改进克里金模型辅助的双档案在线数据驱动进化算法

Improved Kriging Model-Assisted Two-Archive Online Data-Driven Evolutionary Algorithm

在现实世界中,许多进化优化问题难以找到合适的评估函数或评估代价十分昂贵,这在进化优化算法求解现实中的优化问题时提出诸多挑战。近年来,为了解决进化优化算法评估代价昂贵的问题,数据驱动的进化优化应运而生。数据驱动的进化优化的基本思想,就是通过充分利用数据的作用,训练代理模型辅助进化优化过程。一般是将代理模型用于近似真实昂贵函数评估,实现廉价的评估过程,提升算法性能。根据KTA2算法,本文提出了一种改进克里金模型辅助的双档案在线数据驱动进化算法KTA2_addModel4。在KTA2算法中,由训练的三种克里金模型作为代理模型:全部数据集训练的敏感模型、无较大影响点训练的不敏感模型1和无较小影响点训练的不敏感模型2。在改进的算法KTA2_addModel4中,增加了一种同时去掉较小影响点和较大影响点训练的不敏感模型3。通过在测试函数上与KTA2算法和其他代理辅助的数据驱动进化算法对比,证明提出的KTA2_addModel4算法改进了代理模型的质量,提升了算法的性能。

In the real world, many evolutionary optimization problems are difficult to find a suitable evaluation function or the evaluation cost is very expensive, which poses many challenges for evolutionary optimization algorithm to solve real optimization problems. In recent years, in order to solve the problem of expensive evaluation of evolutionary optimization algorithms, data-driven evolutionary optimization came into being. The basic idea of data-driven evolutionary optimization is to train the agent model to assist the evolutionary optimization process by making full use of the role of data. Generally, the agent model is used to approximate real expensive function evaluation, so as to realize the cheap evaluation process and improve the performance of the algorithm. According to the KTA2 algorithm, this paper proposes an improved Kriging Model-assisted two-archive online data-driven evolutionary algorithm KTA2_addModel4. In the KTA2 algorithm, three Kriging models are trained as surrogate models: the sensitive model trained by all data sets, the insensitive model 1 trained without large influence points and the insensitive model 2 trained without small influence points. In the improved algorithm KTA2_addmodel4, an insensitivity model 3 is added to remove both the small influence points and the large influence points. By comparing with KTA2 algorithm and other surrogate-assisted data-driven evolutionary algorithms in the test function, it is proved that the proposed KTA2_addmodel4 algorithm improves the quality of the surrogate model and the performance of the algorithm.