Fuel Cell Performance Prediction under Variable Conditions Based on Extented Kalman Filter and Random Forests

Fuel cell is a promising alternative to battery in heavy duty mobility. Though, efforts are needed for the development of reliable fuel cell aging models that could be integrated in control strategies. The present paper presents a voltage prediction model based on an Extended Kalman Filter that track the key parameters of the different losses of the fuel cell quasi-static model. The ohmic resistance is predicted thanks to a random forests algorithm taking into account the operating variables. The proposed method is evaluated on experimental data from an aging campaign of a 1 kW stack with a heavy transportation profile. The prediction capacity of the method is evaluated for several learning duration. It shows very good results with less than 2 % errors from 400 h of learning. The ohmic resistance modeling based on random forests gives a relevant estimation with about 4 % error, even for short learning duration.