LOCA accidents are the types of accidents that need to be focused on in pressurized water reactor (PWR) nuclear power plants. Currently, the diagnosis of LOCA accidents in PWR nuclear power plants is mainly based on the determination of certain parameter thresholds such as the pressurizer water level and subcooling margin to judge whether there is an obvious break in the primary system. However, there is a relative lack of precise diagnostic means for the size of the break. With the development of artificial intelligence technology, the technical route of using deep learning and other means to further improve the diagnostic accuracy in accident diagnosis has been increasingly emphasized. Nevertheless, due to the requirements of deep learning and other artificial intelligence technologies for large amounts of training data and multiple feature quantities, there is a significant difference from the actual situation in the current design of PWR nuclear power plants where diagnosis is carried out with limited instrument parameters. Based on the actual design of a 1000 MW PWR, this paper develops a LOCA accident diagnosis model of the recurrent neural network (RNN) to evaluate the applicability and accuracy of the RNN diagnosis model for PWR nuclear power plants based on a small number of feature quantities, thus providing a reference for the development of deep learning LOCA diagnosis technology for maturely designed PWR nuclear power plants.