基于深度自编码器网络的逆变器开关管开路故障诊断

Open-circuit fault diagnosis of switch tube in inverters based on deep autoencoder network

逆变器开关管的开路故障是影响逆变器系统安全、可靠运行的关键问题,为此研究了基于深度自编码器网络的逆变器开关管开路故障诊断。首先,给出了逆变器开关管开路故障的模型分析和故障编码,并通过仿真实验平台采集了不同开路故障情况下的原始故障信号数据。为避免人工进行网络结构设计的繁冗工作,提出一种自编码器网络的规则化设计方法。该方法能够快速确定隐含层神经元的数量和网络深度,利用故障数据自动学习开关管的故障特征,实现端到端的智能故障诊断。对比实验表明,提出的自编码器网络能够自动挖掘故障信号中的关键特征信息,对不同类型的故障信号数据均能得到稳定的故障识别率,验证了所提方法的有效性和优越性。

The open-circuit fault diagnosis of the switch in inverters is of critical importance for the safe and reliable operation of the inverter system.This paper studied the open-circuit fault diagnosis of inverters based on deep autoencoder network.It first presented the model analysis and fault coding for the open-circuits fault of inverters and collected the original fault data in different cases through a simulation platform.To avoid the tedious work of manual network design,it then proposed a new design method for the autoencoder network.This method could determine the amount of the hidden neurons and the depth of the network quickly and extract the fault features of the switch tube by using the fault data,realizing the intelligent end-to-end fault diagnosis.The results of several comparison experiments indicate that the proposed autoencoder network can mine the key features from the fault data automatically and achieve reliable fault identification based on different kinds of fault data,thus verifying the effectiveness and superiority of the new method.