Generating Synthetic Data for Electric Vehicle Charging Sessions: Modeling and Validation with Real-World Data

Urban environments are being increasingly equipped with charging stations due 1 to the rapid proliferation of electric vehicles (EVs). The significant changes in the dynam- 2 ics of city-wide power grids resulting from the surge in EV adoption present challenges, 3 such as increased peak demand, while also providing opportunities for intelligent load 4 management and grid flexibility. A major obstacle to data-driven research and develop- 5 ment is the lack of realistic EV charging datasets, which is still restricted by data sparsity, 6 proprietary limitations, and privacy issues. This charging infrastructure is transforming 7 power grid dynamics, introducing new demands for high-resolution synthetic data to 8 support modelling, simulation, and decision-making in smart grid environments. This 9 study addresses the critical gap by developing synthetic data and evaluating multiple 10 synthetic data generation (SDG) techniques tailored explicitly for EV charging datasets. 11 We explore a range of approaches, including rule-based generation strategies informed by 12 domain expertise, as well as several machine learning–based models, such as Conditional 13 Tabular GAN (CTGAN), Conditional GAN (CGAN), Kernel Density Estimation (KDE), 14 and Variational Autoencoder (VAE). Using a real-world EV charging dataset, we evaluate 15 each method’s effectiveness using measures including KL divergence, Maximum Mean 16 Discrepancy (MMD), and Mean Absolute Percentage Error (MAPE) in order to determine 17 statistical resemblance and distributional faithfulness. The findings demonstrate that both 18 heuristic and ML-based generative models can produce realistic synthetic data with varying 19 degrees of fidelity and interoperability. In order to overcome data constraints and facili- 20 tate the creation of robust, data-driven smart grid systems, this work offers a methodical 21 methodology for creating synthetic EV charging data.