Adaptive energy management strategy for fuel cell electric vehicles using online self-tuning neural networks

Abstract

Fuel Cell Electric Vehicles (FCEVs) are considered a promising technology for reducing carbon footprints. However, since these vehicles use more than one energy source—namely, a fuel cell system and a battery pack—an efficient energy management system is necessary. This study proposes a Self-Tuning Neural Network (STNN) Energy Management System (EMS) to improve the performance of a mid-size FCEV. The STNN adapts in real-time and without offline retraining to any changes in driving conditions while balancing new data with existing knowledge. A distinctive feature of the proposed STNN EMS is its ability to simultaneously (i) reduce fuel consumption, (ii) maximize fuel cell efficiency, and (iii) maintain the battery’s state of charge near a target level. The effectiveness of the STNN EMS is demonstrated by comparing its performance against three common EMSs: Power Following, Fuzzy Logic, and Non-Adaptive Neural Network in a simulation environment. The efficacy of the STNN EMS is demonstrated by its performance across various driving cycles, including the UDDS, NYCC, HWFET, and the Amman cycle, particularly it has improved fuel economy for UDDS, NYCC and Amman cycles by at least 4% compared to the baseline energy management systems.