Research On Performance Consistency Of Automotive Fuel Cell Based On Data-driven Model

The PEMFC is widely used in the transportation due to its excellent energy efficiency,zero-emission,and high power density.Automotive fuel cell stacks are usually composed of hundreds of fuel cells connected in series,where voltage consistency between cells is an essential indicator of their performance and lifetime.During real-time operation,fuel cells suffer from local underperformance and uniformity deviations that reduce their reliability and lifetime.Model-based cell voltage uniformity studies facilitate the optimization of operating parameters and performance improvement.Concerning the complex stack geometry,the physics-based multi-dimensional models require tremendous computation resources and time,which are difficult to predict the voltage and uniformity of hundreds of fuel cells.In this thesis,the artificial neural network(ANN)model with high efficiency and online application is proposed to study the cell voltage uniformity of the self-designed 60 k W fuel cell stack with 140 fuel cells in both steady and dynamic states,which is conducive to the optimization of its operating parameters and the evaluation of its performance.The main research work is as follows:Firstly,the critical operating conditions affecting fuel cell voltage output are analyzed mechanistically,and steady-state and dynamic experiments under different operating conditions of automotive fuel cells are designed to obtain experimental data under different cathode outlet pressures and different load currents to provide data sets for steady-state cell voltage uniformity and dynamic cell voltage uniformity studies.Secondly,a steady-state ANN model is established,and the critical operating parameters that affect the performance and uniformity of fuel cells are selected as the input features of the steady-state ANN model according to the fuel cell voltage output model.The model structure is discussed,including the number of hidden layers and neurons.The finalized steady-state ANN model structure with six input features,two hidden layers,and 140-dimensional outputs was used for steady-state voltage and uniformity prediction of 140 fuel cells and compared with experimental results.Then the model’s generalization ability is verified to be used to predict the cell voltage and uniformity under high load current.Finally,to improve the computational efficiency of the model,the output dimension of the model is reduced by average grouping,which provides a theoretical basis for the model to be used for online prediction.Thirdly,the dynamic characteristics of the fuel cell are analyzed and the load current variation and reactant flow variation are added to the existing input features of the steadystate ANN model to build a dynamic ANN model.The effect of the above-added input features on the dynamic cell voltage prediction is also analyzed.The model is used to predict the dynamic voltage and uniformity of 140 fuel cells during load shedding and compared with the experimental results.The model is generalized to predict the dynamic voltage and uniformity of 140 fuel cells under different high cathode outlet pressures and loading processes.The ANN model proposed in this thesis can effectively predict the voltage and uniformity of fuel cells,which is beneficial for an accurate evaluation of fuel cell performance,health state,and reliability.