Electric vehicles (EVs) will play a central role in future energy-efficient and sustainable transportation systems. Predicting the energy use for EVs is a complex issue because the onboard vehicle systems are trying to balance the provision of power to the wheels as well as manage the state of charge (SOC) of the battery pack. Traditional modeling methodologies for estimating real-world EV energy consumption either depend on numerical analysis of laboratory or on-road vehicle test data or the use of full-system EV simulation tools. Unfortunately, full-system simulation tools suffer from scaling problems in the context of large transportation network, necessitating the development of approaches that supports large transportation network projections of modal EV operations and applicable energy use rates.
This study introduces an activity-based, bottom-up modeling approach to estimate EV energy consumption under the expected range of on-road operating conditions. The proposed system integrates outputs from a full EV simulation model called Autonomie. Three analytical efforts were undertaken to develop the activity-based approach for EV fleets using Autonomie simulation outputs. First, a sample of EV technologies was configured in Autonomie, and various operating conditions were simulated in Autonomie to generate a library for on-road operations by technology. Second, a grey box model design, referred to a Bayesian Network method in this study, was used to develop energy consumption models for the variety of EV technologies. This approach combines vehicle performance knowledge and data-driven energy inferences with on-road vehicle operation as inputs. Finally, the proposed EV energy models were verified using a separate testing dataset developed from Autonomie simulation results of another set of driving profiles. In addition, the real-world observed operation and energy use data were collected from select EV models using on-board diagnostic (OBD) devices to verify the energy prediction from the proposed model. The verification results suggested that the proposed model can predict energy use patterns under most driving conditions. The proposed approach was applied at aggregated-level, to a regional-level network, and at individual-level, to households and persons traveling within a region. The scalability of the proposed energy model framework was demonstrated in an Atlanta, GA case study. The results demonstrated that if 6.2% of urban VMT and 4.9% of rural VMT are driven by EVs, the network-level fuel savings are around 4.0% for a normal travel day in 2024. The energy model was also applied to daily trips predicted by the regional travel demand model. The results suggest the actual benefits of EV adoption depends on household travel patterns across deployment scenarios, as well as charger availability, electricity and fuel cost, and ambient environment conditions.