Homogeneous grouping of nuclear fuel cans through simulated annealing and tabu search

Collections of cans containing nuclear fuel have to be grouped in batches that are as homogeneous as possible with respect to several criteria. This highly combinatorial problem, which can be described as grouping or clustering, is tackled using simulated annealing and tabu search. Both approaches are submitted to extensive experimentation on a real data set and several artificial ones. Two variants of the basic approaches called ‘Locally optimized simulating annealing’ and ‘Tabu search with variable offset’ are also tested. Sensitivity to parameter choice and to problem size are investigated. All four algorithms outperform a local search heuristic previously proposed in the literature; on the class of instances dealt with, a remarkably stable ranking of the four algorithms emerges.