Block structure optimization in PEMFC flow channels using a data-driven surrogate model based on random forest

Abstract

ABSTRACT Rational design of flow channel can promote the removal of liquid water and effectively improve the performance of proton exchange membrane fuel cell (PEMFC). Compared with the conventional parallel straight channels, the block flow channel has a better water removal effect and is more beneficial to the fuel cell performance. In this study, a random forest-trained surrogate model is combined with a genetic algorithm to achieve optimization of the block structure applied in a novel two-block flow channel. A three-dimensional multiphase model based on the volume of fluid (VOF) method is employed to obtain the data sets by varying the block structure (length, height, and width) which are used to train a random forest-based data-driven surrogate model for predicting the water removal time (T) and the maximum pressure drop (∆P) of the channel. Then, the data-driven surrogate model is integrated into the genetic algorithm to optimize the design of the block. The results show that the output values of the surrogate model trained by random forest fit well with the actual values of the physical model, indicated by small mean square error (MSE) of T and ∆P which are 0.1191 and 0.0209, respectively. After optimization by the genetic algorithm, the block parameters of 0.6 mm in length, 0.375 mm in width, and 0.75 mm in height are found to be optimal, and the results are validated by the physical model.