Event Actions
Electric Vehicle Adoption and Residential Charging Modelling
Abstract:
As part of this examination, I explore three research questions: (i) Leverage rooftop solar adoption model for Electric vehicle (EV) assignment, (ii) Predict residential EV charging hourly energy consumption, and (iii) Interpretation of the feature contributions for the trained EV adoption model. In the U.S., the transportation sector is considered as one of the highest contributors of greenhouse gas (GHG) emissions. Adopting EVs is one way to address GHG emissions in transportation. However, several challenges arise due to the EV adoption. In this research, we develop two interconnected models: (i) to predict residential EV adoption, including type of home charger and charging location preferences, and (ii) to generate the hourly energy consumption patterns. We leverage transfer learning and semi-supervised learning for the former to extrapolate insights from solar adoption models to EV models. We augment this model with Bayesian optimization to calibrate the model's predicted EV adopters with real-world statistics on EV adoption. Next, we utilize Explainable AI (XAI) techniques to uncover insights about household and demographic features on EV adoption based on the model's prediction. For the latter, departing from traditional simulation-based methodologies, we employ active learning with a multi-output Gaussian process model to capture the hourly correlation and to develop our EV energy consumption models, prioritizing the most informative unlabeled data for querying. This approach efficiently learned the hourly EV residential charging prediction using only < 1% of the data. Finally, we validate our outcomes against real-world data and survey reports for both EV adoption and residential EV energy consumption.
Committee:
- Anil Vullikanti, Committee Chair, (CS, Biocomplexity/SEAS/UVA)
- Marathe Madhav, Advisor (CS, Biocomplexity/SEAS/UVA)
- Yangfeng Ji (CS/SEAS/UVA )
- Upsorn Praphamontripong (CS/SEAS/UVA)