Deep learning based key parameters prediction in nuclear power plants

Accurately obtaining key parameters under accident conditions in nuclear power plants (NPPs) is essential for safety analysis and is critically important for operational decision making. To enable rapid and precise forecasting of parameter trends in NPPs, we introduce an enhanced Non-stationary Transformer (NS Transformer) model. This model enhances the attention mechanism by incorporating a sparse attention mechanism. The sparse attention mechanism, which selectively focuses on key locations in the sequence, thereby significantly reducing computational complexity and memory use. In this paper, we detail the model structure and validate its effectiveness through experiments predicting key parameters in nuclear power plants. The results demonstrate that the improved prediction model can accurately forecast long-term parameter trends. This study enhances the accuracy of critical parameter predictions in NPPs during failure conditions, thereby improving the emergency response capabilities of NPPs in the event of failures and accidents.