文献类型：期刊文献

中文题名：面向复杂拓扑大规模重叠问题的分组方法研究

英文题名：Research on grouping methods for large-scale overlapping problemsof complex topologies

作者：梁辰[1];田茂江[1];陈鸣科[1]

机构：[1]华东理工大学信息科学与工程学院,上海200237

年份：2025

卷号：46

期号：5

起止页码：41

中文期刊名：激光杂志

外文期刊名：Laser Journal

收录：;北大核心:【北大核心2023】；

基金：国家自然科学基金面上项目(No.62173144);上海市自然科学基金面上项目(No.21ZR1416100)。

语种：中文

中文关键词：大规模重叠问题;复杂拓扑;差分分组;计算资源消耗;变量交互性

外文关键词：large-scale overlapping problem;complex topology;differential grouping;computational resource consumption;variable interaction

摘要：大规模重叠问题广泛存在于实际应用中,如激光雷达设计优化问题,其中各系统之间的复杂耦合可视为大规模重叠问题。这类问题通常具有稀疏交互高维度特征,现有的分组方法难以有效识别其底层结构。此外,实际问题中的重叠问题拓扑结构复杂,而现有研究主要关注简单的链式拓扑。为应对这些挑战,扩展了大规模重叠问题的拓扑结构,提出了一种构造复杂拓扑重叠问题的新方法,并设计了一种基于递归分解的差分分组方法(OERDG)。OERDG在保持问题完整结构的前提下,以较低的计算复杂度实现了对复杂拓扑重叠问题的高效准确分解。实验结果表明,OERDG在以100%准确率识别问题结构的前提下,消耗的计算资源仅为传统方法的5%,高效准确地实现了大规模重叠问题底层交互结构的识别。
Large-scale overlapping problems widely exist in practical applications,such as the LIDAR design optimization problem,in which the complex coupling between systems can be regarded as a large-scale overlapping problem.These problems are usually characterized by sparse interaction and high dimensionality,and existing grouping methods are difficult to effectively identify their underlying structures.In addition,the topology of overlapping problems in real problems is complex,while existing studies mainly focus on simple line topology.To cope with these challenges,this paper extends the topology of large-scale overlapping problems,proposes a new method for constructing complex topology overlapping problems,and designs a recursive decomposition-based differential grouping method(OERDG).OERDG realizes efficient and accurate decomposition of complex topology overlapping problems with low computational complexity while maintaining the complete structure of the problem.Experimental results show that OERDG efficiently and accurately realizes the identification of the underlying interaction structure of large-scale overlapping problems while recognizing the problem structure with 100%accuracy and consuming only 5%of the computational resources of traditional methods.