摘要
传统的网络安全态势预测方法依赖于历史态势值的准确性,各种网络安全因素之间存在相关性和重要性差异。针对上述问题,提出了一种基于注意力机制的奇异谱分析(SSA)和时域卷积网络(TCN)与门控循环单元(GRU)的组合预测模型。该模型通过对网络安全态势数据进行奇异谱分析,分解并重构为一系列子序列;对每个子序列建立TCN-GRU神经网络的预测模型,并引入注意力机制动态调整属性的权值;将子序列的预测结果进行叠加,得到最终的预测值。实验结果表明,所提预测方法的拟合度为0.997,其拟合效果和收敛速度均优于其他模型。
Traditional methods of network security situation prediction rely on the accuracy of historical situation values,and there are correlations and differences between various network security factors.To solve the above problems,this paper proposes a combined prediction model of Singular Spectrum Analysis(SSA),Temporal Convolution Network(TCN)and Gated Recurrent Unit(GRU)based on attention mechanism.The model decomposes and reconstructs the data into a series of sub-sequences by SSA.The prediction model of the TCN-GRU neural network for each sub-sequence is established,and the attention mechanism is added to dynamically adjust the weights of attributes.The prediction results of the sub-sequences are superimposed to obtain the final prediction value.The results of the experiments show that the fitting degree of the proposed prediction method is 0.997,and its fitting effect and convergence speed are better than other models.
作者
李成海
孙隽丰
LI Chenghai;SUN Junfeng(Air Defense and Antimissile Academy,Air Force Engineering University,Xi’an 710051,China;Unit 94994 of the People’s Liberation Army,Nanjing 210000,China)
出处
《兵器装备工程学报》
CAS
CSCD
北大核心
2022年第S02期178-185,221,共9页
Journal of Ordnance Equipment Engineering
基金
国家自然科学基金项目(62002362,61703426)
陕西省高校科协青年人才托举计划项目(2019038)
中国陕西省创新能力支持计划项目(2020KJXX-065)。
关键词
网络安全态势预测
奇异谱分析
时域卷积网络
门控循环单元
注意力机制
network security situation prediction
singular spectrum analysis
temporal convolution network
gate recurrent unit
attention mechanism
