A synergistic estimation method of electric vehicle (EV) energy consumption-emission by integrating micro-behavioral modeling and spatial gradient mapping partitioning

Driven by the goal of "dual-carbon", the fine-grained evaluation of the energy consumption and emission characteristics of electric vehicles (EVs) in complex road environments has become a key foundation for the design of intelligent transportation systems (ITSs) and the formulation of policies. In this paper, we propose a synergistic EV energy consumption-emission estimation method that integrates microscopic traffic behavior modeling, road gradient mapping partitioning, and structure-aware map convolutional network. In the behavioral modeling layer, driver reaction delay and acceleration/deceleration strategy features are introduced to construct a time-dependent driving pattern dictionary; in the spatial modeling layer, the terrain influence mapping is generated by coupling the rasterized road slope data with the road network density, and the structure of the estimation network is introduced in the form of a graph neural network (GNN). The joint loss optimization function is further employed to achieve the bi-objective regression fitting of energy consumption and equivalent CO₂ emission metrics. The method is tested on several city-level datasets (Chongqing mountain city and Suzhou plain city), and the results show that the model can control the combined estimation error within ±6.2%, which is a 14.8% improvement in the average accuracy compared with the typical benchmark model (VT-Micro, CMEM). The research results can serve the needs of multi-scale path energy consumption assessment, low-carbon transportation organization and regional emission refinement modeling. Keywords: electric vehicle energy modeling; equivalent carbon emission estimation; microscopic traffic behavior analysis; road gradient mapping partitioning; graph convolutional neural network (GNN); multi-objective regression; urban traffic simulation; low-carbon travel path assessment 1 Introduction Under the background of "dual-carbon" strategy and accelerated transportation electrification, accurately evaluating the energy consumption and equivalent carbon emission characteristics of electric vehicles (EVs) in diverse road environments has become the core foundation of low-carbon transportation organization and path optimization. Most of the existing studies focus on macro-statistical or single dynamics modeling, which is difficult to take into account the coupled effects of driver micro-behavioral differences and spatial topography of roads on energy consumption-emission. In order to break through this limitation, a collaborative estimation method integrating microscopic traffic behavior modeling, road gradient mapping partitioning and structure-aware map convolutional network is proposed to achieve unified spatial and temporal feature modeling and dual-objective regression optimization. Taking mountainous and plain cities as typical samples, it is expected to form a high-precision and interpretable energy consumption-emission estimation framework to provide data and methodological support for refined low-carbon traffic management. 2 Basic Theory of Electric Vehicle Energy Consumption and Emission The energy consumption calculation part introduces vehicle dynamics equations, coupling the instantaneous traction demand with the slope angle and mass distribution to form a power consumption model. The emission estimation part adopts the grid carbon intensity factor to map the time-shared power consumption to equivalent CO₂ emissions. The slope partitioning module utilizes terrain gradient thresholds to segment road segments and matches them with microscopic driving pattern indexes to achieve two-way terrain-behavior mapping. The model training phase captures the spatial dependency of the road network with GNN node feature embedding and introduces a weight adaptive mechanism in the joint loss to balance the energy consumption and emission fitting accuracy. 3 Collaborative estimation method integrating micro-behavioral modeling and spatial slope mapping partitioning 3.1 Microbehavioral modeling method The micro-behavioral modeling first fixes the sampling frequency at 1 Hz, and among the total 22835 s of trajectories, the ICV samples cover 11939 s and the EV samples cover 10896 s, which provides sufficient time-domain density for decomposing the original driving trajectories. The model slices the atomic behaviors of steady state, acceleration and deceleration according to the acceleration