Multi/Many-Objective Optimization Via A New Preference Indicator

The development of preference-based optimizers has become an important trend in multi/many-objective optimization. Among those, the knees play a vital role in environmental selection process. Especially, they can accelerate convergence and maintain a high degree of diversity. Motivated by this, we suggest a new knees driven evolutionary algorithm based on pruning-power indicator to solve multi/many-objective problems. Here, the pruning power of a solution represents the number of points dominated by the solution in the local partition, which is used to identify knees. Then, an efficient pruning-power indicator is developed and then is proven mathematically to be able to characterize the solutions and further reduce the complexity of the hypervolume measurement. Based on this indicator, the algorithm uses angle-based partitioning and the nondominating sorting to accelerate convergence and maintain diversity of solutions. Finally, the algorithm is validated experimentally by using several DTLZ and WFG test problems and common performance measure. Experimental results validate the effectiveness of the proposed algorithm.