PHYSICS-INFORMED REPRESENTATION LEARNING FOR HYBRID ELECTRIC VEHICLE ENERGY MANAGEMENT

Abstract

Integrating physics-based and learning systems enhances fuel cell hybrid electric vehicles (FCHEVs) for better performance control and efficient power source operation. Balancing this diverse mix is challenging, given the uncertainties and fluctuations in complex physics-based modeling. In this framework, our work has a dual purpose. Firstly, precise physics-based modeling enables the effective generation of data. This helps gather diverse data resembling real-world scenarios, aiding in drawing reliable conclusions. Several known driving cycles were utilized to generate sufficient data for the experiments and findings presented in this work. Secondly, the collected data undergoes an advanced representation learning process with adaptive functions, enhancing the interaction between learning models and the FCHEV system's physical phenomena. The effectiveness of the suggested approach is validated through a comprehensive evaluation of developed algorithms using various visual and numerical metrics. In a comparative analysis, the results illustrate the efficacy of the methodology in addressing energy management (EM) challenges in fuel cell hybrid electric vehicles (FCHEVs).