An Intrusion Detection Algorithm For Power Grid Wireless Equipment Based On Radio Frequency Fingerprint

In the booming development of ubiquitous power IoT,with the popularity of wireless terminal devices such as smart meters,the number of wireless terminal devices to be managed by ubiquitous power IoT is increasing.The new business models and complex application scenarios have put forward higher requirements on the intrusion detection of wireless devices terminals in power grids.The previous network layer encryption means can no longer meet the requirements of ubiquitous power IoT applications.Radio frequency(RF)fingerprinting uses the hardware differences inherent in the RF of the terminal device for identification,and the feature cannot be replicated with any other device.Therefore,the physical layer-based,non-replicable alternative RF fingerprinting technology has received a lot of attention.RF fingerprint features of wireless device terminals can be divided into transient signal-based features and steady-state signal-based features.The transient signal interception has high requirements for the sampling rate of the device,once the transient signal interception is not appropriate,the accuracy of RF fingerprint classification recognition slips very obviously.In comparison,the steady-state signal is easier to be detected and extracted,and the steady-state signal contains a large amount of hardware feature information,and the signal is more complete.Therefore,the steady-state signal is chosen to extract the RF fingerprint features of wireless device terminals is more suitable for ubiquitous power IoT application scenarios.This paper mainly discusses the method of extracting RF fingerprint features based on steady-state signals to classify and identify wireless device terminals.The main research results are as follows:This paper investigates and analyzes three important aspects of RF fingerprint classification and recognition technology:signal extraction,RF fingerprint feature selection and extraction,and classification and recognition algorithm.This paper gives a detailed description of the development status of these three links and the mechanism of RF fingerprint feature generation.By analyzing the higher order statistics of the signal,it is concluded that the signal higher order spectral features can suppress Gaussian white noise.Then the properties and specific estimation methods of signal bispectrum are further given,and the characteristics,properties and correlations among the five locally integrated bispectrums are compared one by one,while their projection maps in the class-optimal discriminative subspace are compared.Based on this,an RF fingerprint identification algorithm based on rectangular integral bispectrum and selected bispectrum fusion features is proposed in this paper.Finally,the effectiveness and applicability of the proposed RF fingerprint recognition algorithm based on rectangular integral bispectrum and selected bispectrum fusion features for wireless device terminal identification are demonstrated experimentally,especially in the low signal-to-noise ratio(SNR,Signal to Noise Ratio)environment still maintains a good classification performance,and it is also proved by comparison experiments that the proposed RF.It is also demonstrated through comparative experiments that the proposed RF fingerprint recognition algorithm is generalized to GFSK,OFDM,QPSK,FSK signals,and in the comparison experiments with deep learning models,it also shows the superiority of this scheme in the application scenarios of ubiquitous power IoT.In addition,this paper also proposes an improved scheme for RF fingerprint identification based on federal learning,and it is proved through experiments that this scheme can effectively reduce the model training time.In this paper,we use Tektronix RSA6114 as the receiver device and nRF24LE1 as the transmitter terminal,and extract RF fingerprint features by processing the actual received signals,and also use support vector machine classifier for training and testing,the accuracy of the proposed scheme based on rectangular integral bispectrum and selected bispectrum fusion features as RF fingerprint features is The accuracy of the proposed scheme based on rectangular integral bispectrum and selected bispectrum fusion features as RF fingerprint features is significantly higher than other traditional schemes,which verifies the effectiveness and reliability of the algorithm.