Solid Oxide Fuel Cell (SOFC) are complex systems in which gas-phase mass transport, heat transfer, ionic conduction, chemical reactions and electrical conduction take place simultaneously. Therefore, reliable simulation tools are needed to control and optimize their operation. Machine Learning (ML) can quickly estimate and generalize the relationship between the input values and the output values in a process. ML algorithms are used in various applications such as modelling, simulation, optimization, control, signal processing, pattern recognition, up to systems like power, production and renewable energy systems. Many methods help the successful design of algorithms for SOFC systems. However, a few researchers have studied and compared regression algorithms. This paper includes an in-depth study to compare the two efficient ML algorithms – namely Random Forest (RF) and Support Vector Regression (SVR). These algorithms are used to predict the performance of a SOFC cell. The algorithms were generated by the experimental data which are measured by using the different temperatures and hydrogen flow rates. Additionally, the effect of the amount of pure hydrogen and the total amount of hydrogen in the content of the fuel mixtures, which fed to the anode side of the SOFC, on the experimental voltage were compared. The experimental data set used for developing the model consists of 1272 records regarding the SOFC operated under different operating conditions. 1122 records of the experimental data set are used for training the regression algorithms mentioned above. Accordingly, the algorithms are tested and then, the experimental data are compared to the results generated by the algorithms to validate prediction performances. The model predicts the cell performance (output voltage) in approximately 0.52 s with the mean absolute percentage errors 1.97% for the RF algorithm and as low as 0.92% for the SVR algorithm. In this article, the SVR algorithm is identified as the most promising model. The process parameters effects on the variation of the output voltage of the SOFC can be examined when the developed models are proven reliability and precision after test with unknown data.
Comparison of support vector regression and random forest algorithms for estimating the SOFC output voltage by considering hydrogen flow rates
Baldinelli A.;Bidini G.
2020-01-01
Abstract
Solid Oxide Fuel Cell (SOFC) are complex systems in which gas-phase mass transport, heat transfer, ionic conduction, chemical reactions and electrical conduction take place simultaneously. Therefore, reliable simulation tools are needed to control and optimize their operation. Machine Learning (ML) can quickly estimate and generalize the relationship between the input values and the output values in a process. ML algorithms are used in various applications such as modelling, simulation, optimization, control, signal processing, pattern recognition, up to systems like power, production and renewable energy systems. Many methods help the successful design of algorithms for SOFC systems. However, a few researchers have studied and compared regression algorithms. This paper includes an in-depth study to compare the two efficient ML algorithms – namely Random Forest (RF) and Support Vector Regression (SVR). These algorithms are used to predict the performance of a SOFC cell. The algorithms were generated by the experimental data which are measured by using the different temperatures and hydrogen flow rates. Additionally, the effect of the amount of pure hydrogen and the total amount of hydrogen in the content of the fuel mixtures, which fed to the anode side of the SOFC, on the experimental voltage were compared. The experimental data set used for developing the model consists of 1272 records regarding the SOFC operated under different operating conditions. 1122 records of the experimental data set are used for training the regression algorithms mentioned above. Accordingly, the algorithms are tested and then, the experimental data are compared to the results generated by the algorithms to validate prediction performances. The model predicts the cell performance (output voltage) in approximately 0.52 s with the mean absolute percentage errors 1.97% for the RF algorithm and as low as 0.92% for the SVR algorithm. In this article, the SVR algorithm is identified as the most promising model. The process parameters effects on the variation of the output voltage of the SOFC can be examined when the developed models are proven reliability and precision after test with unknown data.I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.