Gradient Boosting Algorithm for Current-Voltage Prediction of Fuel Cells

Abstract

Catalyst development using trial and error methods by screening several experimental conditions is a conventional and old-fashioned method now. The progress of artificial intelligence (AI) is accelerating at a dramatic rate, and the results from AI have completely shifted scientific paradigms. We developed an overall performance prediction model for an alkaline fuel cell using a machine learning algorithm. From more than 80 I-V curves and 8000 data points, we selected dozens of input features and established models based on two error-scoring methods, which focus on operational conditions rather than catalytic characteristics. Both models exhibited high output predictions with high R2 values (> 0.95). The models further showed that, based on the top-ranked output features, the cathode side is more affected by the fuel cell even though the input data excluded information about the cathode part.

Keywords: performance prediction, alkaline fuel cell, machine learning, artificial intelligence, gradient boosting algorithm