Complex network approach to detect faults in photovoltaic plants: Albanian case study

Abstract

With the trend towards photovoltaic plants in energy production, it is often difficult to describe the system status and fault conditions using traditional methods. We used machine learning and deep learning to predict PV production, but here we are focused on detecting faults in PV plants. Complex networks analysis is presented as an approach which succeeds in detecting faults in PV plants.  Our case study is a photovoltaic plant installed in a factory in Albania. Firstly, we use sliding windows for different periods of time and construct a weighted directed network (functional graph) when each node represents a sensor and each edge represents the strength between the two signals which is determined by the mutual information measure. We have used in total 29 different signals from 4 different inverters for a period of six months (June 2022- December 2022). Secondly, we compute the node’s strength for the signals corresponding to voltage, current and irradiance and show that when a fault occurs the weighted degree centrality of the irradiance node decreases while the voltage and current weighted degree centrality increases.