用最近邻分类方法预测多目标优化d-Pareto支配性

Predicting d-Pareto dominance in multi-objective optimization using nearest neighbor classification method

为进一步提高预测精度,修改候选解间原始Pareto支配性关系,提出了d-Pareto支配性最近邻预测方法。结合多目标优化的自身特点,给出了d-Pareto支配性最近邻预测框架,并论证了d-Pareto支配性预测比Pareto支配性预测具有低平均预测错误率。同时也初步研究了d-Pareto支配性预测与多目标进化算法的交互作用。对几个经典多目标优化问题进行实验,仿真结果表明d-Pareto支配性预测具有一定的可行性和有效性。

To improve predicting accuracy further, this paper modified original Pareto dominance relation among the candidate solutions, and proposed a new method named d-Pareto dominance prediction using nearest neighbor. Combined with the characteristics of multi-objective optimization, it described the framework of d-Pareto and the conclusion that a d-Pareto dominance prediction could obtain a lower average prediction error rate comparing with Pareto demonstrated dominance prediction. Besides,it also explored the interaction between d-Pareto dominance prediction and multi-objective evolutionary algorithms. Experiments on several classic MOPs were conducted and the simulation results show that prediction of d-Pareto dominance is feasible and effective.