Towards large-scale multiobjective optimisation with a hybrid algorithm for non-dominated sorting

Abstract

We present an algorithm for non-dominated sorting that is suitable for large-scale multiobjective optimisation. This algorithm is a hybrid of two previously known algorithms: the divide-and-conquer algorithm initially proposed by Jensen, and the non-dominated tree algorithm proposed by Gustavsson and Syberfeldt. While possessing the good worst-case asymptotic behaviour of the divide-and-conquer algorithm, the proposed algorithm is also very efficient in practice. In our experimental study it is shown to outperform both of its parents on the majority of problem instances, both sampled uniformly from a hypercube and having a single front, with as large as 10 6 points and up to 15 objectives.