A Hybrid Machine Learning Approach to Energy Consumption and Road Emissions Modeling of CNG Vehicles Based on Chassis Dynamometer Data and Road Load Power

This study presents a comparative analysis of energy consumption and gaseous emissions from a compressed natural gas (CNG)-fueled vehicle under real driving emissions (RDE) conditions and values predicted using machine learning (ML) models developed from chassis dynamometer data. The analyzed components included energy consumption (EC) as well as carbon dioxide (CO₂), carbon monoxide (CO), total hydrocarbons (HC), methane (CH₄), and nitrogen oxides (NO_X). The models were trained using a limited set of easily accessible predictors, namely vehicle speed and acceleration. A hybrid modelling approach was proposed, combining laboratory data with validation under real-world conditions. Additionally, road load power (P_rl) was introduced as a novel predictor representing vehicle operating load. The results demonstrate that the models effectively capture emission trends, with the highest agreement obtained for CO, CO₂. The inclusion of P_rl improved prediction accuracy, which increased from approximately 64% to 71% for CO and from 57% to 61% for HC. For CO₂, the model achieved about 80–82% agreement with RDE measurements, with analogous levels obtained for EC. A key advantage of the proposed methodology is its reliance on a limited number of input variables, which enhances practical applicability while maintaining satisfactory accuracy. Furthermore, the use of precise laboratory data improves model robustness, and the approach enables the estimation of methane (CH₄), which is typically not measured by standard portable emissions measurement systems (PEMSs). The results confirm the effectiveness of the hybrid ML framework and highlight the importance of incorporating load-related parameters in real-world emissions and energy consumption modeling.