Abstract
The life and durability problems of proton exchange membrane fuel cell (PEMFC) have limited the commercialization process. The main reason for the degradation of life due to the frequent occurrence of local gas starvation in the dynamic process. The existing research is mainly carried out by experiment or simulation to diagnose local gas starvation, there is almost no research using machine learning methods to predict the local gas starvation through operating parameters. To solve this problem, a snake-shaped five-channel PEMFC model is established in this paper, and obtained source data through CFD simulation. Principal component analysis and k-means clustering algorithm are used to effectively define the local gas starvation state of each sample point, and complete sample labeling. Five operating parameters (temperature, pressure, humidity, gas stoichiometric ratio and current density), were used as model inputs. Three machine learning methods are chosen for training and prediction, and compare their accuracy. The prediction accuracy rate based on the extreme learning machine regression model is the highest, which is 93.49%, and have a fast prediction speed. It can quickly and accurately predict the local gas starvation state under a certain working condition, which has guiding significance for the optimization of op
Keywords Proton exchange membrane fuel cells, Diagnosis and prediction of local gas starvation, Machine learning
Copyright ©
Energy Proceedings