EVENT DOSSIER
The researchers applied the SHRED network to the deployed TRIGA Mark II research reactor, aiming to achieve reliable state estimation for the engineering system. This architecture utilizes synthetic temperature data generated by fluid dynamics models and experimental temperature data, processing sparse measurements and noise through integrated strategies. All relevant fields can be reconstructed in real time without the need for hyperparameter adjustment. Experimental evaluations show that this method is robust under physical constraints and with low dynamic sensor locations, with a relative error of less than 4% in average Euclidean norm and a root-mean-square temperature error of 1.52 K (better than the 1.85 K of the CFD model). It also has self-correction capabilities, making it suitable for constructing interpretable monitoring and control systems for nuclear reactor digital twins.
研究人员将浅层循环解码器（SHRED）网络应用于部署的 TRIGA Mark II 研究堆，旨在实现工程系统的可靠状态估计。该架构利用流体动力学模型生成的合成温度数据与实验温度数据，通过集成策略处理稀疏测量和噪声，无需超参数调整即可在实时重建感兴趣的所有场。实验评估显示，该方法在物理约束和低动态传感器位置下表现稳健，其平均欧氏范数相对误差低于 4%，温度均方根误差为 1.52 K（优于 CFD 模型的 1.85 K），具备自我修正能力，适用于构建核反应堆数字孪生体的可解释监控与控制。