| 引用本文： |
-
陈雨,徐艳云,陈泓钢,张萌,黄伟庆.基于GAT-Transformer的频谱预测和异常检测[J].信息安全学报,已采用 [点击复制]
- chenyu,xuyanyun,chenhonggang,zhangmeng,huangweiqing.Spectrum prediction and anomaly detection based on GAT-Transformer[J].Journal of Cyber Security,Accept [点击复制]
|
|
| 摘要: |
| 随着无线通信技术和物联网的迅速发展，无线信号不断增加使得电磁频谱日益复杂，网络边界与电磁空间逐步融合，以电磁信号为途径的跨网攻击利用正常无线信道实施信息窃密，呈现体系性、隐蔽性和伪装性的特点。传统的信号分析方法很难检测，需要利用长时、海量电磁数据进行规律行为建模、预测和态势感知，进而发现异常。为此，本文提出一种新型的基于GAT-Transformer的频谱预测方法，该方法可实现未来时刻频谱的准确预测，基于预测结果可实现异常电磁频谱的检测识别。该方法结合了图注意力网络（Graph Attention Network, GAT）和Transformer模型的优势，充分考虑了频谱数据在时域上的长时特性和频域上的相关性。通过提取电磁频谱数据的时间占用度、时频相关性等特征，检测具有规律性用频的频段，对该频段频谱数据的使用规律进行深入建模，准确捕捉时频关系，进而预测未来时刻信道的功率值，以实现对频谱变化的高效预测。为验证算法的有效性，本文在公开的真实电磁频谱数据集上进行验证，并与SVR、RNN、LSTM、ConvLSTM、Transformer等基准模型进行了比较，结果表明，基于GAT-Transformer的频谱预测模型在预测准确度方面表现出色。此外，通过多步长预测和消融分析等方面的性能测试，也验证了该方法的稳健性和可靠性。最后本文在频谱预测模型的基础上对频谱异常检测进行了研究，通过在实测频谱数据集的基础上构建异常频谱数据，基于预测模型预测未来的电磁空间安全态势，能够精准地检测出破坏电磁频谱数据用频规律的异常行为。本文首次将GAT和Transformer模型结合应用到频谱预测领域，利用预测值和真实值的差异鉴别异常。该方法无需给所有数据加上标签，为电磁频谱异常检测提供了一种高效可行的解决方案。 |
| 关键词: 频谱预测 异常检测 GAT网络 Transformer模型 电磁安全 |
| DOI： |
| 投稿时间：2024-01-12修订日期：2024-03-29 |
| 基金项目:中国科学院重点资助项目 |
|
| Spectrum prediction and anomaly detection based on GAT-Transformer |
|
chenyu, xuyanyun, chenhonggang, zhangmeng, huangweiqing
|
| (Institute of Information Engineering, Chinese Academy of Sciences) |
| Abstract: |
| With the rapid development of wireless communication technology and the Internet of Things, the continuous increase of wireless signals has made the electromagnetic spectrum increasingly complex. The network boundary has merged with the electromagnetic space. Cross-network attacks using electromagnetic signals use normal wireless channels to carry out information theft, showing systematic, concealed and camouflaged. Traditional signal analysis methods are difficult to detect. It requires the use of long-term and massive electromagnetic data for modeling the regular behavior, prediction and situational awareness to detect anomalies. To this end, this paper proposes a new spectrum prediction method based on GAT-Transformer. This method can achieve accurate prediction of spectrum at future times, and based on the prediction results, it can detect and identify abnormal electromagnetic spectrum. This method combines the advantages of the Graph Attention Network (GAT) and the Transformer model, and fully considers the long-term characteristics of spectrum data in the time domain and the correlation in the frequency domain. By extracting features such as time occupancy and time-frequency correlation of electromagnetic spectrum data, we can detect frequency bands with regular frequency usage, conduct in-depth modeling of the usage patterns of spectrum data in this frequency band, accurately capture time-frequency relationships, and then predict the power value of the channel at future moments to achieve efficient prediction of spectrum changes. In order to verify the effectiveness of the algorithm, this paper conducts verification on the public real electromagnetic spectrum data set and compares it with benchmark models such as SVR, RNN, LSTM, ConvLSTM, and Transformer. The results show that the spectrum prediction model based on GAT-Transformer performs well in terms of prediction accuracy. In addition, the robustness and reliability of the method were also verified through performance tests in multi-step prediction and ablation analysis. Finally, this paper studies spectrum anomaly detection based on the spectrum prediction model. By constructing abnormal spectrum data based on the measured spectrum data set, and predicting the future electromagnetic space security situation based on the prediction model, this method can accurately detect abnormal behaviors that disrupt the frequency pattern of electromagnetic spectrum data. This paper combines GAT and Transformer models for the first time in the field of spectrum prediction, using the difference between predicted values and real values to identify anomalies. This method does not require labeling all data, and provides an efficient and feasible solution for electromagnetic spectrum anomaly detection. |
| Key words: spectrum prediction anomaly detection Graph Attention Network Transformer electromagnetic safety |
