全文下载:
202602020.pdf
文章编号:1672-6987(2026)02-0139-09;DOI:10. 16351/j. 1672-6987. 2026. 02. 020
张俊军 a ,徐倩文 b ,顾成杰 a (安徽理工大学 a. 公共安全与应急管理学院; b. 计算机科学与工程学院,安徽 合肥 231131)
摘 要:贝叶斯网络作为人工智能领域处理不确定性问题的关键工具,其结构学习在高维数 据下属于 NP 难问题,传统群智能方法常因初始种群随机、搜索单一而陷入局部最优,精度有 限。为此,本文提出一种混合学习方法 MMFA,融合最大最小爬山算法(MMHC)与改进萤火 虫策略,构建分阶段优化框架。方法采用 MMHC 筛选候选边并确定局部结构,压缩搜索空间 并提升初始解质量;进而设计自适应萤火虫机制,在全局解空间高效探索,避免早熟收敛;每 代进化后嵌入局部评分搜索,以强化对优质解的精细调整。在多组标准数据集上的实验表 明,MMFA 在结构复原精度上显著优于对比方法,同时保持了较快的收敛速度与良好鲁棒性, 为复杂不确定场景下的贝叶斯网络建模提供了可靠解决方案。
关键词:萤火虫算法; MMHC 算法;贝叶斯网络;结构学习
中图分类号:TP 18 文献标志码:A
引用格式:张俊军,徐倩文,顾成杰 . 改进萤火虫优化算法的贝叶斯网络结构学习[J]. 青岛 科技大学学报(自然科学版),2026,47(2):139-147.
ZHANG Junjun, XU Qianwen, GU Chengjie. Bayesian network structure learning based on an improved firefly algorithm[J]. Journal of Qingdao University of Science and Technology (Natural Science Edition),2026,47(2):139-147.
Bayesian Network Structure Learning Based on an Improved Firefly Algorithm
ZHANG Junjuna ,XU Qianwenb ,GU Chengjiea (a. School of Public Safety and Emergency Management;b. School of Computer Science and Engineering, Anhui University of Science and Technology, Hefei 231131,China)
Abstract:Bayesian networks are pivotal tools for handling uncertainty in AI,yet their struc⁃ ture learning poses an NP-hard challenge in high-dimensional contexts. Conventional swarm intelligence methods often trap in local optima due to random initialization and limited search diversity. This paper proposes MMFA, a hybrid method that systematically combines MaxMin Hill-Climbing (MMHC) with an enhanced firefly algorithm. Our approach applies MMHC to prune edges and initialize promising structures, then employs adaptive firefly opera⁃ tors for global exploration, and finally refines solutions via local scoring. Experiments on benchmark datasets verify that MMFA achieves higher reconstruction accuracy and faster con⁃ vergence while maintaining robustness, offering an effective solution for Bayesian network learning under uncertainty.
Key words:firefly algorithm; MMHC algorithm; Bayesian network; structure learning
收稿日期:2025-06-19
基金项目:国家重点研发计划项目(2022YFB2901305);安徽理工大学高层次引进人才科研启动基金项目(2023yjrc64).
作者简介:张俊军(1978—),男,副研究员 .