Driving Evolutionary Multiobjective Search by a Scalarizing Fitness Function

抄録

In evolutionary multiobjective optimization (EMO), no a priori information about the decision maker's preference is assumed. EMO algorithms are usually designed to search for well-distributed Pareto-optimal solutions. A final solution is chosen from the obtained solutions by the decision maker. It is, however, not easy for EMO algorithms to find a good solution set that well approximates the entire Pareto front of a large-scale multiobjective combinatorial optimization problem. In some cases, only a tiny portion of the Pareto front is approximated by the obtained solutions. In other cases, the obtained solutions are far from the Pareto front. In order to improve the search ability of EMO algorithm to efficiently find Pareto-optimal solutions, we implement a hybrid algorithm by combining a scalarizing fitness function into NSGA-II. In our hybrid approach, parent selection is performed based on a scalarizing fitness function while Pareto dominance relation is used for generation update.