Maximum power point analysis for partial shading detection and identification in photovoltaic systems

Fault diagnosis of photovoltaic (PV) systems is a crucial task to guarantee security, increase productivity, efficiency, and availability. In this regard, numerous diagnosis methods have been developed. Methods requiring the interruption of power production are not adequate for economic reasons. The development of large-scale PV plants and the objective of maintenance cost reduction push toward the emergence of automatic on-line diagnosis methods that use available information. In this study, we propose two data-driven methods for partial shading diagnosis using only the maximum power point's information. It does not require the interruption of production, nor does it require any additional equipment to obtain the I(V) curve. The analyses are conducted with principal component analysis (PCA) and linear discriminant analysis (LDA) to detect and classify the faults. The experimental dataset is collected from a 250 Wp PV module under four states of health (healthy, and three severities of partial shading) for several meteorological conditions. The classification results have a 100% success rate, and are robust to the variations of temperature and irradiance.