Machine learning modeling for proton exchange membrane fuel cell performance

Proton exchange membrane fuel cell (PEMFC) is considered essential for climate change mitigation, and a fast and accurate model is necessary for its control and operation in practical applications. In this study, various machine learning methods are used to develop data-based models for PEMFC performance attributes and internal states. Techniques such as Artificial Neural Network (ANN) and Support Vector Machine Regressor (SVR) are used to predict the cell voltage, membrane resistance, and membrane hydration level for various operating conditions. Varying input features such as cell current, temperature, reactant pressures, and humidity are introduced to evaluate the accuracy of the model, especially under extreme conditions. Two different sets of data are considered in this study, which are acquired from, a physics-based semiempirical model and a 1-D reduceddimension Computational Fluid Dynamics model, respectively. The aspect of data preprocessing and hyperparameter tuning procedures are investigated that are extensively used to calibrate the artificial neural network layers and support vector regressor to predict the fuel cell attributes. ANN clearly shows an advantage in comparison with SVR, especially on a multivariable output regression. However, the SVR is advantageous to model simple regressions as it greatly reduces the level of computation without sacrificing accuracy. Data-based models for PEMFC are successfully developed on both the data sets by adapting advanced modeling techniques and calibration procedures such as ANN incorporating the dropout technique, resulting in an R2 & GE; 0.99 for all the predicted variables, demonstrating the ability to build accurate data-based models solely on data from validated physics-based models, reducing the dependency on extensive experimentation.

Automated summary

This study develops data-based models for the performance attributes and internal states of Proton Exchange Membrane Fuel Cells (PEMFC) using various machine learning methods. Techniques such as Artificial Neural Network (ANN) and Support Vector Machine Regressor (SVR) are employed to accurately predict cell voltage, membrane resistance, and membrane hydration level under different operating conditions. The results show that ANN has an advantage in multivariable output regression, while SVR is beneficial for simple regressions. By incorporating advanced modeling techniques and calibration procedures, accurate data-based models for PEMFC can be built solely on validated physics-based models, reducing the need for extensive experimentation.