Research On Modeling And Parameter Prediction Of Proton Exchange Membrane Fuel Cell

In recent years,research on fuel cells has become increasingly popular and is widely regarded as a potential solution to current environmental and energy problems.Fuel cells have far better endurance and energy conversion rate than traditional batteries.This paper focuses on the research of proton exchange membrane fuel cells,mainly focusing on its operation process,studying static and dynamic simulation models,voltage,current and other parameter prediction methods,laying the foundation for the control and operation optimization of proton exchange membrane fuel cells.First,this article establishes the static mechanism model of the electrochemical reaction of the fuel cell,and uses the optimization method to fit the hyperparameters in the empirical calculation formula in the electrochemical reaction.It is found through research that the method needs to meet several conditions: the fuel cell needs to maintain a specific Maintain a constant environment during internal structure and operation.However,researchers have designed a variety of fuel cells,and the actual environment is also very different from the laboratory environment,so the static mechanism model is difficult to deal with various complex situations in reality.At the same time,the research process found that temperature changes have a great impact on the fuel cell.Therefore,based on the static mechanism model and the thermodynamic characteristics of the fuel cell,a dynamic simulation model of the proton exchange membrane fuel cell was established.The simulation research results show that the dynamic model is more important than the static mechanism The model is closer to reality.Secondly,using the input and output variable data,the parameter prediction method of the proton exchange membrane fuel cell based on XGBoost is studied.The predicted parameters are: stack voltage and stack temperature.The data preprocessing method using the 3sigma principle and wavelet threshold denoising to remove gross errors and noise in the data,by comparing three modeling methods: extreme gradient ascent XGBoost,least square support vector machine LSSVM,deep confidence network DBN,simulation results It shows that the modeling algorithm based on XGBoost can improve the prediction accuracy of battery parameters and greatly reduce the model training time.Finally,using variable time series data,the fuel cell parameter prediction method based on long-term and short-term memory neural network LSTM is studied.The algorithm principle analysis shows that LSTM has a high accuracy of fuel cell parameter prediction.The simulation results on different test sets show that this method is suitable for both theoretical models and practical application scenarios,and can maintain good prediction accuracy and short model training time.