Data-driven photovoltaic installed capacity detection at low voltage substation level

With the increasing penetration of photovoltaic (PV) systems in distribution networks, accurate estimation of PV installed capacity becomes crucial for grid management and planning. This paper proposes a data-driven approach, based on super- vised Machine Learning for estimating total PV installed capacity connected to a Low Voltage (LV) distribution network, using substation level net load measurement data. An XGBoost model is trained to estimate installed PV capacity at the LV side using statistical features from substation net load profiles. Leveraging the availability of residential smart meters, synthetic Medium to Low Voltage substation load profiles are created by sampling and aggregating real consumers’ load profiles, creating a diverse database of synthetic substation load profiles, which is then used to train the model. The proposed methodology is then validated on a synthetic substation dataset generated in a similar manner, demonstrating its effective- ness in accurately estimating PV capacity without the need for expensive and intrusive monitoring equipment. The model performance is assessed in varying degrees of PV penetration and network size, providing a more nuanced evaluation of the proposed model. This data-driven approach provides estimations on unseen synthetic substations with errors of 2.93% and 6.33% for substations with 30 and 150 consumers, respectively, for an actual installed capacity of 150 kWp, which are realistic conditions for European LV networks.