Speaker
Description
The application of AI is in vogue in almost all fields, including nuclear power engineering. A nuclear reactor can be safely and optimally operated when various safety metrics are reliably obtained from the core power distribution. The failure of in-core instrumentation poses a challenge for continued safe and reliable operation, and risks in forced shut down to ensure core integrity. This project sought suitable Machine Learning techniques for In-Core Power Mapping of an HPR1000, based upon reactor core simulations of the first fuel cycle of an HPR1000; using a locally developed depletion calculation program and available operational data. Using available detector data for charge and relative power distribution, LSTMNs and TLFFNs, used in conjunction, have been found suitable to potentially allow extending plant operation for a few hours in the event of more than 25% of detector failures. Prediction of temperatures based on power predictions is also proposed and demonstrated to validate against measured temperatures for practical implementation. GUIs, implementing the best of explored techniques, have also been developed.