The rapid growth of photovoltaic (PV) installations in low-voltage distribution grids presents significant challenges for distribution system operators (DSOs). By the end of 2024, an additional 16.2 GW of PV capacity was installed in Germany, predominantly on residential rooftops and connected to the low-voltage network. This development marks a structural shift from passive, radial grids to networks characterized by volatile renewable generation and distributed energy resources (DERs) with demand-driven behavior. As a result, power flows in these grids are becoming increasingly unpredictable. Since PV is expected to remain a central pillar of the transition toward a carbon-free energy system and in meeting the 1.5°C climate target, associated issues—such as grid congestion and voltage band violations—are expected to intensify. Consequently, accurate forecasting of the grid state is essential for DSOs to enable proactive and preventive grid management.

This work presents a real-time system for nowcasting the state of low-voltage distribution grids using open-source approaches, developed and validated with reference to a typical German distribution grid in Hittistetten. The grid model is kindly provided by the local DSO, Stadtwerke Ulm/Neu-Ulm, to the Smart Grids Research Group as a representative test area, with measurement data available from various grid components. The objective is to demonstrate a forward-looking methodology that enables DSOs to maintain reliable and stable grid operation under increasing PV penetration. The core of the system is a solar radiation forecast based on data from the Meteosat Second Generation (MSG) satellite. This feeds into a physical PV power model to generate feed-in forecasts for the grid area. A complementary load forecast is generated using modified standard load profiles. These feed-in and load forecasts are combined and processed through a load flow algorithm to determine the future grid state. Because the load profiles are based on annual aggregates, the load forecast is further refined using real-time measurements of transformer utilization. The complete model chain is comprehensively validated using measurement data from the operational grid.

All components of the model show strong correlation with observed data and low prediction errors. The system provides accurate PV generation forecasts and, at an aggregated level, reliable grid status predictions for two out of three transformers in the test network. The third transformer could not be effectively validated due to inconsistent measurement data.

This study demonstrates that data-driven nowcasting of the low-voltage grid state is not only technically feasible but also represents a key contribution to ensuring grid stability amid the evolving dynamics of the energy system.