电力时间同步系统防欺骗抗干扰检测技术及方法

Anti-Spoofing and Anti-Interference Detection Technologies and Methods for Power Time Synchronization Systems

摘要:

为解决电力时间同步系统面临的复杂动态干扰与隐蔽欺骗威胁,弥补传统防欺骗抗干扰技术适配性不足、缺乏综合检测的短板,对适用于电力场景的抗干扰与防欺骗检测技术进行了研究。采用基于盲信号分离与时空特征聚合网络的抗干扰识别算法(BSS-TSFANet),与多特征融合CNN-LSTM欺骗检测算法。BSS-TSFANet通过时频域虚拟多通道扩展与改进FastICA实现复合干扰的盲分离,再经轻量化抗噪型TSFANet提取时空特征,实现高精度干扰识别；CNN-LSTM算法经STFT时频转换、PCA降维后,通过轻量化CNN与LSTM联合提取特征实现高精度欺骗检测。经试验测试,BSS-TSFANet在干噪比-20dB环境下对单一干扰平均识别准确率达96.78%,对典型电力复合干扰的成分识别召回率达93.78%；CNN-LSTM算法对转发式与生成式欺骗的检测准确率分别达98.75%和99.95%。两类算法均满足电力终端轻量化、实时性要求,为智能电网电力时间同步系统提供了一体化抗干扰与防欺骗检测方案。

Abstract:

To address the complex dynamic interference and covert spoofing threats faced by power time synchronization systems, and to make up for the shortcomings of traditional anti-spoofing and anti-interference technologies such as poor adaptability and the lack of comprehensive detection capabilities, research was conducted on anti-interference and anti-spoofing detection technologies suitable for power scenarios. A blind signal separation and temporal-spatial feature aggregation network-based anti-interference recognition algorithm (BSS-TSFANet) and a multi-feature fusion CNN-LSTM spoofing detection algorithm were adopted in this study. BSS-TSFANet achieves blind separation of composite interference via time-frequency domain virtual multi-channel expansion and an improved FastICA algorithm, and then extracts temporal and spatial features through a lightweight anti-noise TSFANet to realize high-precision interference recognition. For the CNN-LSTM algorithm, after short-time Fourier transform (STFT) for time-frequency conversion and principal component analysis (PCA) for dimensionality reduction, lightweight CNN and LSTM are combined to extract features for high-precision spoofing detection. Experimental tests show that BSS-TSFANet achieves an average recognition accuracy of 96.78% for single interference and a component recognition recall of 93.78% for typical power composite interference in an environment with an interference-to-signal ratio -20 dB. The CNN-LSTM algorithm reaches detection accuracies of 98.75% and 99.95% for replay spoofing and generative spoofing, respectively. Both algorithms meet the lightweight and real-time requirements of power terminals, providing an integrated anti-interference and anti-spoofing detection solution for power time synchronization systems in smart grids.