Fuel consumption modeling for automotive engines: Physics-based and data-driven approaches

The demand for energy use in vehicle engines has increased steadily in recent years, leading to growing environmental and social concerns. This article provides a comprehensive review and critical assessment of existing methods for estimating vehicle fuel consumption, with a particular focus on their underlying principles, practical applicability, and limitations. Both measurement-based approaches and computational modeling techniques are examined. The analysis shows that direct measurement methods, although reliable, are often complex and costly, which limits their scalability and long-term application. Physics-based models (white-box) rely heavily on mathematical representations of engine and vehicle dynamics and are therefore sensitive to nonlinear effects, parameter uncertainty, and variations in operating conditions. In contrast, data-driven approaches (black-box and gray-box) have demonstrated a strong capability in learning complex relationships from data and providing accurate fuel consumption predictions without detailed knowledge of the system’s structure. However, these methods typically require large and diverse datasets, and their performance may degrade if model parameters are not updated to reflect changing operating conditions. Overall, existing modeling approaches exhibit both strengths and inherent limitations. Addressing these limitations remains a crucial motivation for future research aimed at enhancing the accuracy, robustness, and practical usability of fuel consumption prediction models.