China Mechanical Engineering ›› 2021, Vol. 32 ›› Issue (18): 2153-2158,2164.DOI: 10.3969/j.issn.1004-132X.2021.18.002

Previous Articles     Next Articles

Ensemble Variable Predictive Model Based on Optimal Features and Its Applications

LIU Xiaofeng;TAN Qi;YE Rongting   

  1. College of Mechanical and Vehicle Engineering,Chongqing University,Chongqing,400044
  • Online:2021-09-25 Published:2021-10-13



  1. 重庆大学机械与运载工程学院,重庆,400044
  • 作者简介:刘小峰,女,1980年生,教授、博士研究生导师。研究方向为信号处理、智能算法、大数据和机器学习、预测与健康管理以及智能故障诊断。发表论文50余篇。。
  • 基金资助:

Abstract: Aiming to the problems of low recognition accuracy of conventional variable predictive model resulting in case of small samples, an integrated variable predictive model was proposed based on the optimal selection of recurrence quantification features. Some feature subsets with high weight and low redundancy optimally were formed and then the best feature subsets were selected using the embedded rater. The Gauss function, radial basis function and generalized regression functions were introduced to establish the complex nonlinear interaction relationships among the selected features. The newly established models and conventional models were integrated based on the fitting errors of each model. The experimental results show that the proposed method has higher accuracy and better stability in the bearing fault diagnosis compared with the conventional methods, especially in case of small samples. 

Key words: optimal feature selection, variable predictive model, weighted integration, fault diagnosis

摘要: 针对传统基于变量预测模型的模式识别方法在小样本情况下识别精度较低的问题,提出一种基于递归量化特征优化选择的集成变量预测模型(EVPM)。该模型采用递归量化分析对被分析信号进行特征提取,在交错式最大权值最小冗余规则下选择权重高且冗余度低的优化特征子集,引入高斯函数、径向基函数、广义回归函数建立特征变量间的复杂非线性关系模型,根据各模型的拟合误差进行自适应加权集成。实验结果表明,基于优化特征EVPM的模式识别方法即使是在小样本的情况下也能够有效地进行滚动轴承故障类型与故障程度的自动识别,且在精度和稳定性上明显优于同类型传统方法。

关键词: 特征优化选择, 变量预测模型, 加权集成, 故障诊断

CLC Number: