The output characteristics of photovoltaic modules are affected by many factors, so the fault detection of photovoltaic modules is a severe test. In order to ensure the real-time and accuracy of fault diagnosis, multi-sensor method is used to extract short-circuit and open-circuit fault features, and voltage scanning method is used to obtain the judgment basis of thermal breakdown and electrical breakdown caused by uneven illumination. Based on fault characteristics, an improved RBF neural network based on K-means clustering is proposed for photovoltaic module fault diagnosis. The types of faults occurring in components and fault location are carried out, and the results are compared with those of BP neural network, which verifies the accuracy and effectiveness of the improved RBF neural network fault diagnosis method.