Multiobjective great deluge algorithm with two-stage archive support

Abstract

A multiobjective great deluge algorithm with a two-stage external memory support and associated search operators exploiting the experience accumulated in memory are introduced. The level based acceptance criterion of the great deluge algorithm is implemented based on the dominance of a new solution against its parent and archive elements. The novel two-stage memory architecture and the use of dominance-based level approach make it possible to exploit promising solutions that both lie on better Pareto fronts in objective space and that are diversely separated in variable space. In this respect, the first stage of the external memory is managed as a short-term archive that is updated frequently when a solution that dominates its parent or some individuals over the current Pareto front is extracted whereas the second stage is organized as a long-term memory that is updated only after a number of first stage insertions. The use of memory-based search supported by effective move operators and dominance-based implementation of level mechanism within the great deluge algorithm resulted in a powerful multiobjective optimization method. The success of the presented approach is illustrated using unconstrained (bound constrained) multiobjective test instances used in the CEC'09 contest of multiobjective optimization algorithms. Using the evaluation framework described in CEC'09 contest and in comparison to published results of well-known modern algorithms, it is observed that the presented approach performs better than majority of its competitors and is a powerful alternative for multiobjective optimization.