The nuclear reactor coolant system (RCS) serves as a critical barrier for the safe operation of nuclear power plants, and the accuracy of its fault diagnosis is of paramount importance for ensuring nuclear safety. In response to the shortcomings of traditional fault diagnosis methods in feature extraction and temporal modeling, as well as the numerous limitations of conventional diagnostic systems in installation, maintenance, and cross-platform data interaction, a convolutional neural network-gated recurrent unit (CNN-GRU)-based RCS fault diagnosis model was developed, and an RCS fault diagnosis system based on a browser/server (B/S) architecture was designed following this approach. First, a CNN model was employed to extract local spatiotemporal features from RCS operational data. Then, these extracted features were fed into a GRU for temporal modeling, constructing the CNN-GRU diagnostic model. Finally, an RCS fault diagnosis system was designed based on this model. The system achieved fault diagnosis through front-end data upload and back-end model inference. Experiments were conducted using simulated data generated by the personal computer transient analyzer (PCTRAN) for nuclear power plant simulation and severe accident analysis. The results demonstrate that the system can accurately identify 8 types of RCS fault conditions, achieving an average accuracy of 99.85%, thereby providing reliable fault diagnosis support for the safe operation of nuclear power plants.