Performance Evaluation of Neural Network Models for Fuel Cell Power Prediction in Hybrid Electric Vehicles

Abstract

This paper evaluates the performance of different neural network models: Fitting Net, Pattern Recognition Net, and NARX Net (Nonlinear Autoregressive with Exogenous Inputs) in predicting fuel cell power in hybrid electric vehicles (HEVs) when integrated with an energy management system optimized by a genetic algorithm. The models were trained and tested using a comprehensive dataset consisting of 15 drive cycles under varying scenarios. The evaluation metrics include Mean Absolute Error (MAE), Root Mean Square Error (RMSE), Relative Absolute Error (RAE), and correlation coefficient. The results indicate that while all models perform adequately, the NARX Net demonstrates superior predictive accuracy, achieving the lowest MAE, RMSE, and RAE, as well as the highest correlation coefficient. Additionally, the NARX Net showed the lowest hydrogen consumption among the tested models, suggesting its effectiveness in real-time energy management for HEVs. The adaptive nature of the trained networks allows for dynamic adjustments to changing driving conditions, further enhancing their practical utility in real-world applications.