Fault classification and detection for photovoltaic plants using machine learning algorithms

Abstract

Using photovoltaic (PV) energy has increased in recently, due to new laws that aim to reduce the global use of fossil fuels. The efficiency of a PV system relies on many types of malfunctions which may cause significant energy loss during the system’s operation, besides the ecological factors. Consequently, a monitoring system (MS) capable of measuring both the environmental and electrical factors is described in order to gather real-time and historical data and estimate the plant efficiency metrics. Additionally, a recursive linear model for detecting problems in the system is presented, where the input is the irradiance and temperature of the PV module, whereas the output is the power, using the same MS. The achieved fault detection’s accuracy for the 5-kW power plant reached 93.09 percent, based on 16 days and 143 hours of failures under various situations. After detecting a defect, a machine-learning-based algorithm categorizes each defect problem as short circuit, partial shadowing, deterioration, or open-circuit. The performance of the four most prevalent supervised machine learning (ML) approaches for this assignment (Naïve Bias, decision tree, LDA, and KNN) was evaluated according to their results.