Nuclear power plants have a multitude of online real-time monitoring parameters that are easily affected by internal and external disturbances, changes in operating conditions, equipment start-ups and shutdowns, etc. Additionally, there is a lack of autonomous analysis and diagnostic capabilities. To detect potential anomalies in the unit early and ensure its safe and stable operation, research on artificial intelligence autonomous monitoring technology is particularly necessary. This paper introduces deep learning methods to analyze the process flow and fault characteristics of nuclear reactors, constructing a fault early warning model to predict potential slow-changing faults or accidents in nuclear power units. Considering the complex operational conditions and scarcity of fault samples in nuclear power plants, this paper proposes a fault early warning method based on an auto-encoding neural network (AENN). By analyzing typical accidents in nuclear power plants such as steam generator tube rupture (SGTR), modeling parameters are determined. Combined with typical data samples generated during the operation of the simulation verification platform, an autonomous fault monitoring model is established. Leveraging the powerful self-learning capabilities of deep neural networks, the non-linear characteristics of unit operation are simulated, and the predicted values of the current state are outputted for similarity comparison with actual values. This allows for the identification of faults at their initial stages, achieving parameter early warning functions based on SAE. The performance of the proposed autonomous fault monitoring method and model is tested using a test set, and the results indicate that it can effectively perform accident early warning functions, significantly enhancing the level of intelligence in nuclear power plants.