Nuclear reactor modeling is an important task for understanding the system design, safe operating conditions, and interdependencies among system variables in the nuclear power plant. This task is challenging because nuclear power plants exhibit complex, high-dimensional, nonlinear dynamics that are highly sensitive to parameter variations. Since it is impossible to conduct all experiments in practice due to safety concerns, highly accurate simulations are essential for nuclear reactor studies. Traditional system modeling requires an in-depth understanding of the underlying physical phenomena, including the system of coupled governing equations. Highly accurate methods, such as the Finite Difference, Finite Element, and Finite Volume methods, have been widely used. However, they require significant computational resources, often making them unsuitable for real-time operation and control. Given these drawbacks and recent advancements in data science, model reduction and Machine Learning techniques have emerged as promising alternatives. These approaches aim to predict system dynamics within reasonable computational times, enabling quasi-real-time prediction. In particular, data-driven modeling does not rely on governing equations but instead learns complex nonlinear relationships directly from data, enabling the efficient reconstruction and prediction of unseen scenarios. These approaches are also well-suited for handling and extracting insights from noisy datasets. In this work, a novel Data-Driven Reduced Order Modeling (DDROM) method, the Parametric Dynamic Mode Decomposition (Parametric DMD), is implemented to study the parametric dynamics of the Molten Salt Fast Reactor (MSFR). Unlike other reactors, the MSFR is described by Delay Differential Equations (DDEs), which are highly nonlinear and more complex to solve than standard PDEs. The goal of this work is therefore to assess whether the novel Parametric DMD algorithm can effectively model such a complex system, providing a simplified representation suitable for future applications in online monitoring and control.
La modellizzazione dei reattori nucleari è un compito importante per comprendere la progettazione del sistema, le condizioni operative di sicurezza e le interdipendenze tra le variabili del sistema in una centrale nucleare. Si tratta di un compito impegnativo perché le centrali nucleari presentano dinamiche complesse, ad alta dimensionalità e non lineari, altamente sensibili alle variazioni dei parametri. Poiché è impossibile condurre tutti gli esperimenti nella pratica per motivi di sicurezza, per gli studi sui reattori nucleari sono essenziali simulazioni altamente accurate. La modellizzazione tradizionale dei sistemi richiede una comprensione approfondita dei fenomeni fisici sottostanti, compreso il sistema di equazioni governanti accoppiate. Sono stati ampiamente utilizzati metodi altamente accurati, come il metodo delle differenze finite, degli elementi finiti e dei volumi finiti. Tuttavia, essi richiedono notevoli risorse computazionali, che spesso li rendono inadatti al funzionamento e al controllo in tempo reale. Considerati questi inconvenienti e i recenti progressi nella scienza dei dati, la riduzione dei modelli e le tecniche di apprendimento automatico sono emerse come alternative promettenti. Questi approcci mirano a prevedere le dinamiche del sistema con tempi di calcolo ragionevoli, consentendo una previsione quasi in tempo reale. In particolare, la modellazione basata sui dati non si basa su equazioni governanti, ma apprende relazioni non lineari complesse dai dati, consentendo la ricostruzione e la previsione efficiente di scenari non visibili. Questi approcci sono anche adatti per gestire ed estrarre informazioni da set di dati rumorosi. In questo lavoro, viene implementato un nuovo metodo di modellazione a ordine ridotto basato sui dati (DDROM), la decomposizione dinamica parametrica (Parametric DMD), per studiare le dinamiche parametriche del reattore veloce a sali fusi (MSFR). A differenza di altri reattori, l'MSFR è descritto da equazioni differenziali con ritardo (DDE), che sono altamente non lineari e più complesse da risolvere rispetto alle PDE standard. L'obiettivo di questo lavoro è quindi quello di valutare se il nuovo algoritmo Parametric DMD sia in grado di modellare efficacemente un sistema così complesso, fornendo una rappresentazione semplificata adatta a future applicazioni nel monitoraggio e controllo online.
Data-Driven Reduced Order Modeling using parametric dynamic mode decomposition: an application on molten salt fast reactors
TRUONG, KIM ANH NGOC
2024/2025
Abstract
Nuclear reactor modeling is an important task for understanding the system design, safe operating conditions, and interdependencies among system variables in the nuclear power plant. This task is challenging because nuclear power plants exhibit complex, high-dimensional, nonlinear dynamics that are highly sensitive to parameter variations. Since it is impossible to conduct all experiments in practice due to safety concerns, highly accurate simulations are essential for nuclear reactor studies. Traditional system modeling requires an in-depth understanding of the underlying physical phenomena, including the system of coupled governing equations. Highly accurate methods, such as the Finite Difference, Finite Element, and Finite Volume methods, have been widely used. However, they require significant computational resources, often making them unsuitable for real-time operation and control. Given these drawbacks and recent advancements in data science, model reduction and Machine Learning techniques have emerged as promising alternatives. These approaches aim to predict system dynamics within reasonable computational times, enabling quasi-real-time prediction. In particular, data-driven modeling does not rely on governing equations but instead learns complex nonlinear relationships directly from data, enabling the efficient reconstruction and prediction of unseen scenarios. These approaches are also well-suited for handling and extracting insights from noisy datasets. In this work, a novel Data-Driven Reduced Order Modeling (DDROM) method, the Parametric Dynamic Mode Decomposition (Parametric DMD), is implemented to study the parametric dynamics of the Molten Salt Fast Reactor (MSFR). Unlike other reactors, the MSFR is described by Delay Differential Equations (DDEs), which are highly nonlinear and more complex to solve than standard PDEs. The goal of this work is therefore to assess whether the novel Parametric DMD algorithm can effectively model such a complex system, providing a simplified representation suitable for future applications in online monitoring and control.| File | Dimensione | Formato | |
|---|---|---|---|
|
Master_Thesis_TRUONG.pdf
accessibile in internet solo dagli utenti autorizzati
Dimensione
9.26 MB
Formato
Adobe PDF
|
9.26 MB | Adobe PDF | Visualizza/Apri |
I documenti in POLITesi sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.
https://hdl.handle.net/10589/243770