基于改进深度置信网络的故障诊断方法  

Fault Diagnosis Method Based on Improved Deep Belief Network

作　　者：陈子兆 矫文成 孙慧贤 李月武 CHEN Zizhao;JIAO Wencheng;SUN Huixian;LI Yuewu(The Army Engineering University of PLA, Shijiazhuang 050003, China)

机构地区：[1]陆军工程大学石家庄校区,河北石家庄050003

出　　处：《探测与控制学报》2020年第4期98-105,共8页Journal of Detection & Control

摘　　要：针对深度置信网络对小样本数据集故障诊断时准确率低的问题,提出了基于改进深度置信网络的故障诊断方法。该方法通过优化网络特征提取能力,提升网络学习和分类能力以减少网络训练对数据的依赖程度;利用网络公开数据集测试改进的深度置信网络模型的性能;并将改进网络模型应用于小样本故障数据集上。实验验证结果表明,较基于传统网络模型的故障诊断而言,基于改进模型的方法通过添加新的隐藏层,强化了模型特征提取能力,提高了故障诊断准确率。Aiming at the problem of low accuracy of deep belief network for small sample data set fault diagnosis,a fault diagnosis method based on improved deep belief network was proposed.This method optimized the network feature extraction ability,enhanced the network learning and classification ability to reduce the dependence of network training on the data.This method used the open dataset of the network to test the performance of the improved deep belief network model and applies the improved network model to small sample faults dataset.The experimental verification results showed that,compared with fault diagnosis based on the traditional network model,the method based on the improved model strengthened the model feature extraction capability and improved the accuracy of fault diagnosis by adding new hidden layers.

关 键 词：故障诊断 深度置信网络 特征提取 自适应谐振 

分 类 号：TP206[自动化与计算机技术—检测技术与自动化装置]