Data-driven Prognostics for PEM Fuel Cell Degradation by Long Short-term Memory Network

Proton exchange membrane fuel cell (PEMFC) degradation prediction is essential especially in transportation applications, since one of the major issues that hinder its worldwide commercialization is represented by its durability. However, due to the complex physical phenomena inside the fuel cell, which are strongly inter-coupled, the conventional semi-empirical model-based prognostics approach may fail to predict the aging phenomena under varies fuel cell operating conditions. In order to improve prognostics accuracy, this paper proposed a data-driven approach to predict the fuel cell performance based on the long short-term memory (LSTM) recurrent neural network (RNN). Compared with traditional RNN, LSTM can be used to avoid gradient exploding and vanishing problems. Such a prediction model for the short-term memory can last for a long period of time, which makes LSTM suitable for time series forecasting. In order to validate the performance of the proposed LSTM approach, two different types of PEMFC along with five aging experimental data sets have been used. The results show that the proposed LSTM approach can accurately predict PEMFC degradation. An accurate degradation prediction plays an important role in PEMFC optimization used in transportation applications.