Transformers and Long Short-Term Memory Transfer Learning for GenIV Reactor Temperature Time Series Forecasting DOI Creative Commons
Stella Pantopoulou, A.C. Cilliers, Lefteri H. Tsoukalas

et al.

Energies, Journal Year: 2025, Volume and Issue: 18(9), P. 2286 - 2286

Published: April 30, 2025

Automated monitoring of the coolant temperature can enable autonomous operation generation IV reactors (GenIV), thus reducing their operating and maintenance costs. Automation be accomplished with machine learning (ML) models trained on historical sensor data. However, performance ML usually depends availability large amount training data, which is difficult to obtain for GenIV, as this technology still under development. We propose use transfer (TL), involves utilizing knowledge across different domains, compensate lack TL used create pre-trained data from small-scale research facilities, then fine-tuned monitor GenIV reactors. In work, we develop Transformer long short-term memory (LSTM) networks by them measurements thermal hydraulic flow loops water Galinstan fluids at room Argonne National Laboratory. The are re-trained minimal additional perform predictions time series high obtained Engineering Test Unit (ETU) Kairos Power. LSTM investigated varying size lookback window forecast horizon. results study show that have lower prediction errors than Transformers, but increase more rapidly increasing horizon compared errors.

Language: Английский

Transformers and Long Short-Term Memory Transfer Learning for GenIV Reactor Temperature Time Series Forecasting DOI Creative Commons
Stella Pantopoulou, A.C. Cilliers, Lefteri H. Tsoukalas

et al.

Energies, Journal Year: 2025, Volume and Issue: 18(9), P. 2286 - 2286

Published: April 30, 2025

Automated monitoring of the coolant temperature can enable autonomous operation generation IV reactors (GenIV), thus reducing their operating and maintenance costs. Automation be accomplished with machine learning (ML) models trained on historical sensor data. However, performance ML usually depends availability large amount training data, which is difficult to obtain for GenIV, as this technology still under development. We propose use transfer (TL), involves utilizing knowledge across different domains, compensate lack TL used create pre-trained data from small-scale research facilities, then fine-tuned monitor GenIV reactors. In work, we develop Transformer long short-term memory (LSTM) networks by them measurements thermal hydraulic flow loops water Galinstan fluids at room Argonne National Laboratory. The are re-trained minimal additional perform predictions time series high obtained Engineering Test Unit (ETU) Kairos Power. LSTM investigated varying size lookback window forecast horizon. results study show that have lower prediction errors than Transformers, but increase more rapidly increasing horizon compared errors.

Language: Английский

Citations

0