The article introduces an innovative probabilistic algorithm designed to predict automotive fuel cell performance degradation. Traditional point estimation algorithms struggle with the uncertainty caused by external factors and changing conditions. The proposed algorithm combines the diffusion model and Transformer to quantify this uncertainty effectively. By incorporating a self-guiding sampling method and integrating Transformer architecture, the model achieves enhanced generation control and noise reduction. Comparative analysis highlights the superior performance of the guidance method and prediction model, with a notable 35.10% improvement in prediction accuracy. The study emphasizes the importance of accurate and robust predictions to extend the lifespan of PEMFCs. The Prognostics and Health Management (PHM) technique plays a vital role in monitoring and predicting performance degradation to enable timely maintenance and prevent fatal damage. The complexity of automotive fuel cells, involving multiple technologies, presents challenges in understanding degradation mechanisms. Various predictive methods have been proposed, combining data-driven and mechanistic approaches. Data-driven methods utilize historical data for prediction, while mechanistic models simplify the process by making constant assumptions. The article showcases different predictive models and methods used to forecast fuel cell performance degradation, emphasizing the significance of accurate predictions for extending service life in the hydrogen energy sector.