With the advancement of artificial intelligence technology, intelligent diagnostic technology has been gradually implemented across various industries. This study proposes the use of convolutional neural networks–long short-term memory (CNNs-LSTM) for diagnosing faults in CPR1000 nuclear power plants (NPPs). To automatically extract data related to different types and levels of faults in the PCTRAN program, the study utilizes a self-developed AutoPCTRAN software and selects several key nuclear parameters as feature quantities. The study uses random sampling to create the training, validation, and test sets in an 8:1:1 ratio and identifies acceptable parameters to build the CNN-LSTM model. Test results show that the CNN-LSTM–based model for diagnosing CPR1000 NPP faults achieves a problem recognition rate of 99.6%, which validates the efficacy of the CNN-LSTM–based nuclear power fault diagnosis model.