Abstract: Most photovoltaic power stations are located in harsh environment,suffer from wind,sand,rain and snow corrosion,panels are prone to multiple types of failure.How to effectively identify and locate the fault is particularly important. Therefore,a strategy for photovoltaic fault diagnosis based on the clockwork-recurrent neural network（CW-RNN）was proposed. Firstly,a simulation model of photovoltaic array system was established,the causes of photovoltaic power generation faults were analyzed,and the output characteristics of photovoltaic array under different faults were simulated. Then,the CW-RNN method was used to establish a fault diagnosis model to identify and locate photovoltaic array faults. Finally,a photovoltaic power generation fault analysis platform was built based on the real-time database system,and the performance of the proposed fault diagnosis model was verified.The effectiveness and accuracy were verified,which has certain reference significance for the efficient and accurate fault identification and location of photovoltaic power stations.