A Comprehensive Understanding of CO ₂ Emissions from Hybrid Vehicles Using Long Short-Term Memory

Insufficient exploration of hybrid electric vehicles (HEVs) in relation to carbon dioxide (CO 2 ) emissions may lead to significant policy-related problems. Previous studies emphasize comparisons between internal combustion engine vehicles and HEVs, rather than an exclusive focus on HEVs. A detailed investigation of HEV-derived CO 2 emissions is needed to understand their emission profiles across different road types. Moreover, the limited existing studies on HEV emissions have not considered trip lengths and spatial attributes, which are essential factors considering the operating functions of HEVs. This study aims to fill this gap by integrating data of engine characteristics, trip lengths, and spatial attributes of road types to assess the impact of these factors on CO 2 emissions using a real-world, second-by-second vehicle energy data set. A deep learning framework, long short-term memory model, is employed to predict the CO 2 emissions of HEVs. Permutation feature importance is used to quantify the relative importance of key predictors. Fuel consumption, mass air flow, trip lengths, and spatial attributes emerged as the key predictors of the model. To better explore the relationship between key predictors and CO 2 emissions, we also visualized their distributions. The results revealed that CO 2 emissions increase significantly with higher speed. Motorways generally involve longer trips, resulting in higher CO 2 emissions than short trips, despite a significant difference in the proportion of trips. Short trips tend to emit more CO 2 on local roads. The findings from this data-driven modeling study have practical implications in the transportation sector.