Anomaly detection by comparing photovoltaic systems with machine learning methods

The digital interconnection of several photovoltaic systems enables a comprehensive monitoring and detection of failures. The concept of this work is based on the similar behaviour of individual photovoltaic systems under same conditions. By recording electrical data, individual strings can be compared with each other by machine learning methods. For this purpose, a k-Nearest-Neighbours algorithmn and a Multi-layer Perceptron are used for processing the data and detecting divergent characteristics of the electrical current. With this approach, a reliable and reproducible fault monitoring can be guaranteed without large measuring effort.