Abstract
Electric vehicle (EV) charging at home is transforming the electric load profiles of residential buildings, creating both challenges and opportunities for the grid. Using
data-driven methods to extract EV charging information from smart meters can provide valuable insights for grid management. However, prior research has mainly
focused on developing methods tailored to residential buildings without electric heating, where EV charging dominates peak electricity use, and seasonal variation in
the load is limited. This article presents a case study that addresses this gap by combining smart meter data from 296 single-family houses in Norway with EV
charging data from 82 residential chargers, creating semi-synthetic load profiles for households with and without EV charging and electric heating. A three-step
supervised method with domain-informed features is proposed for: (1) estimating EV charger capacity, (2) classifying charging events (on/off), and (3) estimating
electricity use for EV charging from hourly household smart meter data. Results show that Step 1 estimates charger capacity with 93% accuracy, Step 2
detects charging events with 81% recall/84% F1-score and Step 3 estimates charging electricity use with an average R2 of 0.80–0.86/NMAE of 0.005–0.006. Applying
a physics-informed filter to Step 3 can further improve estimates of electricity use for charging by eliminating physically impossible solutions, improve real-world
applicability and significantly reduce MAE. Each step can be used independently for applications such as demand-side flexibility estimations and user nudging,
making the framework flexible and useful for a range of stakeholders in the energy sector.