"TopoChange - Topology Change Synthetic Dataset"

"Topology changes of power networks degrade the performance of state estimators that assume a known topology. Physics-Informed Neural Networks (PINNs) and data augmentation techniques have the potential to handle topology changes. We introduce a dataset (TopoChange) that enables researchers to evaluate their methods under topology changes. This dataset contains processed real load profiles, synthetic load profiles, and power flows under different topologies. Original real load profiles are obtained from the Spanish smart meter dataset and preprocessed to ensure consistent load patterns and homogeneous recording intervals. Additional synthetic load profiles are subsequently generated from the processed dataset. Customers in both the real and synthetic datasets are assigned to the load buses in the CIGRE MV network to simulate power injections. The topology of the CIGRE MV network is alternated by changing the status of switches. For each load snapshot and topology, power flow is computed using the Power Grid Model to obtain voltages. A PINN was used to test how the synthetic data helps with topology changes. This dataset enables researchers to assess whether their model is robust to topology changes and whether it benefits from synthetic training data."