Model-Free Detection of Distributed Solar Generation in Distribution Grids Based on Minimal Exogenous Information

In recent years, the importance of PV generation data for distribution system operations has increased. However, some behind-the-meter solar installations are still not registered with the system operator and are not necessarily monitored at a centralized level. This “hidden” generation, therefore, increases the difficulty to operate securely and efficiently the distribution grid. This paper introduces a tool dedicated to the automatic detection of such a generation. It is designed to discriminate the nodes with and without local PV generation and is aimed at high accuracy, without local measurements, thus preserving privacy and increasing security. The tool consists of a neural network coupled with a rule-based classification algorithm, which considers only a very limited volume of data (i.e., node consumption and temperature data). Open-access consumption and solar radiation data are used to feed the simulation of a 14-nodes CIGRE distribution grid used to validate the proposed approach. The implemented solution is tested across all the nodes of the selected grid. The sensitivity of the results is analyzed by the level of PV penetration and the period of observation. The tool can recognize the nodes with a new PV installation with an accuracy of up to 100%, depending on exogenous conditions.