基于门控循环单元神经网络的燃料电池系统故障监测

A Fault Monitoring Method of Fuel Cell System Based on the Gated Recurrent Unit Neural Network

质子交换膜燃料电池运行机理复杂,对工作环境要求高,在运行中可能出现多种类型故障.为了对燃料电池进行故障监测,保障系统正常运行,提高安全可靠性,采用门控循环单元神经网络建立燃料电池系统的输入输出模型,将燃料电池系统实际输出电压与门控循环单元神经网络的预测电压进行比较产生电压残差,对电压残差进行评价判断燃料电池系统有无故障.结果表明,该方法可以很好地拟合燃料电池的输入输出模型,合理选取电压残差阈值可有效监测燃料电池的运行状态是否正常.不同神经网络的实验表明,门控循环单元神经网络在网络参数规模上比全连接神经网络大大减少,比简单循环神经网络更容易训练.

The Proton Exchange Membrane Fuel Cell(PEMFC) has a complicated operating mechanism and high requirements on the working environment,thus various types of faults may occur on operation for PEMFC system.The fault monitoring of the fuel cell is significant to ensure the normal operation of the system and improve the reliability of the system.In this paper,a Gated Recurrent Unit(GRU) neural network is presented to build the model of PEMFC system.The residual voltages generated by the subtraction of fuel cell system voltages and predicted voltages of GRU neural network are processed to diagnose the faults and normal states.The results show that the proposed method can fit the model of PEMFC system well and be applied to distinguish between faults and normal states by a reasonable selection of voltage residual thresholds.Experiments with different neural networks have shown that gated recurrent unit neural networks have greatly reduced network parameter scales than fully connected neural networks,and are easier to train than recurrent neural networks.