Research On The State Assessment Method Of Integrated Water-cooled Proton Exchange Membrane Fuel Cell System Integrating Multi-source Perceptio

Due to the complex configuration of high-power integrated PEMFC systems,the associated system condition assessment remains a huge challenge.For the initial system failure,failure to detect and take necessary repair measures in time will further lead to serious system failure or even downtime,which will result in irreversible component failure or even system damage.Therefore,real-time condition assessment of high-power PEMFC systems,timely detection of potential safety hazards,effective fault warning,so as to avoid the deep development of faults and prevent major system operation accidents,has become one of the key problems that need to be solved in the current commercial application of high-power PEMFC systems.In this paper,we propose an online condition assessment architecture for high-power integrated PEMFC system based on random matrix analysis.The proposed architecture consists of two cascading processes,which are the data flow representation of the random feature matrix and the system state evaluation based on the random matrix analysis.The signal clustering transformation performed by the fusion of model-driven and data-driven methods allows the extraction of residuals characterizing the system anomalies to form the data flow stochastic matrix.Based on this,matrix dimension expansion is performed using random tensor augmentation in order to enhance the significance level of matrix eigenvalue distribution for matrix highdimensional analysis.Further,the eigenvalues of the random covariance matrix are updated by recursive eigenvalue updating method,and the system state assessment indicators are derived based on the M-P law and the circular law.Based on the proposed state assessment indicators,a practical process of online system state assessment incorporating multiple indicators is proposed,which is divided into two parts: In the prediction stage,STLES and TMFR status evaluation indicators based on the tensor augmented characteristic transformation matrix are used.In the validation phase,SLES and MFR status evaluation indicators based on the original data-driven feature transformation matrix are used.Taking temperature anomaly perception as an example,detailed experimental results show that the exported indicators are more sensitive to system anomalies than traditional threshold-based state evaluation methods,and can more effectively realize the online evaluation of the running state of integrated PEMFC systems under the proposed system state evaluation architecture.