ML24263A256
| ML24263A256 | |
| Person / Time | |
|---|---|
| Issue date: | 09/17/2024 |
| From: | Nathanael Hudson, Tehrani N, Peter Yarsky NRC/RES/DSA/AAB |
| To: | |
| References | |
| Download: ML24263A256 (1) | |
Text
Autonomous Control Algorithms to Simulate Boiling Water Reactor Cycle Depletion Using BALTO NRC Team:
Nate Hudson, Ph.D. Nathanael.Hudson@nrc.gov Nazila Tehrani, Ph.D. Nazila.Tehrani@nrc.gov Peter Yarsky, Ph.D. Peter.Yarsky@nrc.gov
NRC exploring innovative tools
- Perform independent transients/accidents analysis
- Increase staff efficiency
- Identify/focus safety significant issues
- Address emergent issues 2
Models not associated with any licensing action No licensee information Automate LWR core/cycle design Generic core designs for modern fuel types Autonomous control methods 3
Motivation
Control excess reactivity during cycle
- burnable poisons
- control blades
- flow control windows Competing goals
- meeting desired cycle energy
- maintaining safety margins
- minimizing duty related fuel failures High-dimensionality of BWR cores
- issues with symmetry
- very large number of types of fuel elements BWR cores complex to design 4
BWR cores complex to design 5
Reactivity Balance in BWR 6
TRACE/PARCS for LWR transients
- Coupled neutronics/TH transient
- normal
- anticipated transients
- accident conditions
- current reactors
- advanced reactors
- 12 major plant types (5 BWR)
- Standard models
- Efficient confirmatory calculations 7
BALTO :
BWR Autonomous Learning Tasks Optimizer 8
Picture from Bean
9 NRC Models: Traditional or Autonomous?
RES is developing generic LWR TRACE/PARCS models for transient analysis (DBAs and AOOs)
BALTO is a software package to autonomously generate equilibrium cores for Boiling Water Reactors satisfying desired economic and safety constraints Equilibrium Cycle is an analysis concept characterized by the indefinite repetition of bundle shuffle and loading patterns between sequential cycles, with the same control blade pattern within each of the cycles.
General optimization objectives: Minimizing fuel cost; Maintaining criticality throughout the cycle; Ensuring thermal limits are satisfied ; Minimizing rod movement, if possible
BALTO: Flowchart 10 BALTO is scalable and generic, it works for any BWR core size and any fuel assembly design
Genetic Algorithm
- Inspired by the Theory of Evolution
- Variables of interest are represented by 1 or more genes inside a chromosome
- GAs create a diverse population of chromosomes
- New generations of offspring are created by crossing over the genes of two parents and introducing random mutations
- Probability of being selected as a parent depends on cost or "fitness" function 11 Example of loading pattern crossover from Chen et al.
Training PARCS Simulations 1.
Generate random population of loading patterns 2.
Distribute LPs to multiple nodes of HPC 3.
Generate CRP and CF for each LP using RL 4.
Evaluate each LP/CRP/CF combo using PARCS 5.
Save results of each run 12 6.
Update RL algorithm with PARCS results 7.
Return to Step 3 until cycle reaches equilibrium or max iterations reached.
8.
Return to step 2 until a sufficient amount of data is generated 9.
Train surrogate models
Reference:
M. R. Oktavian, J. Nistor, J. Gruenwald, and Y. Xu (2023).
Summary 13
- Automate LWR core/cycle design
- Create models not associated with any specific licensing action
- Use autonomous control methods
References C.-H. Chen, P.-F. Lin, and Y.-N. Huang, PLAYGO: Program for BWR LP-CRP automatic yielding by using GA optimization, Nuclear Engineering and Design, vol. 322, pp. 427-443, 2017, ISSN: 0029-5493. DOI: https://doi.org/10.1016/j.nucengdes.2017.07.
007. [Online]. Available: https://www.sciencedirect.com/science/article/pii/
S0029549317303394.
M. R. Oktavian, J. Nistor, J. Gruenwald, and Y. Xu (2023),Preliminary development of machine learning-based error correction model for low-fidelity reactor physics simulation, https://www.sciencedirect.com/science/article/pii/S030645492300107X 14
Definitions 15 AOO: Anticipated Operational Occurrences BALTO: BWR Autonomous Learning Tasks Optimizer BWR: Boiling water reactor CF: Core Flow CR: Control Rod CRP: Control Rod Patterns DBA: Design Basis Accident EOC: End of cycle FA: Fuel Assembly GA: Genetic Algorithm HPC: High Performance Computing Keff: Effective Multiplication Factor LP: Loading Patterns LWR: Light water reactor MAPRAT: Maximum Average Planar Linear Heat Generation Rate MFLCPR: Maximum Fraction of Limiting Critical Power Ratio MFLPD: Maximum Fraction of Limiting Power Density PARCS: NRC Core Simulator PWR: Pressurized water reactor RES: Office of Nuclear Regulatory Research RL: Reinforcement Learning SDM: Shutdown Margin TH: Thermal-Hydraulics TRACE: NRC Thermal-Hydraulics code
Questions 16