基于Res-LSTM的燃料电池系统故障检测

A Fault Detection Method of Fuel Cell System Based on the Res-LSTM

质子交换膜燃料电池系统由多个子系统组成,其运行机理复杂,在运行中易发生多种故障,对其进行故障检测有利于控制策略调整,提高系统的可靠性.本文运用残差-长短期记忆网络(Residual Network with Long Short-Term Memory, Res-LSTM)对质子交换膜燃料电池系统的输入输出关系进行建模,运用基于模型的方法对燃料电池系统进行故障检测.实验结果表明,残差-长短期记忆网络能准确预测正常状态的燃料电池输出电压,通过电压残差阈值比较,可以进行故障检测.与长短期记忆网络相比,残差-长短期记忆网络的预测结果更精确,训练速度更快,电压残差阈值更小,从而可以获得更准确的故障检测结果.

The Proton Exchange Membrane Fuel Cell(PEMFC) system is composed of multiple subsystems, and it’s operating mechanism is complex, so various faults are prone to occur during operation.Fault detection is beneficial to the adjustment of control strategies and improves the reliability of the system.In this paper, the Residual Network with Long Short-Term Memory(Res-LSTM) is used to model the input-output relationship of the proton exchange membrane fuel cell system, and the model-based method is used to detect the faults of the fuel cell system.The experimental results show that the Res-LSTM can accurately predict output voltages of the fuel cell system in the normal state, and the voltage residual thresholds can be compared to detect faults.Compared with the Long Short-Term Memory(LSTM) network, the Res-LSTMs have more accurate prediction results, faster training speed and smaller voltage residual thresholds, thus more accurate fault detection results can be obtained.