Part 1 of 2 Parts
Genetic algorithms (GA) are a type of computer software that utilize evolutionary processes imitating biology. A bunch of solutions to a problem are developed that are all slightly different from each other. Then they are set loose to solve the problem. A few of the best solutions are then used to create a new set of slightly different solutions. In some cases, two solutions are combined to make a new solution and then the process is run again. Through successive “generations”, better and better solutions are “evolved.”
When nuclear fuel rod assemblies (FAs) are burned in a nuclear reactor core, they are not all consumed at the same rate. After eighteen to twenty-four months, the reactor is shut down and the reactor vessel is opened up to insert new FAs. This process is expensive and time consuming. Usually, about one third of the FAs that show the most depletion is replaced. The new assemblies and the two thirds of the assemblies that are left are rearranged to form a new configurations of FAs in the core. The arrangement of FAs is referred to as a “Loading Pattern” (LP). The new arrangement is supposed to satisfy several criteria but some of these are in conflict. Maximizing the production of energy might work against insuring safety regulations and operational constraints and visa versa.
Figuring out the best arrangement of FAs to simultaneously satisfy the conflicting criterion is an example of what is called a “classical discrete optimization problem”. The search space of possible arrangements is huge. This is a multi-objective, nonlinear, nonconvex, NP-hard combinatorial problem. In a standard pressurized water reactor, the most common kind of commercial power reactor, there are about two hundred FAs. There are also about two hundred different possible ways to arrange those assemblies. These different LPs each have to be evaluated, analyzed and characterized using sophisticated computer simulation with complex calculations. Assuming that it takes one second to evaluate a single core, it would take about 10360 years to explore the entire search space of possible configurations. The current age of the universe is only about 1010 years. Obviously, a better and faster way is needed to evaluate possible core configurations.
The genetic algorithms mentioned above are well-known methods of solving this type of optimization problem. They generate and evolve solutions to move through the huge search space of possible LPs to find the best one. Each LP is considered as an individual. Parent LPs are selected based on how well they fit the criterion and then they are mutated and recombined to create the next generation. After many generations, optimal LPs emerge from the process.
Dr. Erez Gilad and Ph.D. candidate Ella Israli are researchers at the Unit of Nuclear Engineering at Ben-Gurion University of the Negev (BGU). They recently published a study in the Annals of Nuclear Energy on the use of GAs for optimizing core design in nuclear reactors. They mentioned some of the problems that can be found with some uses of the GAs. They said that many GAs studies use old and outmoded GA implementation. These might disregard important and critical information such as the geometric structure of the core. Another problem would be the imposing of symmetric restrictions just to reduce the runtime of the algorithm. In their report, the authors develop, implement and evaluation novel GA methods using different case studies of LP design.
Please read Part 2