Lifetime prediction of proton exchange membrane fuel cell based on long short-term memory neural network optimized by improved Grey Wolf algorithm

Le résumé fourni par la source

Proton exchange membrane fuel cells (PEMFCs) are widely employed in stationary and mobile power applications due to their zero-pollution emissions and high power generation efficiency. Insufficient service life remains a key obstacle to large-scale commercial deployment. Predictive and health management (PHM) techniques can estimate the remaining useful life (RUL) of PEMFC systems, thereby preventing catastrophic failure. Long Short-Term Memory (LSTM) neural networks represent one of the most prevalent forecasting methodologies. This study employs voltage as the health metric for PEMFCs, proposing a global optimisation approach for LSTM networks based on a multi-strategy enhanced grey wolf optimisation algorithm. This is combined with wavelet threshold denoising to filter voltage decay signals, thereby establishing a PEMFC lifespan prediction methodology. Results demonstrate that compared to the Grey Wolf Optimisation-based LSTM neural network (GWO-LSTM), the Multi-Strategy Grey Wolf Optimisation-based LSTM neural network (MSGWO-LSTM) reduces multiple prediction errors by approximately threefold under steady-state conditions and by approximately one to twofold under dynamic conditions. Concurrently, training time is halved, achieving simultaneous enhancements in prediction accuracy and efficiency. Particularly in dynamic scenarios, the MSGWO-LSTM neural network demonstrated outstanding long-term forecasting performance.

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.