Researchers from the Hefei Institutes of Physical Science, Chinese Academy of Sciences, have developed NPP-GPT, an AI framework that uses pre-trained large language models to accurately forecast nuclear power plant (NPP) operating parameters over the long term.

Their work has been published in Applied Energy.

Accurate long-term prediction of key NPP operating parameters is essential for ensuring safety and improving operational efficiency. However, the complexity of reactor systems and varying operating conditions generate high-dimensional data with strong correlations and long-range dependencies. Traditional small-scale models often struggle to capture these features, limiting their predictive accuracy and practical value.

In this study, the NPP-GPT framework addresses these challenges using a cross-modal transfer learning strategy. It first aligns numerical time-series data with the representation space of a pre-trained language model through input embedding reconstruction and self-supervised learning with random masking. Domain knowledge is then incorporated via LoRA (parameter-efficient fine-tuning) by adjusting the Q/V projections in GPT-2’s self-attention modules. This approach preserves the general capabilities of the pre-trained model while enhancing forecasting performance and maintaining training and deployment efficiency.

Tests show that NPP-GPT performs exceptionally across six typical operating-condition datasets. In multivariate, multi-step forecasting tasks, it outperforms several mainstream time-series methods, maintaining high accuracy even as the forecasting horizon extends. Evaluations under cross-condition transfer, noise interference, and missing-data scenarios demonstrate strong robustness and generalization.

This framework provides more reliable predictive information for online safety monitoring and operational decision support in nuclear power plants and opens new possibilities for applying large language models in the nuclear energy field.

Article link: https://doi.org/10.1016/j.apenergy.2026.127438

The framework of the proposed NPP-GPT in nuclear energy systems