Power-reduction detection and classification in photovoltaic modules using clustering of parametric deviations

Power reduction in photovoltaic modules may be caused by environmental conditions, faults, and degradation mechanisms, whose correct identification is essential to improve the reliability, safety, and energy yield of photovoltaic systems. This article proposes a classification approach for power-reduction scenarios based on parametric deviations extracted from a Global Nonlinear Model (GNLM) derived from the Single-Diode Model (SDM). The methodology starts with the construction of a GNLM representative of a healthy reference photovoltaic module, capable of describing its electrical behavior under different irradiance and temperature conditions. For each evaluated experimental current–voltage ( I–V ) curve, selected SDM parameters associated with this healthy reference model are adjusted to approximate the electrical behavior of the analyzed module. The deviations required for this adjustment, calculated with respect to the healthy reference condition, are used as diagnostic markers for clustering-based classification. Different power-reduction conditions were evaluated, including soiling, short-circuited bypass diodes, open-diode conditions under different shading levels, front-glass breakage, and partial shading (PS). To reduce the overlap between scenarios with distinct electrical behaviors, the samples were initially separated according to the number of maxima in the power–voltage ( P–V ) curve, denoted by N peak , distinguishing cases with N peak = 1 from those with N peak > 1 . Based on these diagnostic markers, two strategies are investigated according to the set of SDM parameters used in the classification process. Strategy 1 uses deviations in the diode ideality factor ( n ) and photogenerated current ( I p h ), achieving the best performance, with an overall accuracy of 99.6%. Strategy 2 uses deviations in the n , series resistance ( R s ), and shunt resistance ( R s h ), achieving an overall accuracy of 98.3%.