朱文,胡亚平,聂涌泉,江伟,谢虎.基于深度机器学习的电网虚假数据入侵检测方法研究[J].电测与仪表,2025,62(6):126-133. ZHU Wen,HU Yaping,NIE Yongquan,JIANG Wei,XiIE Hu.False data intrusion detection method for power grid based on deep machine learning[J].Electrical Measurement & Instrumentation,2025,62(6):126-133.
基于深度机器学习的电网虚假数据入侵检测方法研究
False data intrusion detection method for power grid based on deep machine learning
综合能源为主体的新型电力系统运行过程中，容易被虚假数据入侵，且识别虚假数据入侵时易受数据噪声干扰。为了提升其电能质量与运行稳定性，提出了基于深度机器学习的电网虚假数据入侵检测方法。对新型电网数据实施去噪的预处理，利用相量测量单元（phasor measurement unit，PMU）预测出新型电力系统等综合能源的实时系统状态。通过在PMU中不断添加错误测量向量得出虚假数据注入攻击（false data injection attacks，FDIAs），判断电网是否已经被虚假信息攻击以及预测可能攻击的位置值。利用基于小波去噪的BP(back propagation)神经网络对预测结果训练，利用其中的输入层，隐含层以及输出层实时更新出实际值，与阈值比较得出偏差结果，即可检测出电网存在的虚假数据。实验结果表明，所提方法能够提前有效去除噪声，提高了电网虚假数据入侵检测精度高、且检测所需时间较短。
英文摘要:
During the operation of the novel power system with integrated energy as the main body, it is easy to be invaded by false data, and it is easy to be disturbed by data noise when identifying false data intrusion. In order to improve its power quality and operation stability, a false data intrusion detection method for power grid based on deep machine learning is proposed. The novel power grid data is preprocessed by de-noising, and the phasor measurement unit (PMU) is used to predict the real-time system state of comprehensive energy such as the novel power system. False data injection attacks (FDIAs) are obtained by continuously adding error measurement vectors in PMU to judge whether the power grid has been attacked by false information and predict the location value of possible attacks. Back propagation (BP) neural network based on wavelet de-noising is used to train the prediction results, the input layer, hidden layer and output layer are used to update the actual value in real time, and the deviation result is obtained by comparing with the threshold, which can detect the false data in power grid. Experimental results show that the proposed method can effectively remove noise in advance, improve the accuracy of false data intrusion detection in power grid, and the detection time is short.
