Interactive and Interpretable Knee Point-Driven Multiobjective Optimization for Multiobjective Decision Making

In the optimization of real applications, it is challenging to find a representative solution set to cover the whole Pareto optimal front (PoF) of the problems due to the resource-intensive, time-consuming, and complex manifold of the PoF. Accordingly, preference information from the decision makers (DMs) is commonly utilized to guide the optimization process towards the regions of interest (ROIs) on the PoF. However, it is non-trivial for decision makers to articulate preferences without a priori knowledge. To fill the gap, this paper introduces the naturally interesting "knee point" on the PoF into the optimization and designs an interactive knee point driven optimization framework with interpretable preference information, aiming to facilitate DMs in preference articulation, decision making and further investigation on the optimization problems. Experimental results on a number of benchmark problems have verified the effectiveness of the proposed framework.