Topographical global optimization applied to nuclear reactor core design: Some preliminary results

The nuclear reactor core design optimization problem consists in adjusting several reactor cell parameters, such as dimensions, enrichment and materials, in order to minimize the average peak-factor in a three-enrichment-zone reactor, considering restrictions on the average thermal flux, criticality and sub-moderation. This problem is highly multimodal, requiring optimization techniques that overcome local optima. In order to do so, we use a clustering optimization technique based on the topographical information on the objective function called Topographical Global Optimization (TGO). This algorithm consists of three steps: a uniform random sampling of solutions in the search space, the construction of the topograph, and the application of a local optimization algorithm using the topograph minima as starting points. In this work, we use the Sobol quasi-random sequence to perform the first step and the Hooke-Jeeves direct search method (HJ), which is one of the less sophisticated algorithms of this type, for the third step. In spite of HJ's simplicity, the results are competitive in terms of fitness function values, being obtained at a computational cost one order of magnitude lower than the efforts required for achieving the best results so far. This fact suggests that better results can be obtained employing more modern and effective direct search methods. Nevertheless, as the problem attacked is quite challenging, the preliminary results show the potential of TGO to be applied to other nuclear science and engineering problems. For the best of our knowledge, this is the first time that TGO is applied to an engineering optimization problem. (C) 2013 Elsevier Ltd. All rights reserved.