Application of Differential Evolution algorithms to multi-objective optimization problems in mixed-oxide fuel assembly design

Highlights: • Performance is comparable to Genetic Algorithms on assembly optimization. • Algorithms can handle complex nuclear assembly optimization problems. • Demonstrated no statistical performance impact due to control parameter sensitivity. - Abstract: Multi-objective optimization of nuclear engineering fuel assembly design problems is particularly difficult due to the highly non-linear interactions of a large number of possible variables. In addition, effective optimization algorithms are often highly problem-dependent and require extensive tuning, which reduces their applicability to the real world. To address this issue, Differential Evolution (DE) algorithms have been proposed as a new and effective method for heterogeneous fuel assembly optimization design problems. This paper presents the first complete study to investigate their applicability and performance. Firstly, two multi-objective DE algorithms have their performance compared against an Evolutionary Algorithm (EA) from the literature in optimizing a CORAIL mixed-oxide (MOX) fuel assembly for maximum plutonium content and minimum power peaking factor. Statistical analysis of the results shows the DE algorithms exhibit superior performance to the EA. The DE algorithms are then used to optimize a MOX fuel assembly with gadolinia poison, with results showing DE produces assembly designs comparable in performance to those in the literature. Finally, a sensitivity study is conducted on the control parameters of the better performing of the DE algorithms. Results indicate DE performance remains consistent for a wide range of values of both control parameters, suggesting the algorithm is able to perform effectively without requiring user expertise or effort to find the 'optimal' control parameter settings.