Graph Neural ODE Digital Twins for Control-Oriented Reactor Thermal-Hydraulic Forecasting Under Partial Observability
2026-04-08 • Machine Learning
Machine Learning
AI summaryⓘ
The authors developed a smart computer model that predicts how heat and fluid move inside advanced reactors, even where there are no physical sensors. They used a special kind of neural network that understands the system like a connected graph and updates predictions continuously over time. Their model is fast, accurate, and can fill in missing information about unmeasured parts. They also showed it can adapt well from simulated data to real experimental data with minimal retraining. Overall, this helps monitor reactors in real-time without needing sensors everywhere.
Graph Neural NetworkNeural Ordinary Differential EquationThermal-hydraulicsSupervisory ControlSurrogate ModelPartial ObservabilityReynolds NumberUncertainty QuantificationSim-to-Real TransferEnsemble Modeling
Authors
Akzhol Almukhametov, Doyeong Lim, Rui Hu, Yang Liu
Abstract
Real-time supervisory control of advanced reactors requires accurate forecasting of plant-wide thermal-hydraulic states, including locations where physical sensors are unavailable. Meeting this need calls for surrogate models that combine predictive fidelity, millisecond-scale inference, and robustness to partial observability. In this work, we present a physics-informed message-passing Graph Neural Network coupled with a Neural Ordinary Differential Equation (GNN-ODE) to addresses all three requirements simultaneously. We represent the whole system as a directed sensor graph whose edges encode hydraulic connectivity through flow/heat transfer-aware message passing, and we advance the latent dynamics in continuous time via a controlled Neural ODE. A topology-guided missing-node initializer reconstructs uninstrumented states at rollout start; prediction then proceeds fully autoregressively. The GNN-ODE surrogate achieves satisfactory results for the system dynamics prediction. On held-out simulation transients, the surrogate achieves an average MAE of 0.91 K at 60 s and 2.18 K at 300 s for uninstrumented nodes, with $R^2$ up to 0.995 for missing-node state reconstruction. Inference runs at approximately 105 times faster than simulated time on a single GPU, enabling 64-member ensemble rollouts for uncertainty quantification. To assess sim-to-real transfer, we adapt the pretrained surrogate to experimental facility data using layerwise discriminative fine-tuning with only 30 training sequences. The learned flow-dependent heat-transfer scaling recovers a Reynolds-number exponent consistent with established correlations, indicating constitutive learning beyond trajectory fitting. The model tracks a steep power change transient and produces accurate trajectories at uninstrumented locations.