高维多目标进化算法的研究与应用

Multi-objective optimization problems widely exist in engineering practice and social science. Compared with traditional methods, evolutionary algorithms are more suitable for solving multi-objective optimization problems. With the development of computer technology and engineering practice, many-objective optimization is becoming a hot issue in the field of multi-objective optimization. Nowadays, the research and application of many-objective optimization still has a broad space for development. To further improve the search capability of algorithms for many-objective optimization problems by aiming at the problem that deterioration of multi-objective algorithms arise when the number of objectives increase, in this project, we focus on the dimensionality reduction techniques for many-objective optimization problems based on feature selection and its self-adaptive mechanism, the group of feature selection, the Pareto front approximation, and propose novel many-objective evolutionary algorithms based on them. To test the performance of the proposed many-objective optimization algorithms, they will be used for solving the classical continuous many-objective optimization problems and the vehicle routing problem.

多目标优化问题广泛存在于工程实践和社会科学中。和传统方法相比，进化算法更适用于求解多目标优化问题。随着计算机技术与工程实践的发展，高维多目标优化逐渐成为多目标优化领域中的一个热点问题。如今，高维多目标优化依然有很大的研究和应用空间。为进一步提高高维多目标进化算法的搜索能力，本项目针对目标数量过多导致算法搜索能力下降的问题，研究基于目标选择的高维多目标降维方法及其自适应机制，研究目标选择的群体化，以及研究Pareto前沿近似方法，从而提出新的更优秀的高维多目标进化算法。最后，把提出的新算法用于求解经典连续高维多目标优化问题和车辆路径问题以测试算法的性能。

多目标优化问题是工程实践中的一类常见问题，高维多目标优化问题的求解一直是进化计算领域的一个研究热点。针对目标数量增加导致选择压力下降的问题，本项目提出了双聚类算子。该算子根据算法的搜索状态，通过寻找冲突目标对种群进行基于目标选择的分组降维，实现了目标选择的自适应和群体化。另一方面，针对当前已有算法对决策空间分析的不足，在基于参考点的Pareto前沿近似方法的基础上，本项目提出了基于角度的决策空间聚类算子，通过对种群在决策空间分组提高了算法在决策空间上的局部搜索能力。本项目研究了高维多目标优化在车辆路径问题、心电信号分类以及关联规则挖掘上的应用。基于上述研究，发表了期刊论文6篇、会议论文1篇以及申请获批软件著作权1项。