State-of-charge-aware charging opportunity detection for electric vehicles using data-driven learning and digital twin simulation

• SoC-COD detects charging opportunities via data-driven digital twin logic. • XGBoost classifier achieves ROC-AUC = 0.9915 and F1-score = 0.9123. • Grid energy and cost reduced by 52.5% with 85% fewer charging events. • Minimum SoC improved by 24%; battery cycle accumulation cut by 85% • Edge-deployable in 3.25 ms; no V2G, grid data, or driver input needed. As the number of electric vehicles (EVs) rises, ineffective and needless charging during parking events is reducing battery life, user comfort, grid stability, and charging infrastructure usage. Current charging schedules do not take into account whether a charging activity is required in the first place; instead, they optimize for power or scheduling after the charging choice. Additionally, many methods scale poorly and cannot be implemented in practice because they rely on grid-side data, battery physics models, or static heuristics. Using historical operational data, a completely data-driven, vehicle-centric State-of-Charge–Aware Charging Opportunity Detection (SoC-COD) framework is provided to determine if charging during a parking event can successfully enable SoC recovery without utilizing any real-time information. SoC-COD, which combines optimization-based labeling and probabilistic machine learning to infer the charging opportunities (i.e., whether a vehicle would charge) and counterfactual SoC trajectories if they could, is a parking-event level approach that combines churn prediction and charging opportunity identification. Validation using digital twin simulations and the California Audi EV Dataset is shown; SoC-COD achieves near-perfect opportunity discrimination (ROC–AUC = 0.9915) with a strong SoC prediction accuracy (R 2 = 0.8297), while also reducing charging events by 85.00% and lowering energy and cost by 52.50%. It also improves the minimum SoC by 24%.