Application of Genetic Algorithm to Optimize Burnable Poison Placement in Pressurized Water Reactors

1
Application of Genetic Algorithm to Optimize
Burnable Poison Placement in Pressurized Water
ReactorsPresented by Serkan Yilmaz, Faculty
Advisor Prof. Kostadin IvanovSpecial Thanks to
Prof. Samuel Levine and Moussa Mahgerefteh
(Exelon Nuclear Company) Mechanical and Nuclear
Engineering Department, The Pennsylvania State
University
1) Why?
3) Genetic Algorithm Model
5) Conclusions

• Based on natural genetics and natural selection
• Robust search algorithm
• Binary representation
• Use black-box approach

• The innovation is to search all of the possible
  U/Gd fuel pin designs with variable number of
  U/Gd pins and concentration of Gd2O3 instead of
  using limited number of available designs
• An ordinary nuclear engineer can develop good BP
  designs automatically in hours with this tool.
  This would take an expert several weeks to
  achieve the same results
• This study contributes to the efforts to make
  nuclear electricity generation cost competitive
• As a result, the potential of 12.5 effective full
  power days (EFPD) savings in cycle length was
  achieved during optimization process
• 1 EFPD savings in cycle length refers to
  1,000, 000 more income due to electricity sale

• The current deregulated power generation market
• Minimize fuel cycle cost to reduce cost of
  electricity
• Develop an efficient and a practical optimization
  tool

2) What is the problem?

• Neutron absorbers (Burnable Poisons or BPs) such
  as Gadolinium are placed in the core for control
  of reactor operation. These BP disappear during
  the cycle due to absorption of neutron
• BP optimization problem refers to developing a BP
  loading pattern for a given core loading
  configuration that minimizes the total Gd amount
  in the core without violating peak pin power and
  soluble boron constraints

Figure 5 Solution Space
Representation of BP loading pattern in Genetic
Algorithm
Figure 4 Genetic Algorithm (GA) Flow Diagram
Figure 6 Fitness Functions
4) Results
Figure 2 Selection of Two Decision Variables (No
of U/Gd fuel pins and Gd2O3 conc.)
Figure 10 Good Solutions
6) Future Work
Figure 1 Three Mile Island (TMI)-1 Reference Core
Loading Pattern

• Accelerate the process with hybrid GA and
  artificial Neural Networks (ANNs) methodology

Figure 9 Good Solutions
Figure 7 GA run results in the solution space
Table 1 Best Solution
Table 2 Good Design Data
Figure 8 Fitness value during evaluation
Figure 3 Phenotype and Genotype Structure
Figure 11 Hybrid GA-ANN Algorithm