基于元学习推荐的自适应群集智能最优化模型研究

In the operation and optimization area, addressing global optimization problems using swarm intelligence (SI) algorithms has gained huge popularity most recently. Previous research mostly proposed new SI algorithms or improved existing SI algorithms for a number of specific problems. As a result, their performance may vary significantly from problem to problem, which impairs the applicability of SI algorithms. Inspired by the meta-learning processing in machine learning, in this project we propose to develop an adaptive swarm intelligence model (ASI) which enables adaptive decision for a diverse range of global optimization problems. By dynamically mapping the meta-features of problem space and the performance of SI algorithms, it adaptively selects the most appropriate algorithms (strategies) from the algorithm depository for different problems. The main research topics include: (1) develop a set of meta-features for capturing unknown problem space of global optimization; (2) build a mapping and recommender system to model the correlations between the meta-features of unknown problem space and the performance of SI algorithms; (3) Dynamically build multiple sub-swarms corresponding to the dynamic recommended algorithms (strategies), introduce new co-existing relationship (cooperation-competition)to the sub-swarms and construct an adaptive SI model; (4) The proposed SI model is verified through theoretical analysis, comparison experiment and a real world application in terms of effectiveness and applicability. This project is a multidisciplinary investigation on operation optimization and machine learning, where the correlations between the meta-features of global optimization problems and the performance of SI algorithms would be modeled and investigated. The study is envisioned to bridge existing gaps and provide a new perspective for current optimization research.

基于群集智能（SI）的最优化算法对全局优化问题的研究是当前运筹优化领域的一个热点。目前研究主要针对特定问题进行算法设计或改进，因而对不同优化问题的性能波动较大，适用性难以保证。本项目结合机器学习领域元学习思想，研究一种自适应群集智能最优化模型，对全局优化问题进行自适应决策。通过问题元特征与优化性能的动态匹配，智能地选择算法（策略）组合进行快速有效的求解。研究内容包括：（1）构建对全局优化情境下描述问题空间的元特征集；（2）通过元学习建立问题元特征与算法性能间的映射推荐模型；（3）对推荐的算法组合进行动态转化，引入新型共生关系（协同-竞争）构建自适应异质群集智能模型；（4）通过理论分析、数值实验与供应链优化实例，研究该模型的有效性与适用性。本项目是运筹优化与机器学习的交叉研究，对全局优化问题元特征与群集智能算法性能间的映射关联进行了建模与展示，为现有最优化算法的研究提供了新的理论视角与研究。

基于群集智能（SI）的最优化算法对全局优化问题的研究是当前运筹优化领域的一个热点。目前研究主要针对特定问题进行算法设计或改进，因而对不同优化问题的性能波动较大，适用性难以保证。本项目结合机器学习领域元学习思想，研究一种自适应群集智能最优化模型，对全局优化问题进行自适应决策。通过问题元特征与优化性能的动态匹配，智能地选择算法（策略）组合进行快速有效的求解。研究内容包括：（1）构建对全局优化情境下描述问题空间的元特征集；（2）通过元学习建立问题元特征与算法性能间的映射推荐模型；（3）对推荐的算法组合进行，引入新型共生关系（协同-竞争）构建自适应异质群集智能模型；（4）通过理论分析、数值实验与在物流、交通、能源领域的实例，研究该模型的有效性与适用性。本项目研究进展顺利，完成了既定的研究目标与计划。共发表学术论文13篇，其中项目主持人以第一作者+通讯作者身份发表SCI/SSCI共7篇。项目的成果已经成功运用于能源、交通等管理与工程领域。在医学智能诊断方面的应用也在探索中。本项目是运筹优化与机器学习的交叉研究，对全局优化问题元特征与群集智能算法性能间的映射关联进行了建模与展示，为现有最优化算法的研究提供了新的理论视角与研究。