Development of the batcher tool for the optimization of the batch distribution

This thesis is the result of a 600-hours internship at ENEA. The objective of the internship was to develop an optimization algorithm aimed at determining the optimal distribution of fuel assemblies into batches within a nuclear reactor. An optimal batch distribution meets certain criteria concerning reactivity and temperature distribution within the core map. Since the cooling and reactivity control systems must be dimensioned on a worst-case basis, it is necessary that all the cycles are as similar as possible. We employed a fitness function based on the first-order perturbation theory to determine the goodness of a distribution. Knowing the perturbed cross-sections, which were provided as input, it was possible to calculate the variation from the reference states of the reactivity and a power functional in each phase of the cycle. After the fitness function was defined, it was optimized by using two methods. The first is an analytical algorithm that attempts to minimize the nonlinear objective function by exploring the possible solutions through successive linearization to find the best solution. The second method is a genetic algorithm, based on natural selection processes, such as cross-over, mutation and elitism. The two algorithms were applied to the test-case of ALFRED and the obtained results have been compared. In general, the analytical algorithm shows better results compared to the genetic one, both in terms of execution time and of the fitness function obtained. In any case, the results obtained from the two algorithms meet the requirements imposed during the modeling of the problem very well. However, due to the algorithms’ structure, it is expected that the choice of the best optimization algorithm will depend on the model of the problem to be solved. If the fitness function becomes very complicated, the computation of the analytical solution will take much longer, consequently making this method a worse choice.