基于改进卷积神经网络的燃气调压器故障识别研究

针对传统故障识别方法不仅过分依赖专家经验对故障特征进行提取且识别准确率不高的问题,在深度学习理论基础上,提出了一种将一维卷积神经网络与SVM分类器相结合的改进深度卷积神经网络,实现调压器“端到端”的故障识别。首先,介绍了传统卷积神经网络结构;其次,将改进后的一维卷积神经网络与SVM相结合,提出了基于1-MsCNN-SVM算法的调压器故障识别模型,并对模型的组成部分进行了介绍;然后,通过对比实验确定了模型的卷积核长度和卷积层组数;最后,为验证模型的有效性,基于燃气调压器故障数据集,开展了燃气调压器故障识别研究。研究结果表明,改进后的1-MsCNN-SVM算法故障识别准确率高达99.20%,模型具有较好的分类准确率。

Fault identification of gas pressure regulator based on improved convolutional neural network

Aiming at the problems of traditional fault identification method which not only relies on expert experience excessively to extract fault features,but also has low identification accuracy,on the basis of deep learning theory,an improved deep convolutional neural network combining one-dimensional convolutional neural network with SVM classifier was proposed,the end-to-end fault identification of voltage regulator was realized.First,the structure of traditional convolutional neural network was introduced.Then,combining the improved one-dimensional convolutional neural network with SVM,a fault identification model of voltage regulator based on 1-MsCNN-SVM algorithm was proposed,and the components of the model were introduced.Furthermore,the convolutional kernel length and convolutional layer group of the model were determined by experimental comparison.Finally,in order to verify the effectiveness of the new model,based on the fault data set of gas pressure regulator,the fault identification research of gas pressure regulator was carried out.Experimental results show that the accuracy of the improved 1-MsCNN-SVM algorithm is up to 99.20%,the model has a better classification accuracy.