Paper detail

An improved hyperboxing algorithm for calculating a Pareto front representation

When solving optimization problems with multiple objective functions we are often faced with the situation that one or several objective functions are non-convex or that we can not easily show the convexity of all functions involved. In this case a general algorithm for computing a representation of the nondominated set is required. A suitable approach consists in a so-called hyperboxing algorithm that is characterized by splitting the objective space into axis-parallel hyperrectangles. Thereby, only the property of nondominance is exploited for reducing the so-called search region. In the literature such an algorithm has already shown to provide a very good coverage of the Pareto front relative to the number of representation points calculated. However, the computational cost for the algorithm was prohibitive for problems with more than five objectives. In this paper, we present algorithmic advances that improve the performance of the algorithm and make it applicable to problems with up to nine objectives. We illustrate the performance gain and the quality of the representation for a set of test problems. We also apply the improved algorithm to a real world problem in the field of radiotherapy planning.