Anomaly Detection in Photovoltaic Production Factories via Monte Carlo Pre-Processed Principal Component Analysis

This paper investigates a use case of robust anomaly detection applied to the scenario photovoltaic production factory—namely, Enel Green Power’s 3SUN solar cell plant in Catania, Italy—by considering Monte Carlo based pre-processing technique as valid alternative other typically used methods. In particular, proposed method exhibits following advantages: (i) Outlier replacement, by contrast with traditional methods which are limited outlier only, and (ii) preservation temporal locality respect training dataset. After pre-processing, authors trained an model on principal component analysis defined suitable key performance indicator for each sensor line errors. this way, running algorithm unseen data streams, it is possible isolate anomalous conditions monitoring above-mentioned indicators virtually trigger alarm when exceeding reference threshold. The approach was tested both standard operating scenario. With considered case, successfully anticipated fault equipment advance almost two weeks, but also demonstrated its robustness false alarms during normal conditions.