A CTGAN-based approach for synthetic data generation of low-voltage distribution networks

The development of Low-Voltage (LV) distribution networks to support low-carbon heating and transport technologies requires substantial load and network characteristic data. However, privacy concerns and the costs of data monitoring equipment limit observations at this network extremity. Generating representative data is challenging due to system heterogeneity, especially in the last mile of LV networks. To address this, we developed a method that captures network topology to accurately generate synthetic node locations, feeder positions, and cable types. Using Conditional Tabular Generative Adversarial Networks (CTGAN), our approach produces high-fidelity synthetic data closely resembling real LV distribution networks. Quality metrics including Jensen-Shannon Divergence (JSD) and Maximum Mean Discrepancy (MMD) yield results of 3% and 1% respectively, validating data fidelity. Synthetic LV networks topology graphs confirm our method’s accuracy, while additional robustness verification using Coverage and Precision metrics on summer and winter load profiles further strengthens model validation. This synthetic data enables advanced power system analyses while safeguarding data privacy.