复杂网络社区检测中的高维多目标优化算法研究

Accurate community detection in real complex networks has become a hot and difficult problem demanding prompt solution in the area of complex networks, which is essentially a complex Many-Objective Optimization Problem(MaOP). In this project, key technologies of many-objective optimization evolutionary algorithms, such as the evolution framework, elitist selection mechanism and decomposition mechanism, will be redesigned to significantly enhance algorithms’ performance in terms of proximity, diversity, velocity and stability. The characteristic and innovation of this project lie in not only make many-objective evolutionary algorithms suitable for complex many-objective optimization, but also realize their application in accurate community detection, by which the quality of the community partitions in terms of topological structure, attribute relevancy, fuzzy overlapping, dynamic evolution and users’ preference will be satisfied maximally at the same time, and thus improving the accuracy, stability and users’ satisfaction degree of community detection in fuzzy overlapping networks and dynamic networks.This project possesses wide actual application prospects and significant theoretical research value, which can not only reveal in depth the information contained in the community structure of complex networks, and satisfy users’ diversified demands to the fullest, but also deepen the theoretical development and wide application of many-objective optimization, and promote the improvement and incorporation of complex networks and many-objective optimization theory.

真实复杂网络环境中高精度的社区检测是目前复杂网络领域亟待解决的研究热点和难点，本质上是高维复杂多目标优化问题。本课题通过对高维多目标优化进化算法中模型框架、精英选择、多目标分解等核心关键技术进行改进，综合提升算法的收敛精度、解集分布性、收敛速度及稳定性。该课题的特色和创新之处在于，不仅提升了高维多目标优化进化算法对高维复杂多目标优化问题的求解能力，而且实现了基于高维多目标优化的高精度复杂网络社区检测，同时最大化满足真实复杂网络对拓扑结构、属性关联、模糊重叠、动态演化及用户偏好等特性的检测质量需求，提升模糊重叠社区检测和动态网络社区检测的精确性、稳定性及用户满意度。本项目具有广阔的实际应用前景和重要的理论研究价值，不仅能深度挖掘复杂网络社区结构中潜在的深层次信息，最大程度满足用户多样化检测需求，又能深化高维多目标优化理论的发展及实际应用范围，促进复杂网络与高维多目标优化理论的发展与融合。

社区结构广泛存在于现代复杂网络中，如社交网络、生物网络、工程网络等。高精度的社区检测不仅能够加强对于网络拓扑结构特性的理解，而且能够深入挖掘网络的功能及动态特性。真实复杂网络环境中高精度的社区检测，是目前复杂网络领域亟待解决的研究热点和难点。本课题将高精度的社区检测转化为高维复杂多目标优化问题进行求解，一方面通过对高维多目标优化算法中的核心关键技术进行改进，提升算法收敛精度、解集分布性、收敛速度及稳定性；另一方面，将提出的高性能群智能改进算法及高维多目标优化算法，应用于离散、模糊重叠和动态网络社区检测，提升社区检测的精确性、稳定性及用户满意度。本项目研究中首先我们提出了高性能的群智能优化算法，约束及动态高维多目标优化算法。其次，发现了影响模块度优化社区检测性能的两项关键因素，有效提升了离散社区硬划分检测的精确性和稳定性。再次，有效克服了社区结构显著性和重叠性之间的冲突，同时实现模糊社区划分的两种形式(即离散硬划分和模糊重叠划分)，提升了模糊社区划分的精确性和稳定性。最后，成功将零模型应用到社区检测研究中，不仅丰富了现有社区检测测试网络的数量、种类和功能多样性，增强了对于社区检测算法性能的评估能力，而且加强了对网络多尺度结构特性的分析能力。本项目具有广阔的实际应用前景和重要的理论研究价值，不仅能精确挖掘出复杂网络社区结构中潜在的深层次信息，最大程度满足用户多样化检测需求，又能促进复杂网络与群智能优化、高维多目标优化等理论的发展与融合。