firstly, a series of kernel functions with different parameters are selected to project the nonlinear data into the kernel feature space, and the eigenvalues and eigenvectors that contribute a lot to Renyi entropy are selected to obtain the score matrix, and the entropy component analysis model of multiple subkernels is established; Then, the kernel matrix of the original data is projected onto the subspace formed by the principal axis of KPCA, and the statistics of each sub KECA model are calculated to obtain the detection results; Finally, the detection results of each KECA sub model are converted into probability by Bayesian decision, and the total probability of the detection results is calculated by ensemble learning method to get the statistics and judge whether it is beyond the control limit. The algorithm is applied to a numerical example and the TE process. Simulation results show that the algorithm can effectively improve the fault detection rate and reduce the false alarm rate compared with traditional EKPCA, KECA and other algorithms. This method provides a reference for the traditional KECA algorithm to solve the problem of selecting different fault kernel parameters.

    Cited by
Get Citation
Article Metrics
  • Abstract:
  • PDF:
  • HTML:
  • Cited by:
  • Received:June 03,2021
  • Revised:June 21,2021
  • Adopted:November 08,2021
  • Online:
  • Published: