Utilizing physics-informed synthetic data to predict reactor operations

The ability to predict how a reactor will operate, understand when anomalous conditions arise, and ensure a reactor is being operated as expected is crucial for deploying new nuclear facilities. Digital twins serve as a unique solution for recognizing reactor behavior; however, they require data to be effectively utilized. For next-generation reactors, these data may not currently be available. The following work aims to validate the efficacy of using synthetic data for modeling reactor operations and enabling digital twin development of nuclear reactors prior to physical data acquisition. Physics-informed synthetic data is generated using surrogate models of reactor physics simulations of the Idaho State University AGN-201 research reactor for efficient data generation. This synthetic data generation process is validated on AGN-201 operational data to understand biases in the synthetic data generation process. Finally, the physics-informed synthetic data is used to train a long short-term memory (LSTM) recurrent neural network to dynamically predict reactor operations and allow for consistent model updating as more data becomes available. The LSTM was validated using both synthetic and real operational data. By training an LSTM model on synthetic data and validating the model on operational data, this approach can help bridge the gap between the virtual and physical sides of a digital twin.