Graph Mining for Classifying and Localizing Solar Panels in Distribution Grids

Abstract

The growing photovoltaic (PV) penetration level in distribution networks has presented significant challenges to topology recovery of the distribution systems. Unfortunately, many utilities not only lack accurate topology information on the distribution grids but also have no record of photovoltaic panel locations. To acquire approximate locations of solar users, we propose a Graph Mining (GM) approach for solar panel localization. Due to the graphical structure of the grid and the temporal features of the power demand (Pd), we employ a solar panel classification algorithm that identifies graphical topology with time series data. Based on this time-series information, we design a graph construction algorithm and convert the time-series data to graph-type data. In the end, the graph-type data are fed into a graph neural network. By doing so, we transfer this problem into a graph classification problem and recognize the buses that are connected with solar panels. We validate the proposed method on several benchmark distribution grids and evaluate the model's capability under different system scenarios. The numerical results show that our algorithm can accurately detect solar panel locations in distribution feeders, thus improving the situational awareness of the secondary distribution grid.