Transformer-Based Inference of Residential EV Charging Patterns from Smart Meter Time Series

Accurate prediction of electric vehicle (EV) charging behavior in residential settings is essential for optimizing grid load management and supporting the transition to sustainable transportation. This paper introduces a novel transformer-based methodology for inferring EV charging events directly from smart meter time series data. Our approach, which employs a divide-and-conquer strategy, effectively captures intricate charging patterns by processing segmented load data in parallel. This enables high-resolution, short-term forecasting of EV charging occurrences without relying on explicit charging records. The efficacy of our method is validated using real-world smart meter data, demonstrating its potential to provide valuable predictive insights for grid operators and facilitate more efficient energy distribution and management strategies.