基于图卷积神经网络的区域制空权智能评估方法

doi:10.16182/j.issn1004731x.joss.25-0936

摘要/Abstract

摘要：

针对现代空战制空权评估中量化标准缺失与空战样本稀疏的双重难题，提出了一种基于图卷积神经网络的区域制空权智能评估方法。采用空间网格方法将整个战场空域的空中态势编码为一种新的数据表示—制空权矩阵。运用一种新型制空权量化模型，对空间网格的制空权标签值进行量化，选取与制空权争夺紧密相关的4种关键作战能力作为空间网格节点的属性特征。基于图神经网络的半监督学习能力，设计了融合标签传播算法(label propagation algorithm， LPA)的图卷积神经网络模型GCN-LPA，利用LPA生成高质量伪标签，与原始特征拼接后输入GCN网络，形成特征增强机制。采用K折交叉验证方法，提升模型的泛化能力和训练效果。GCN-LPA模型可实时生成全域战场制空权分布热力图，为指挥决策提供精准态势支持。实验结果表明：该方法在制空权数据集上显著优于基线模型，且基于LPA伪标签的特征增强策略其效果也显著优于在GCN损失函数中引入LPA约束项的策略。

关键词: 制空权评估, 态势认知, 指挥控制, 图卷积神经网络, 标签传播

Abstract:

To address the dual challenges of the lack of quantification standards and sparse air combat samples in the superiority evaluation of modern air combat, an intelligent evaluation method for regional air superiority based on a graph convolutional neural network was proposed. The spatial grid method was employed to encode the air situation of the entire battlefield airspace into a new data representation, namely the air superiority matrix. A novel air superiority quantification model was applied to quantify the air superiority label values of spatial grids, and four key operational capabilities closely related to air superiority contention were selected as the attribute features of spatial grid nodes. Based on the semi-supervised learning capability of a graph neural network, a graph convolutional neural network model fused with the label propagation algorithm (LPA), named GCN-LPA, was designed. LPA was utilized to generate high-quality pseudo-labels, which were concatenated with the original features and then input into the GCN network to form a feature enhancement mechanism. The K-fold cross-validation method was adopted to enhance the generalization ability and training effect of the model. The GCN-LPA model can generate a full-domain battlefield air superiority distribution heatmap in real time, providing precise situational support for command decision-making. The experimental results indicate that this method is significantly superior to the baseline model on the air superiority dataset, and the effect of the feature enhancement strategy based on LPA pseudo-labels is significantly better than that of the strategy of introducing an LPA constraint term into the GCN loss function.

Key words: air superiority evaluation, situation awareness, command and control, graph convolutional neural network, label propagation

中图分类号:

E816

贺鹏,唐宇波,刘经德等 . 基于图卷积神经网络的区域制空权智能评估方法[J]. 系统仿真学报, 2026, 38(8): 2294-2311.

He Peng,Tang Yubo,Liu Jingde,et al . Intelligent Evaluation Method for Regional Air Superiority Based on Graph Convolutional Neural Network[J]. Journal of System Simulation, 2026, 38(8): 2294-2311.

图/表 22

图1

图2

图3

表1

表2

图4

图5

图6

图7

图8

表3

表4

表5

图9

图10

图11

图12

表6

表7

图13

图14

图15

参考文献 31

[1]	Endsley M R. Design and Evaluation for Situation Awareness Enhancement[J]. Proceedings of the Human Factors and Ergonomics Society Annual Meeting, 1988, 32(2): 97-101.
[2]	Air Force U.S.. Counterair Operations. Air Force Doctrine Publication 3-01[Z]. Washington: U.S. Air Force.
[3]	朱延广, 张驭龙, 张晓辉, 等. 基于时间影响网络的联合打击方案评估方法[J]. 计算机仿真, 2017, 34(8): 33-38.
	Zhu Yanguang, Zhang Yulong, Zhang Xiaohui, et al. Assessment of Joint Fire Strike Alternatives Based on Timed Influence Net[J]. Computer Simulation, 2017, 34(8): 33-38.
[4]	陈敏, 黄谦, 祁文豪, 等. 面向兵棋推演实验的夺取制权行动效果评估研究[J]. 舰船电子工程, 2024, 44(12): 115-120.
	Chen Min, Huang Qian, Qi Wenhao, et al. Research on the Effectiveness Evaluation of Seizing the Initiative Actions in Wargaming Exercises[J]. Ship Electronic Engineering, 2024, 44(12): 115-120.
[5]	孙巨为, 贾春雨. 陆军进攻战役防空效率指标及制空权评估[J]. 火力与指挥控制, 2008, 33(4): 138-142.
	Sun Juwei, Jia Chunyu. Research on Air-defence Efficiency Indexes and Air Domination Evaluation in Offensive Combat of the Army[J]. Fire Control & Command Control, 2008, 33(4): 138-142.
[6]	韩文生, 邸奇光. 夺取制空权量化分析研究[J]. 军事运筹与系统工程, 2002(4): 47-49.
[7]	Uzair Aslam Bhatti, Tang Hao, Wu Guilu, et al. Deep Learning with Graph Convolutional Networks: An Overview and Latest Applications in Computational Intelligence[J]. International Journal of Intelligent Systems, 2023, 2023(1): 8342104.
[8]	袁立宁, 冯文刚, 刘钊. 基于标签传播增强的多通道图卷积网络[J]. 计算机科学, 2024, 51(8): 304-312.
	Yuan Lining, Feng Wengang, Liu Zhao. Multi-channel Graph Convolutional Networks Enhanced by Label Propagation Algorithm[J]. Computer Science, 2024, 51(8): 304-312.
[9]	Veličković Petar, Cucurull Guillem, Casanova Arantxa, et al. Graph Attention Networks[EB/OL]. (2018-02-04) [2025-07-15]..
[10]	Wu F, Souza A, Zhang Tianyi, et al. Simplifying Graph Convolutional Networks[C]//Proceedings of the 36th International Conference on Machine Learning. Chia Laguna Resort: PMLR, 2019: 6861-6871.
[11]	Zhao Yunxiang, Qi Jianzhong, Liu Qingwei, et al. WGCN: Graph Convolutional Networks with Weighted Structural Features[C]//Proceedings of the 44th International ACM SIGIR Conference on Research and Development in Information Retrieval. New York: Association for Computing Machinery, 2021: 624-633.
[12]	Jin Wei, Derr T, Wang Yiqi, et al. Node Similarity Preserving Graph Convolutional Networks[C]//Proceedings of the 14th ACM International Conference on Web Search and Data Mining. New York: Association for Computing Machinery, 2021: 148-156.
[13]	Guo Kan, Hu Yongli, Sun Yanfeng, et al. Hierarchical Graph Convolution Network for Traffic Forecasting[C]//Proceedings of the Thirty-Fifth AAAI Conference on Artificial Intelligence and the Thirty-Third Conference on Innovative Applications of Artificial Intelligence and the Eleventh Symposium on Educational Advances in Artificial Intelligence. Reston: AIAA, 2021: 151-159.
[14]	Bei Yuanchen, Chen Weizhi, Chen Hao, et al. Correlation-aware Graph Convolutional Networks for Multi-label Node Classification[C]//Proceedings of the 31st ACM SIGKDD Conference on Knowledge Discovery and Data Mining V.1. New York: Association for Computing Machinery, 2025: 37-48.
[15]	You Yuning, Chen Tianlong, Sui Yongduo, et al. Graph Contrastive Learning with Augmentations[C]//Proceedings of the 34th International Conference on Neural Information Processing Systems. Red Hook: Curran Associates Inc., 2020: 5812-5823.
[16]	Zhu Yanqiao, Xu Yichen, Yu Feng, et al. Graph Contrastive Learning with Adaptive Augmentation[C]//Proceedings of the Web Conference 2021. New York: Association for Computing Machinery, 2021: 2069-2080.
[17]	Hassani Kaveh, Amir Hosein Khasahmadi. Contrastive Multi-view Representation Learning on Graphs[C]//Proceedings of the 37th International Conference on Machine Learning. Reston: AIAA, 2020: 4116-4126.
[18]	Raghavan U N, Albert Réka, Kumara S. Near Linear Time Algorithm to Detect Community Structures in Large-scale Networks[J]. Physical Review E, 2007, 76(3): 036106.
[19]	Jia Junteng, Benson A R. A Unifying Generative Model for Graph Learning Algorithms: Label Propagation, Graph Convolutions, and Combinations[J]. SIAM Journal on Mathematics of Data Science, 2022, 4(1): 100-125.
[20]	Wang Hongwei, Leskovec J. Combining Graph Convolutional Neural Networks and Label Propagation[J]. ACM Transactions on Information Systems, 2022, 40(4): 73.
[21]	Ghayekhloo Mohadeseh, Nickabadi Ahmad. CLP-GCN: Confidence and Label Propagation Applied to Graph Convolutional Networks[J]. Applied Soft Computing, 2023, 132: 109850.
[22]	Li Liang, Zhou Jianye, Jiang Yuewen, et al. Propagation Source Identification of Infectious Diseases with Graph Convolutional Networks[J]. Journal of Biomedical Informatics, 2021, 116: 103720.
[23]	毛翔, 褚睿, 邢蓬宇. 美军作战评估理论与实践[M]. 北京: 知识产权出版社, 2017.
	Mao Xiang, Chu Rui, Xing Pengyu. US Armed Forces Operational Assessment Theory and Practice[M]. Beijing: Intellectual Property Publishing House, 2017.
[24]	袁艺, 张玉军, 李志飞. 把握即时优势制胜的要义[EB/OL]. (2022-10-28) [2025-08-08]. .
[25]	梁荣晓, 司光亚, 王艳正, 等. 基于作战势和优势窗口的网络空间作战效能评估方法[J]. 指挥与控制学报, 2023, 9(6): 742-752.
	Liang Rongxiao, Si Guangya, Wang Yanzheng, et al. Evaluation Method of Combat Effectiveness in Cyberspace Based on Combat Potential and Superiority Window[J]. Journal of Command and Control, 2023, 9(6): 742-752.
[26]	中国人民解放军军事科学院国防科技创新研究院. 可计算即时优势窗口形式化表征方法及装置: CN202311239400.3[P]. 2023-10-31.
[27]	中国人民解放军军事科学院国防科技创新研究院. 基于人工经验的即时优势窗口挖掘方法及装置: CN202311239334.X[P]. 2023-10-31.
[28]	Hu Yang, You Haoxuan, Wang Zhecan, et al. Graph-MLP: Node Classification without Message Passing in Graph[EB/OL]. (2021-06-08) [2025-08-16]. .
[29]	Hamilton W L, Ying R, Leskovec J. Inductive Representation Learning on Large Graphs[C]//Proceedings of the 31st International Conference on Neural Information Processing Systems. Red Hook: Curran Associates Inc., 2017: 1025-1035.
[30]	Zhou Dengyong, Bousquet Olivier, Thomas Navin Lal, et al. Learning with Local and Global Consistency[C]//Proceedings of the 17th International Conference on Neural Information Processing Systems. Cambridge: MIT Press, 2003: 321-328.
[31]	Jiang Bo, Zhang Ziyan, Lin Doudou, et al. Semi-supervised Learning with Graph Learning-convolutional Networks[C]//2019 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). Piscataway: IEEE, 2019: 11305-11312.

实验分组	编号	模型名称	LPA特征	LPA损失
对照组	A0	GCN	×	×
对照组	A1	GCN-LPA-LOSS	×	√
实验组	A2	GCN-LPA	√	×
实验组	A3	GCN-LPA-FULL	√	√

参数名称	参数值	参数名称	参数值
隐藏层维度	256	Dropout率	0.100 4
输出层维度	16	学习率	0.004 2
神经网络层数	3	权重衰减系数	1.655 6×10^-5
批量大小	64	训练周期	100

编号	E_MS	E_MA	R²
A0	0.154 7/0.142 3	0.169 4/0.155 1	0.601 6/0.642 5
A1	0.132 4/0.141 6	0.143 1/0.144 6	0.624 5/0.644 5
A2	0.044 3/0.041 9	0.061 2/0.061 0	0.881 5/0.894 8
A3	0.042 6/0.042 0	0.051 5/0.064 9	0.885 7/0.894 5

类别	模型名称	E_MS	E_MA	R²
ML	LPA	0.795 0/0.796 3	0.888 3/0.884 9	-2.249 2/-2.402 1
	Ridge回归	0.011 5/0.017 6	0.021 9/0.029 0	0.094 8/0.101 6
	KNN	0.011 6/0.017 2	0.022 0/0.034 1	0.102 7/0.122 6
	随机森林回归	0.011 1/0.017 2	0.021 2/0.030 5	0.109 9/0.131 6
	贝叶斯回归	0.011 5/0.017 6	0.021 8/0.029 1	0.094 7/0.101 6
MLP	MLP	0.011 5/0.017 7	0.020 8/0.027 1	0.094 3/0.098 8
MLP	Graph-MLP	0.011 5/0.017 7	0.020 8/0.027 1	0.094 3/0.098 8
GNN	GAT	0.183 2/0.161 5	0.222 6/0.182 2	0.489 2/0.549 3
	SGC	0.132 9/0.140 6	0.144 9/0.141 0	0.623 1/0.646 7
	GraphSAGE	0.146 4/0.142 7	0.148 8/0.143 9	0.605 9/0.641 7
	GCN-LPA	0.036 8/0.042 1	0.048 7/0.050 8	0.895 7/0.894 1

参数类型	参数名称	优化区间
离散参数	隐藏层维度	{32,64,128,256}
	输出层维度	{16,32,64}
	批量大小	{4,8,16,32,64}
	神经网络层数	{2,3,4,5,6}
连续参数	学习率	[1×10^-5,1×10^-2]
	Dropout率	[0.1,0.7]
	权重衰减系数	[1×10^-6,1×10^-3]
	训练周期	[50,200]