Fault Diagnosis Method Of PEMFC System Base On Data-driven

Proton exchange membrane fuel cell(PEMFC),as a new type of efficient green energy device,has been highly concerned by all countries in the world and is at the forefront of research.PEMFC has been widely used in the field of household cars,heavy trucks,locomotives and trams.However,in the complex and changeable application scenarios,the PEMFC system is prone to various failures,which lead to the degradation of stack performance and seriously affect its commercial development.Therefore,aiming at solving the fault problem of the PEMFC system in practical application,it is of great significance to study the corresponding fault diagnosis strategy to ensure the safe and stable operation of the system.In the background of fault detection and isolation,this paper conducts the research of fault diagnosis method of the PEMFC system based on data-driven framework.The main contents of this paper are as follows:(1)In order to solve the problem of fault identification of high-power PEMFC system for trams,the working principle and the characteristics of typical fault types of high-power water-cooled PEMFC system are studied,including the leakage of hydrogen,the low pressure of deionized water humidification pumps,deionized ethylene glycol inlet high temperature,signal voltage over-range of deionized ethylene glycol outlet temperature and the low pressure of input air.By cleaning the data collected during the actual operation of the tram,the original data sample sets containing five types of faults are obtained.(2)For the purpose of analyzing the characteristics of water fault in PEMFC stack,the change trend of output voltage of fuel cell under the state of water flooding and membrane drying faults is analyzed,according to the experimental data.On this basis,in order to fully mine the feature information of fault time series data,a feature extraction method based on twelve statistical theories is proposed,which includes absolute energy value,first-order differential absolute sum and autoregressive coefficient,etc.This method realizes the comprehensive extraction of fault features.Multiple data samples under normal,flooding and membrane drying fault conditions are obtained.(3)Considering the fault classification and recognition problem of the high-power PEMFC system,a fault classification method based on random forest(RF)combined classifier is proposed.This method resampled based on Bootstrap method to generate multiple sample subsets.Multiple classifiers models were constructed by CART algorithm in the decision tree concept.At last,the optimal classification results were selected by voting as prediction categories.Through case analysis,compared with the classification results of support vector machine(SVM)and k-nearest neighbor(KNN),RF method can effectively identify five fault categories in the actual operation of the hybrid tramway with the classification accuracy up to 94.4%.The result shows that RF method can obviously increases classification accuracy.(4)Aiming at the problem of fault diagnosis of PEMFC water management,a fault diagnosis method based on truncated singular value decomposition(TSVD)and light gradient boosting machine(LightGBM)is proposed to realize the fault recognition of water flooding and membrane drying faults.TSVD algorithm is used to compress data and reduce feature dimension,so as to reduce the complexity of calculation time.Then,the LightGBM classification and recognition model is constructed to realize accurate and fast fault diagnosis.The feasibility of the novel method is verified by measured data,and the result shows that the proposed method can diagnose the normal,flooding and membrane drying fault states quickly and accurately with the overall accuracy up to 98.7%.Compared with the diagnosis results of decision tree algorithm,the proposed algorithm is more suitable for fault diagnosis of PEMFC water management subsystem.