Machine learning for predicting fuel cell and battery polarisation and charge–discharge curves

Energy Reports

F. Yu¹  A.A. Shah²  W.W. Xing²  P.K. Leung³ 

[1] Corresponding authors.;Key Laboratory of Low-grade Energy Utilization Technologies and Systems, MOE, Chongqing University, Chongqing 400030, China;School of Microelectronics, Beijing University of Aeronautics and Astronautics Address: No. 37 Xueyuan Road, Haidian District, 100191, China;

关键词: Fuel cells and batteries;    Machine learning;    Dimensionality reduction;    Gaussian process regression;    Support vector regression;    Deep learning;   

DOI  :  

来源: DOAJ

【 摘 要 】

Predictions of whole polarisation or charge–discharge curves for fuel cells and batteries using machine learning has rarely been explored, despite the now vast efforts to apply such methods to the analysis of these technologies. The aforementioned curves contain vital signatures of performance and degradation, and are usually therefore amongst the main targets for more traditional modelling techniques. In this paper, we investigate state-of-the-art multivariate machine-learning methods for predicting polarisation and charge–discharge curves. We employ a reduced-dimensional approach that enables powerful Gaussian process and support vector regression models to approximate multivariate outputs. Both models are shown to be able to accurately approximate polarisation and charge–discharge curves from complex physics-based models, including the sequence of times associated with charge–discharge cycles under constant load. We provide comparisons to standard deep networks, which fail to yield satisfactory results with the volume of data available. The methods developed in this paper can be used to in a variety of applications, including optimisation and battery health management.

【 授权许可】

Unknown