Preferred Region Based Evolutionary Multi-objective Optimization Using Parallel Coordinates Interface

This work proposes a novel preference based evolutionary multi and many-objective optimization approach to search a specific region of the Pareto front. First, to know the overview of the entire Pareto front, the proposed approach roughly approximates it by using a representative MOEA/D with uniformly distributed weight vectors. Then, the obtained solutions are plotted on the parallel coordinates user interface (UI). In the proposed approach, the decision maker's preference can be specified as a region in the objective space while the conventional approaches use a single preference point in the objective space. It has an advantage when the decision maker has poor knowledge about the target problem. Next, the proposed approach rearranges the weight vectors to determine the search directions in the objective space inside the preferred region and performs MOEA/D with the rearranged weight vectors. The parallel coordinates UI is particularly suited to rearrange weight vectors and compatible with MOEA/D. Experimental results using DLTZ2 problems with 2-6 objectives show the proposed approach improves the approximation performance of the specific region of Pareto front.