I–V characteristic and its fractal dimension for performance’s fault detection

Failures in photovoltaic systems are a major problem since they cause a decrease in the production of electrical energy. It is a challenge for the scientific community to obtain algorithms that adapt to existing systems, reducing the probability density of false positives. This paper solves this problem, presenting two contributions aimed at detecting faults in photovoltaic systems. The first contribution is aimed at a new algorithm based on non-coherent detection. Such algorithm is adaptable to any photovoltaic system and uses the box-counting procedure to estimate the fractal dimension of the normalized signal. The second contribution are to two equations that allow calculating the detection threshold under a failure prediction of suchalgorithm. The prediction of failures is based on a probability density of false positives set a priori. The algorithm was experimentally validated using 300 signals acquired from a photovoltaic system in series and parallel configurations. The results show that the algorithm had a behaviour, under a probability density of false positives of 2%, higher than those reported in the literature.