A Prognostic Framework for PEMFC Based on Least Squares Support Vector Regression-Particle Filter

Proton exchange membrane fuel cell (PEMFC), a promising energy device, has been widely used thanks to its high energy density and low pollutant emissions. Prognostics of PEMFC has attracted increasing attention nowadays, which aims at predicting the health degradation of PEMFC and estimating the residual useful life (RUL) for better management. However, majority existing prognostic methods for PEMFC are restricted for real applications due to the following reasons: 1) accurate models which rely on complex underlying mechanism are hard to obtain; 2) data-driven methods usually depend on large amounts of data; 3) the prediction results are usually an estimated value rather than a probability distribution of RUL. Aiming at solving these problems, this paper proposes a prediction framework for PEMFC based on least squares support vector regression-particle filter (LSSVR-PF). As an intelligent prognostic method, LSSVR can achieve a good performance with limited training samples. The predicted values of LSSVR act as new observations of PF to obtain a probability distribution of the RUL. Effectiveness of the proposed method is verified based on PEMFC dataset provided by FCLAB Research Federation. The results confirm that the proposed method has a high accuracy of RUL prediction for PEMFC.