"A Hybrid Multi-Domain Electric Vehicle Dataset for AI Model Training "

"I present EV-HybridDataset-2026, a comprehensive hybrid dataset for AI model training in the electric vehicle (EV) domain of power and energy systems. The dataset integrates three interconnected sub-datasets:1. EV Charging Sessions: 50,000 records covering AC Level 1\/2 and DC Fast charging (up to 350\u202fkW), five connector standards, four location types, over the period 2022\u20132024.2. Battery Degradation Checkpoints: 37,506 records across 300 vehicles with three Li-ion chemistries (NMC, LFP, NCA), tracking state-of-health (SOH), internal resistance, open-circuit voltage, and EIS impedance over up to 1,200 equivalent full cycles.3. Trip Energy Consumption: 80,000 records for six commercial EV platforms across urban, highway, mixed, and rural scenarios, including full physics-correlated contextual metadata.Parameter distributions are calibrated against eight peer-reviewed real-world sources. Identified data gaps are filled through physics-informed stochastic synthesis, clearly flagged with a `data_source` column.All 167,506 records across 51 variables are validated with zero missing values, zero range violations, and zero physical logic violations. The dataset supports AI tasks including:- Charging Load Forecasting: LSTM (R\u00b2 = 0.883)- SOH Estimation: XGBoost (R\u00b2 = 0.971, MAE = 0.42%)- Energy Consumption Regression: Random Forest (R\u00b2 = 0.921)- User Behavior Clustering: K-Means (silhouette = 0.67)The full data generation and validation pipeline is provided as an open-source Python script."