Anomaly Detection in Photovoltaic Systems via Deep Learning Autoencoder

The usage of photovoltaic (PV) systems has been rapidly increasing in recent years to generate green electricity. Anomaly detection in PV systems is essential to improve reliability, ensure electricity production and equipment safety, and decrease their negative impact on the economy of the operation system. This paper proposes an effective anomaly detection model based on a deep neural network autoencoder, which accurately identifies PV system anomalies. The presented model only uses measured PV power production as input and does not need additional equipment data collection. The results demonstrate that the proposed model is superior to existing data-driven solutions regarding different evaluation accuracy metrics to detect anomalies in PV systems. The proposed model is validated and assessed on a PV power plant in Genoa, Italy. Obtained numerical results of the case study show the superiority of the developed model compared to principal component analysis and other machine learning classification models. In comparison to other machine learning models, the proposed model accurately detected all faults in the test dataset, achieving an F1-score accuracy of 82%.