REVI-Twin: An Integrated AI-Driven Methodology for Creating Digital Twin of Residential Electric Vehicle Infrastructure

Integrating electric vehicles (EVs) into homes and the electrical grid introduces complex dynamics that can overwhelm planning tools. Accurately estimating hourly residential EV charging demand typically requires computationally intensive agent-based simulations that depend on large volumes of input data. Such requirements limit scalability of the digital twins and hinder timely decision-making, highlighting the need for more efficient modeling approaches. We introduce REVI-Twin, an AI-driven digital twin of residential EV infrastructure scaling without heavy agent-based simulation. The digital twin includes the following information: (i) ownership of EV at a household-level; (ii) user behavior and charging preferences; (iii) hourly power consumption by the EVs; and (iv) planned trips taken using EVs. Our framework tackles two tasks: (i) predicting EV adoption from scarce labels via transfer learning, semi-supervised learning, and Bayesian optimization; (ii) synthesizing hourly consumption with active-learning multi-output Gaussian processes learning robust patterns from <1% of data. Furthermore, we develop and publicly release an integrated, hourly residential energy dataset comprising profiles of synthetic residential loads, EV charging, rooftop photovoltaics, and battery energy storage systems. Our case study with the integrated residential energy framework shows that each 1% of battery adoption reduces Virginia's net imports by ~0.06% daily and ~0.085% during peak hours, with the expected battery capacity ranging between 23-28 kWh. REVI-Twin supports policymakers, utilities, and planners with scenario analysis of adoption, charging demand, and infrastructure needs for resilient electrification.