Advanced Spatiotemporal Anomaly Detection in Photovoltaic Power Plants Using Temporal Graph Networks Integrated with Publicly Available Weather Data

For purposes of performance analysis and minimization of maintenance expenses, it is vital to identify an effective way in which to monitor anomalies in photovoltaic (PV) power plants. This paper proposes a novel framework to address the problem of anomaly detection based on the inherent graph structure of PV plant configurations using Temporal Graph Networks (TGNs). TGNs are built to capture spatial and temporal dependencies of the PV systems which allows detecting discrepancies affecting the power production, which traditional NN might fail not to notice. The model uses external weather data collected from NOAA and tracks how external parameters influence the performance of photovoltaic panels and therefore electricity generation. Our methodology solely employs TGNs to analyse the intricate relationship in the PV plant data to improve the recognition of abnormality associated with the equipment and weather conditions. The integration of NOAA's open source weather data into the extent of analysis further raises sophistication by offering a variety of features and conditions affecting PV results. Our approach achieves high accuracy of anomaly detection and can be considered to be better compared with the baselines that do not take graph dependencies or time aspect into account. This advancement has important implications for enhancing the operation and maintenance recommendations in the PV power plants and hence the reliability of the satellite power systems. Allying our efforts to one highly specialized method, this research presents a new state of the art in leveraging centralized data for anomaly detection in renewable energy.