基于多点加点准则的代理模型辅助社会学习微粒群算法

代理模型辅助的进化算法目前已广泛用于解决计算代价高的复杂优化问题.然而,大多数现有的代理辅助进化算法只适用于低维问题且仍然需要数千次昂贵的真实适应值评价来获得较优解.为此,提出一种基于多点加点准则的代理模型辅助的社会学习微粒群算法,用于解决高维问题并使用更少的评价次数.该算法选用高斯过程构造代理模型,以社会学习微粒群算法(SLPSO)作为优化器,提出一种基于相似度的多点加点规则(SMIC),用于选取需要使用原函数进行实际计算的候选解.在仿真实验中将该方法与现有研究成果进行比较,通过对50维sim100维的基准函数的测试,验证了所提出算法在有限的适应值计算次数下拥有更好的寻优性能,尤其是在高维优化问题上拥有更显著的优势.

Similarity-based multipoint infill criterion for surrogate-assisted social learning particle swarm optimization

The research on surrogate-assisted evolutionary algorithms has attracted increasing attention for time or resource consuming optimization problems over the past decades. However, most existing surrogate-assisted evolutionary algorithms still require thousands of expensive function evaluations to obtain acceptable solutions and are only applied to low-dimensional problems. Therefore, a surrogate-assisted optimization, called SMIC-SLPSO (Similarity-based multipoint infill criterion for surrogate-assisted social learning particle swarm optimization), is proposed for high dimensional problems with less computation. In the proposed algorithm, using Gaussian process as the surrogate model and considering SLPSO algorithm as an optimizer, a similarity-based multipoint infill criterion (SMIC) is proposed for searching the solutions to re-evaluation by using the original expensive problems. Simulation of experiments comparing the proposed algorithm with a few state-of-the-art surrogate-assisted evolutionary algorithms on benchmark functions from 50 to 100 dimension is carried out. The results proposed algorithm is able to achieve better or competitive solutions with a limited budget of exact evaluations, especially on higher dimensional problems.