A particle filter and long short term memory fusion algorithm for failure prognostic of proton exchange membrane fuel cells

Abstract

Prognostics and Health Management (PHM) appears to be a promising maintenance strategy which can enhance reliability and reduce maintenance costs of the target system. In the process of PHM, Prognostics is the most important and crucial. Prognostic approaches can be roughly divided into two categories: model-based methods and data-driven methods, both of which have advantages and limitations. To overcome the limitations of these methods and improve the accuracy and precision of the forecasting, we propose a novel fusion prognostic method. This method fuses the Particle Filter (PF, model-based) and Long Short Term Memory (LSTM, data-driven) algorithms. In the literature, PF is used by estimating the system state and identifying the parameters of the model for the purpose of Prognostic. However, it does not have ideal performance due to the lack of measurements in the prediction phase. To solve this problem, LSTM is used to forecast the measurements and use the results as the observation of PF. The experiment is applied to the data of Proton Exchange Membrane Fuel Cell Stack from IEEE PHM 2014 Data Challenge. The results demonstrate that the proposed method can effectively integrate the advantages of PF and LSTM, which leads to a better forecasting performance than naive PF approach.