Research on an improved lp -RWMKEELM fault diagnosis model
-
摘要: 本文针对装备各类故障样本分布不平衡、现有算法故障诊断精度较低的问题,定义了一种p范数约束下正则化加权多核集成超限学习机的故障诊断模型。该模型在p范数约束下,基于各类故障样本自身规模,分别进行了两种自适应的样本权重分配;同时将多核学习(Multiple Kernel Learning,MKL)的多源数据融合能力和超限学习机(Extreme Learning Machine,ELM)运算高效的特点相结合,将样本的权重W融入到多核超限学习机的优化目标函数中;通过Adaboost集成策略,自适应提升富含信息的样本在模型中的权重,从而显著提升故障诊断的精度。以6个UCI公共数据集以及1个实装案例为例,进行了故障诊断实验。结果表明,模型与KELM、WKELM(W(1)和W(2)加权方式)以及多核超限学习机l1-MKELM、lp-MKELM、TRMKELM相比,诊断精度有显著提升;范数约束形式对模型的诊断性能影响有限。Abstract: This paper defines a fault diagnosis model based on regularized weighted multiple kernel ensemble extreme learning machine under p-norm constraints for the problems of imbalanced distribution of various types of fault samples and the low accuracy of fault diagnosis of existing algorithms. Under the p-norm constraint, the model performs two kinds of adaptive sample weight allocation based on the size of various types of fault samples. At the same time, it combines the multi-source data fusion ability and extreme learning machine of multiple kernel learning (MKL) combined with the characteristics of high efficiency of Extreme Learning Machine (ELM) operation, the weight W of the sample is integrated into the optimization objective function of the multiple kernel extreme learning machine. Through the Adaboost integration strategy, the weight of the information-rich samples in the model is adaptively increased, thereby significantly improving the accuracy of fault diagnosis. Taking 6 UCI public data sets and 1 actual installation case as examples, fault diagnosis experiments were carried out. The results show that compared with KELM, WKELM (W(1) and W(2) weighted methods) and l1-MKELM、lp-MKELM、TRMKELM, the diagnostic accuracy is significantly improved. The diagnostic performance of the model is limited。
-

计量
- 文章访问数: 412
- HTML全文浏览量: 33
- PDF下载量: 4
- 被引次数: 0