Micro- to Macro-Level Fuel Consumption Modeling of Diesel Heavy-Duty Trucks: A Hierarchical Bayesian Approach

The transportation sector faces growing pressure to reduce fuel consumption, driving the need for highly accurate and stable energy-use prediction models. However, developing these models is often constrained by limited data for certain vehicle categories, leading to prediction bias. This study introduces a Bayesian hierarchical modeling approach for predicting fuel consumption, leveraging extensive data from trucks operating throughout China. We assess the model’s applicability across varying data durations (from 100 to 100,000 s) and across analytical levels, from micro (detailed vehicle-level data) to macro (aggregate data). Our approach is compared with conventional models, including ordinary least squares regressions for individual vehicles and fleet-level data. Findings indicate that the Bayesian hierarchical model provides robust estimates, performing well in scenarios that require detailed, vehicle-specific data and in broader contexts with high-level aggregate data. Additionally, this model effectively mitigates prediction bias in small samples while preserving unique fuel consumption characteristics for each vehicle. This paper thus offers a strong framework for precise fuel consumption prediction in heavy-duty trucks, particularly advantageous in small-sample contexts.