Fuel loading pattern optimization of ALLEGRO fast reactor using genetic algorithms

Abstract

In this work the genetic algorithmic (GA) optimization technique is used to find the optimal fuel loading pattern in the experimental fast reactor ALLEGRO. The penalty function method is used to solve this multi-objective combinatorial optimization problem. To evaluate each loading pattern, an interface between the genetic algorithm code and the ERANOS 2.3 deterministic neutronic simulation code was developed. For the simulation of the fuel cycle in ERANOS, the diffusion approach and a 7-group energy structure are used to reduce the computational cost. An integer variable encoding scheme was defined to represent the fuel loading patterns. Two crossover operators, the partially mapped crossover (PMX) and the order crossover (OX), combined with the random mutation operator were implemented as an alternative to the classical operators used in GA. To evaluate the performance of the code the results are compared using two objective functions and two crossover operators. In the first case the operating cycle length is maximized by maximizing the k-eff value at the EOC by satisfying the constraints on the power peaking factor over the cycle, the excess reactivity at the BOC, and the linear heat generation rate constraints. In addition, a condition was imposed that there should be no more than two contiguous positions in the pattern to preserve the power distribution uniformity and to avoid high power peaking factors. The code works very well, and a good solution was found in a reasonable computation time. The best solution corresponds to the PMX operator, and the running time is twice as short compared to the OX operator. In the second case study, a power distribution flatness term is included in the objective function. In this case, a more uniform radial power distribution is obtained but associated with a decrease in cycle length due to the penalty of the flatness term.