Abstract: Aiming at the problem that the fault diagnosis methods for photovoltaic（PV）arrays based on traditional machine learning algorithms need a large number of training sets, a CatBoost algorithm based fault diagnosis method is proposed to achieve the accurate diagnosis of different degrees of faults in small-scale training sets. The equivalent circuit model of PV modules is presented. Different degrees of PV array faults including short circuit, open circuit, aging and partial shading are considered.Changing characteristics of the I-V characteristic curves of a PV array including bypass diodes and blocking diodes are analyzed.Characteristics are built to reflect different fault characteristics and selected as the input vector of the fault diagnosis method for PV arrays. The CatBoost algorithm is used to train the small-scale training set, and the CatBoost algorithm based fault diagnosis model is established. In order to verify the effectiveness of the proposed method, simulation and experimental analysis are carried out,respectively. The proposed method is compared with traditional neural network algorithms and other decision tree algorithms to verify the accuracy and stability of the proposed method in the small-scale training sets.