Preventive automated network control: enhancing low voltage grid management with recurrent neural networks

The integration of volatile, emission-free energy generation and new power-intensive consumers into low-voltage distribution grids presents significant challenges for grid stability, operational safety, and transformer reliability. To address these challenges, this paper introduces Preventive Automated Network Control (PANC), a novel approach designed to optimize grid management through secure, runtime-efficient, and automated solutions. PANC builds on the regulatory framework in Germany, advancing beyond autonomous curative load congestion management to implement preventive strategies. Using recurrent neural networks, PANC provides short-term load forecasting for grid state estimation, enabling early detection of critical states and reducing reliance on curative measures. The methodology integrates real-time electrical measurements, grid topology data, and weather information to train models capable of ex-ante grid state prediction, even with incomplete datasets. A strong focus is placed on capturing gradients and seasonal trends influencing load and generation dynamics. The system leverages secure MQTT-based communication interfaces for scalable integration into operational environments. The communication interfaces were tested in cooperation with “Stadtwerke München” (SWM). First validation results, derived from transformer measurements in real low-voltage grids, demonstrate the practical feasibility of PANC under real-world conditions. This paper highlights the methodologies, interfaces, and technologies enabling its deployment, offering a foundation for future grid management innovations.