Estimation of distribution algorithms for nuclear reactor fuel management optimisation

In this paper, estimation of distribution algorithms (EDAs) are used to solve nuclear reactor fuel management optimisation (NRFMO) problems. Similar to typical population based optimisation algorithms, e.g. genetic algorithms (GAs), EDAs maintain a population of solutions and evolve them during the optimisation process. Unlike GAs, new solutions are suggested by sampling the distribution estimated from all the solutions evaluated so far. We have developed new algorithms based on the EDAs approach, which are applied to maximize the effective multiplication factor (Keff) of the CONSORT research reactor of Imperial College London. In the new algorithms, a new 'elite-guided' strategy and the 'stand-alone' Keff with fuel coupling is used as heuristic information to improve the optimisation. A detailed comparison study between the EDAs and GAs with previously published crossover operators is presented. A trained three-layer feed-forward artificial neural network (ANN) was used as a fast approximate model to replace the three-dimensional finite element reactor simulation code EVENT in predicting the Keff. Results from the numerical experiments have shown that the EDAs used provide accurate, efficient and robust algorithms for the test case studied here. This encourages further investigation of the performance of EDAs on more realistic problems.