Long-Term Performance Prediction of PEMFC Based on LASSO-ESN
Kai He, Lei Mao, Jianbo Yu, Weiguo Huang, Qingbo He, Lisa Jackson
Abstract
In recent years, with wide application of proton exchange membrane fuel cell (PEMFC) in vehicles and portable applications, researches regarding PEMFC lifetime behavior and associated prognostic techniques receive more interest. In this article, a least absolute shrinkage and selection operator-echo state network (LASSO-ESN)-based prognostic strategy is proposed for the optimization of input parameters and long-term PEMFC behavior prediction. In the analysis, ESN is selected to predict PEMFC long-term behavior iteratively, while input parameters to ESN are optimized using LASSO. With LASSO, the contribution of input parameters to PEMFC prediction can be evaluated, and those with the minimum weight are eliminated iteratively during the prediction. From the findings, the most accurate predictions and corresponding optimized input parameters can be determined. Furthermore, effectiveness of proposed strategy is investigated using PEMFC data at different operating conditions. Results demonstrate that with proposed strategy, optimized input parameters at different operating conditions can be determined, and accurate PEMFC predictions can be provided.