The impact of the ANN’s choice on PV systems diagnosis quality

•The impact of the ANNs’ choice on PV systems diagnosis quality is addressed.•The efficiency of an ANNs based intelligent diagnosis algorithm is analysed.•Five ANNs are considered: BPNN, Two RBF, PNN and GRNN.•The comparison takes into account the accuracy, the specificity, the sensitivity and the rapidity.•The results identify the PNN as the best candidate for the studied diagnosis task. Fault diagnosis has become an indispensable part of PV installations to ensure their safety and reliability. The accuracy, rapidity, specificity, sensitivity and the precision of faults detection and isolation are the most pertinent criterions of the diagnosis quality. The present work examines the impact of the Artificial Neural Networks choice on these criterions. For this, five ANNs are studied: back-propagation ANNs (BPNN), generalized regression ANNs (GRNN), probabilistic ANNs (PNN) and two radial basis function ANNs (RBF). These ANNs are used to identify and locate the most frequently faults encountered in PV installations: short-circuit cases and open-circuit string cases in PV generator. Comparison study using the same PV installation, working conditions, data and the same diagnosis algorithm have been carried to confront the five ANNs to the same faults. Based on experimental data, the study shows that RBF ANNs affect the rate of reaction of the algorithm in presence of faults while BPNNs and GRNN present the best results from the point of view of its speed and its important high precision with good classification efficiency. In the other hand, the PNN marks its importance by its best results displaying 100% for all key statistical concepts comparing to the other algorithms.