Umsetzung von KI-Algorithmen zur Anomaliedetektion in PV-Anlagen und Entwicklung einer Data Handling Plattform

With the increasing expansion of renewable energy sources, the importance of efficient monitoring systems for photovoltaic installations is also growing. In particular, the early detection of anomalies such as leaf shading plays a crucial role in optimizing system performance. The aim of this thesis is the development, implementation, and evaluation of two AI-based models - Random Forest and Long Short-Term Memory (LSTM) - for the detection of such shading effects.
To this end, an appropriate communication structure was first developed and then connected to a data handling platform based on a Timescale database, Node-RED, and Grafana. Data collection was carried out at the string level and included both electrical and meteorological measurements. Using simulated leaf shading scenarios on a real system, the two AI models were trained, validated, and subsequently integrated into the overall system. In a practical test, both models demonstrated strong performance in detecting significant leaf shading on a module, achieving F1 scores of 0.87 (Random Forest) and 0.94 (LSTM), respectively. Weaker shading conditions, however, were only reliably identified by the Random Forest model. The results show that artificial intelligence models can reliably detect pronounced characteristic patterns and be effectively integrated into monitoring systems. The approach presented here thus offers potential to support anomaly detection in large-scale PV systems and could, in principle, be adapted to other types of faults.