Supervision and Fault Detection System for Photovoltaic Installations Based on Classification Algorithms

This article presents a methodology for the supervision and fault detection on photovoltaic installation, through the information gathered by their SCADA system. The proposed methodology consists of the use of a multi-clustering approach to analyse and classify the operating behaviour of the photovoltaic installations, using information of their DC voltage, generated current (per string), as well as information related to the climatic conditions of the park (i.e. solar irradiance, temperature). The proposed methodology uses a supervised training algorithm, based on a decision tree learning algorithm, allowing to determine the appearance of anomaly behaviours in the installations of photovoltaic plants, including soiling detection, hot-spots, tracker deviations, electric connections and faults in sensors. The presented methodology has been developed in the framework of a CORFO R&D project and validated under real operating conditions in a utility-scale photovoltaic power plant of one axis mount, located in Chile, with a total power of 40 MWp.