Train satellite positioning unsupervised deception detection method based on zero deception sample

Abstract

The invention provides a train satellite positioning unsupervised deception detection method based on a zero deception sample. The method comprises the following steps: extracting satellite signal related statistical features, constructing a satellite signal feature sequence diagram by taking time as a dimension, and generating a label-free zero-cheating training feature sample set, a label-free zero-cheating verification feature sample set and a test feature sample set by utilizing the satellite signal feature sequence diagram; training the diffusion reconstruction network by using the label-free'zero-cheating 'training feature sample set, and verifying the diffusion reconstruction network by using the verification feature sample set and the test feature sample set to obtain a trained diffusion reconstruction network; inputting the signal characteristic time sequence diagram of the train in the in-transit operation process into the diffusion reconstruction network, calculating a reconstruction error index value, and judging whether the train is subjected to spoofing attack or not based on the reconstruction error index value. According to the invention, real-time detection of deception signals in the train operation process is realized, and effective support is provided for interference protection of a train positioning system.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of train satellite positioning technology, and in particular to a train satellite positioning unsupervised spoofing detection method based on zero spoofing samples. BACKGROUND

[0002] According to the process from satellite signal generation to navigation positioning, spoofing detection techniques can be divided into spoofing detection based on navigation data information, spoofing detection based on spatial processing, spoofing detection based on radio frequency front-end, spoofing detection based on baseband digital signal processing, spoofing detection based on positioning and navigation results, and spoofing detection based on machine learning. The spoofing detection based on navigation data information mainly prevents spoofing attacks from tampering with navigation information by encrypting and authenticating navigation signals from the source; the spoofing detection based on spatial processing mainly detects the spoofing by detecting the angle of arrival and direction of arrival; the spoofing detection based on baseband digital signal mainly identifies and realizes spoofing detection by monitoring the received signal strength, signal quality, Doppler shift consistency, and characteristic differences of signal arrival time; the spoofing detection based on positioning and navigation results usually needs to introduce auxiliary sensors that are not affected by electromagnetic interference, and compares the measured data information with the results calculated by the receiver terminal to realize spoofing detection; the spoofing detection based on machine learning algorithm mainly distinguishes spoofing signals from real signals by learning the differences between input features, and realizes spoofing interference detection.

[0003] The prior art has a scheme that proposes a sensor fusion method based on a particle filter for satellite positioning spoofing interference detection. This method enhances the robustness of the fusion of satellite navigation systems and inertial navigation systems by using the output data of the IMU (Inertial Measurement Unit), vehicle route information, and road features such as lanes and edges. This method detects abnormal changes in pseudo-range measurements under spoofing attack conditions.

[0004] The prior art also has a scheme that proposes a multi-parameter spoofing interference signal detection algorithm based on a gated mechanism parallel CNN-Transformer neural network. This algorithm considers the influence of spoofing interference on each parameter of the receiver in the tracking stage, and extracts multiple feature information to form a multi-dimensional time sequence. After preprocessing the sequence, the PCTN network is used to extract the time before and after the sequence and the feature information in multiple dimensions, and the feature information is sent into the network for processing to realize spoofing interference detection. Experimental results show that this method can effectively improve the universality and generalization of the spoofing interference detection algorithm in multiple scenarios.

[0005] The spoofing interference detection methods in the above prior art have the following disadvantages:

[0006] Existing deception jamming detection methods usually need to introduce additional hardware devices, such as high-precision sensors or additional antennas, to achieve a relatively ideal detection effect. This not only increases the implementation complexity of the system, but also brings additional costs, so that these technologies are difficult to be widely applied in the railway field which requires high security.

[0007] In existing deep learning-based deception jamming detection methods, most methods rely on a large number of labeled deception jamming samples for training. However, in the railway field, real deception jamming data is relatively scarce, which leads to problems such as insufficient generalization ability, poor reliability, and low accuracy when these methods are applied to the railway scene, making it difficult to meet the actual needs of train positioning. SUMMARY

[0008] The present application provides a train satellite positioning unsupervised deception detection method based on zero deception samples to effectively improve the credibility of satellite positioning such as Beidou.

[0009] In order to achieve the above-mentioned purpose, the present application adopts the following technical solutions.

[0010] A train Beidou positioning unsupervised deception detection method based on zero deception samples, comprising:

[0011] Collecting digital intermediate frequency signals of train satellite positioning, constructing a "zero deception sample" training dataset and a "zero deception sample" validation dataset that do not contain any deception jamming, and generating a test feature dataset containing deception jamming by injecting deception signals;

[0012] Extracting satellite signal related statistical features, constructing a satellite signal feature time sequence graph in time dimension, and generating a label-free "zero deception" training feature sample set, a label-free "zero deception" validation feature sample set, and a test feature sample set using the satellite signal feature time sequence graph;

[0013] Constructing a diffusion reconstruction network suitable for unsupervised deception jamming detection tasks in the field of rail transit, training the diffusion reconstruction network using the label-free "zero deception" training feature sample set, validating the diffusion reconstruction network using the validation feature sample set and the test feature sample set, quantifying the reconstruction error of the diffusion reconstruction network using mean square error and structural similarity index, and obtaining a trained diffusion reconstruction network;

[0014] During the in-transit operation of the train, real-time satellite digital intermediate frequency signals are received, satellite signal related statistical features are extracted and a signal feature time sequence graph is generated, the signal feature time sequence graph is input into the trained diffusion reconstruction network, the diffusion reconstruction network outputs a model reconstruction graph, a reconstruction error index value is calculated, and whether the train is subjected to deception attack is determined based on the size of the reconstruction error index value.

[0015] Preferably, the collection of train satellite positioning digital intermediate frequency signal, construct "zero spoof sample" training data set and "zero spoof sample" validation data set which do not contain any spoofing interference, and generate test feature data set containing spoofing interference by injecting spoofing signal, including:

[0016] In the process of train running on a specific line, the positioning data is collected by the train position reference system, the train running trajectory file is generated, the satellite navigation signal simulation and collection environment is built, the train running scenario is reproduced by injecting the train running trajectory file into the simulation system, and the satellite positioning digital intermediate frequency signal is collected for a certain time to generate "zero spoof sample" training data set and "zero spoof sample" validation data set which do not contain any spoofing interference;

[0017] In the process of reproducing the train running scenario, the spoofing interference signal is injected into the "zero spoof sample" training data set and "zero spoof sample" validation data set which do not contain any spoofing interference according to the predetermined time window, the satellite positioning digital intermediate frequency signal of the corresponding period is collected, and the test feature data set containing spoofing interference is generated.

[0018] Preferably, the satellite signal related statistical features are extracted, and the satellite signal feature time sequence diagram is constructed with time as the dimension, and the unlabeled "zero spoof" training feature sample set is generated by using the satellite signal feature time sequence diagram, including:

[0019] The collected satellite positioning digital intermediate frequency signal of the train is sent into the software receiver for processing, and according to the processing flow of the capture and tracking loop, the signal power feature, signal correlation peak feature and signal tracking stability feature of the satellite positioning digital intermediate frequency signal are extracted at the frequency f1:

[0020] (1) The carrier noise power spectral density C / N0 is used as the signal power feature, and the signal power feature reflects the ratio between the demodulation signal power C and the noise power spectral density N0:

[0021]

[0022] Wherein, P C represents the power of the carrier signal, P N is the noise power in the passband;

[0023] (2) The signal correlation peak feature includes Delta index, Ratio index, ELP index and Q channel index;

[0024] The calculation formula of Delta index is:

[0025]

[0026] Wherein, IE (t),I L (t),I P (t) respectively represent the in-phase component output of the early code, late code and prompt code correlators;

[0027] The calculation formula of the Ratio index is:

[0028]

[0029] The calculation formula of the ELP index is:

[0030]

[0031] wherein, Q E (t),Q L (t) respectively represent the in-phase component output of the early code, late code and prompt code correlators;

[0032] The calculation formula of the Q channel index is:

[0033]

[0034] m SQM The index identifies the injection of a spoofing signal by measuring the abnormal energy of the in-phase channel. If the energy of the in-phase channel is abnormally high, it indicates that the signal modulation structure or phase consistency is destroyed;

[0035] On the basis of setting a fixed window length, the sliding mean and sliding variance of different types of characteristic quantities are calculated respectively. The calculation formulae of the sliding mean and sliding variance are:

[0036]

[0037] wherein, μ x (n) represents the mean of the characteristic x at the n th sliding time step, σ Delta (n) represents the variance of the characteristic x at the n th sliding time step, n represents the number of sliding windows, L represents the size of the sliding window, x(i) represents the characteristic value at the i th time, and the means of the Delta index, the Ratio index, the ELP index and the Q channel index at the n th sliding time step are respectively represented as μ Ratio (n), μ ELP (n) and The variances at the n th sliding time step are respectively represented as σ

[0038] (3) The signal tracking stability features include code tracking loops and carrier tracking loops. The output of the code loop discriminator represents the code phase tracking error, which is calculated by the following formula:

[0039]

[0040] where I E and Q L represent the in-phase component outputs of the early and late correlators, respectively. E and Q L represent the quadrature component outputs of the early and late correlators, respectively.

[0041] The output of the carrier loop discriminator represents the carrier phase tracking error, which is calculated by the following formula:

[0042]

[0043] where I P and Q P represent the in-phase and quadrature component outputs of the prompt correlator, respectively.

[0044] The mean values of the code phase tracking error and the carrier phase tracking error at the nth sliding time step are denoted as μ dllDiscr (n) and μ pllDiscr (n), respectively, and the variances are denoted as and The extracted signal power features, signal correlation peak features, and signal tracking stability features together contain 19 specific feature indicators, and the sampling frequency of the instantaneous features is set to be consistent with the sliding window update interval The signal statistical feature vector fv at time step t is denoted as:

[0045]

[0046] The extracted features of various types are aligned according to the time dimension, and a feature time sequence diagram is constructed according to the preset spoofing detection frequency f3. The feature time sequence diagram F T at time T is denoted as:

[0047]

[0048] where the size of the satellite signal feature time sequence diagram is (f2 / f3) x 19, (f2 / f3) represents the number of time points contained in the feature time sequence diagram, and 19 represents the feature dimension. The satellite signal feature time sequence diagram is used to generate the unlabeled "zero spoofing" training feature sample set.

[0049] Preferably, the diffusion reconstruction network suitable for the unsupervised spoofing interference detection task in the field of rail transit is trained using the unlabeled "zero spoofing" training feature sample set, including:

[0050] Based on the time sequence-feature two-dimensional structure of the satellite signal feature time sequence diagram, the noise prediction network in the diffusion model structure is optimized and improved. Residual network ResNet modules are introduced in the encoder and decoder parts of U-Net to replace the convolutional blocks. Channel attention ECA modules are introduced. One-dimensional time convolution down-sampling and up-sampling modules are designed and embedded into the encoder and decoder for time dimension feature learning. The down-sampling module of the encoder in the network is connected to the corresponding up-sampling module in the decoder through a jump connection to construct a diffusion reconstruction network suitable for unsupervised fraud interference detection tasks in the field of rail transit;

[0051] The diffusion reconstruction network is trained using the unlabeled "zero fraud" training feature sample set. The training process includes the following steps:

[0052] (1) Initialize the training data:

[0053] The unlabeled "zero fraud" training feature sample set is standardized;

[0054] (2) Define the key hyperparameters for model training:

[0055] Including the maximum number of time steps T of the diffusion process, the noise scheduling function, the learning rate, the optimizer type, the number of training rounds epochs, and the batch size;

[0056] The noise scheduling function is set to linear scaling, represented as:

[0057]

[0058] Where, represents and T uniform numbers are generated between β t represents the noise intensity at the t-th time step, β start and β end are the starting value and ending value of the noise intensity, respectively;

[0059] (3) The training process of the diffusion reconstruction network is a Markov modeling process aimed at denoising. The training process includes a forward diffusion process and a backward denoising process. After training, the diffusion reconstruction network realizes the reconstruction and generation of the feature time sequence diagram based on the backward sampling strategy:

[0060]

[0061] Where, x t represents the noisy feature time sequence diagram at time step t, ε θ (x t , t) represents the noise component predicted by the model, and σ tx t represents the noise amplitude sampled at the time step t, x t-1 x t-1 represents the feature time series after denoising at the previous time step, the process is iteratively performed in the full time domain of t=T→0, and the "zero deception" feature time series is reconstructed

[0062] Preferably, the reconstruction error of the diffusion reconstruction network is quantified by using the mean square error and the structural similarity index to obtain the trained diffusion reconstruction network, comprising:

[0063] The reconstruction error between the original feature time series x0 and the reconstructed feature time series is calculated, and the reconstruction error includes MSE and SSIM, and the mathematical expression of MSE is:

[0064]

[0065] Wherein, D represents the total number of pixels in the sample, x0 j represents the value of the jth pixel point of the original feature time series, x j represents the value of the jth pixel point of the reconstructed feature time series;

[0066] The mathematical expression of SSIM is:

[0067]

[0068] Wherein, and are the average brightness of images x0 and , respectively, and are the variances of images x0 and , respectively, is the covariance of images x0 and , C1 and C2 are constants for stabilizing calculation, C1=(K1L) 2 and C2=(K2L) 2 , wherein L is the dynamic range of pixel value;

[0069] The reconstruction error of the ith sample in the unlabeled "zero deception" verification feature sample set is represented as The mean μ and the standard deviation σ of the reconstruction error of the entire unlabeled "zero deception" verification feature sample set val and val are:

[0070]

[0071] Wherein, N val represents the number of "zero deception" verification feature samples;

[0072] The calculation process of the reconstruction error of the ith sample in the feature sample set without label "zero fraud" verification feature is shown in the following table The calculation process of the reconstruction error of the ith sample in the feature sample set without label "zero fraud" verification feature is shown in the following table The corresponding Z-score is:

[0073]

[0074] Wherein, Z (j) Reflects the Z-score of the jth test feature sample;

[0075] Taking the Z-score as the abnormal detection score index, combining the real label of the sample in the test feature data set, the ROC curve is constructed, and by traversing different Z-score thresholds, the false positive rate FPR and the true positive rate TPR under the threshold are calculated:

[0076]

[0077] The area AUC value under the ROC curve is obtained by integral method or numerical calculation method:

[0078]

[0079] When the difference between the AUC value and the threshold value 1 is less than the set numerical range, it is judged that the training process of the diffusion reconstruction network is ended, and the trained diffusion reconstruction network is obtained.

[0080] Preferably, during the in-transit operation of the train, satellite digital intermediate frequency signals are received in real time, satellite signal related statistical features are extracted and signal feature time sequence diagrams are generated, the signal feature time sequence diagrams are input into the trained diffusion reconstruction network, the diffusion reconstruction network outputs a model reconstruction diagram, a reconstruction error index value is calculated, and whether the train is attacked by fraud is judged based on the size of the reconstruction error index value, including:

[0081] During the in-transit operation of the train, satellite digital intermediate frequency signals are received in real time, signal related statistical features are extracted and satellite signal feature time sequence diagrams are generated, the trained diffusion model is called and a model reconstruction diagram is obtained, the mean square error, structural similarity and other reconstruction error indexes are calculated, and whether the train is attacked by fraud is judged based on the size of the reconstruction error index value.

[0082] During the in-transit operation of the train, satellite digital intermediate frequency signals are received in real time, satellite positioning digital intermediate frequency signals are sent into a software receiver for processing, signal power features, signal peak value features and signal tracking stability features of the satellite digital intermediate frequency signals are extracted, and a signal feature time sequence diagram F T at time T is obtained, which is represented as:

[0083] The signal feature time sequence diagram F T is input to the trained diffusion reconstruction network, and a reconstruction error index including a mean square error and a structural similarity is calculated. Taking the "zero deception" verification feature sample set as a benchmark, a Z-score index of the current reconstruction error is calculated, and the Z-score value is compared with a preselected threshold τ * . When the Z-score value is greater than the threshold τ * , it is judged that the train is attacked by deception at the current time; and when the Z-score value is not greater than the optimal threshold τ * , it is judged that the train is not attacked by deception at the current time.

[0084] As can be seen from the technical solutions provided by the above embodiments of the present application, the method can learn and reconstruct the feature time sequence diagram of the "zero deception" navigation signal, and produce a differential response to the feature time sequence diagram of the deceptive satellite signal, thereby realizing real-time detection of the deceptive signal during train operation. This method provides effective support for interference protection of the train positioning system, thereby ensuring the credibility of the train Beidou positioning.

[0085] Additional aspects and advantages of the present application will be described in the following description and will be apparent from the description or will be learned. BRIEF DESCRIPTION OF DRAWINGS

[0086] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings needed in the embodiment description will be briefly introduced below. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor.

[0087] Figure 1 A flow chart of a train Beidou positioning unsupervised deception detection method based on zero deception samples is provided for the embodiments of the present application.

[0088] Figure 2 A schematic diagram of data processing using a sliding window mechanism is provided for the embodiments of the present application.

[0089] Figure 3 A model architecture diagram for realizing zero deception sample unsupervised deception interference detection based on a diffusion model is provided for the embodiments of the present application.

[0090] Figure 4 A noise prediction network structure diagram designed for Beidou satellite positioning deception interference detection is provided for the embodiments of the present application.

[0091] Figure 5 An ROC curve diagram of deception interference test results based on different reconstruction error indexes is provided for the embodiments of the present application.

[0092] Figure 6 A comparison chart of test performance of deception jamming based on different reconstruction error indicators is provided for the embodiments of the present application. DETAILED DESCRIPTION

[0093] Embodiments of the present application are described in detail below with reference to several specific embodiments thereof, examples of which are illustrated in the accompanying drawings, wherein like or similar elements are designated with the same or similar reference numerals throughout the several views. The embodiments described below are exemplary only, and are not to be construed as limiting the present application.

[0094] As used herein, the singular forms "a", "an" and "the" are intended to include the plural forms as well, unless the context clearly indicates otherwise. It will be further understood that the terms "comprises" and / or "comprising," when used in this specification, specify the presence of stated features, integers, steps, operations, elements, and / or components, but do not preclude the presence or addition of one or more other features, integers, steps, operations, elements, components, and / or groups thereof. As used herein, "connected" or "coupled" can mean directly connected or coupled or can mean indirectly connected or coupled with intervening elements.

[0095] As used herein, the term "and / or" includes any and all combinations of one or more of the associated listed items. Unless otherwise defined, all terms (including technical and scientific terms) used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs. It will be further understood that terms, such as those defined in commonly used dictionaries, should be interpreted as having a meaning that is consistent with their meaning in the context of the relevant art and will not be interpreted in an idealized or overly formal sense unless expressly so defined herein.

[0096] For the purpose of promoting an understanding of the principles of the present application, reference will now be made to the embodiments illustrated in the drawings, and specific language will be used to describe the same. It will, nevertheless, be understood that no limitation of the scope of the application is intended by illustrative discussion of the embodiments.

[0097] The application provides an application scenario of a train satellite positioning unsupervised spoofing detection method based on zero spoofing samples.

[0098] The processing flow of the train satellite positioning unsupervised spoofing detection method based on zero spoofing samples provided by the embodiment of the application is as shown in Figure 1

[0099] Step S1, collect the digital intermediate frequency signals of train satellite positioning, construct the "zero spoofing sample" training data set and the "zero spoofing sample" verification data set not containing any spoofing interference, and generate the test feature data set for detection and verification by injecting spoofing signals.

[0100] During the operation of the train on the Qinghai-Tibet line (Lhasa-Shigatse), the train positioning data are collected by using the high-precision train position reference system, and the 30-minute train operation trajectory file is continuously generated, serving as the basic data for subsequent experimental test and verification.

[0101] The satellite navigation signal simulation and collection environment is built, the 30-minute train operation trajectory file is injected into the simulation system, the train operation scene is reproduced, the 30-minute digital intermediate frequency signals are collected by using the Beidou satellite navigation system intermediate frequency signal collector, and the "zero spoofing sample" training data set and the "zero spoofing sample" verification data set not containing any spoofing interference are constructed, wherein the collection time length of the training data set is 20 minutes, and the collection time length of the verification data set is 10 minutes.

[0102] ​During the above scenario reproduction process, a spoofing interference test feature dataset is constructed simultaneously. A spoofing interference signal is injected at PRN=30 channel when the scenario runs for 5 minutes, the injection interference type is pseudo-range spoofing, the slope increases at a speed of 1 m / s, the power increases at a rate of 1 dB / s, and the injection duration is 15 minutes. The satellite digital intermediate frequency signal in this period is collected by the intermediate frequency signal collector, the collection duration is 20 minutes (including 5 minutes of non-spoofing data and 15 minutes of spoofing data), and finally a test feature dataset containing spoofing interference is generated, which is used for subsequent model performance test verification.

[0103] Step S2, extract satellite signal related statistical features, construct satellite signal feature time series graph with time as the dimension, and form unlabeled feature sample set corresponding to three different data sets.

[0104] The collected satellite positioning digital intermediate frequency signal of the train is sent to the software receiver for processing, and according to the processing flow of the capture and tracking loop, three types of statistical features are extracted at frequency f1, in this embodiment, f1=1000Hz, specifically:

[0105] (1) Signal power feature

[0106] The carrier-to-noise power spectrum density (C / N0) is used as the signal power feature, which reflects the ratio between the demodulation signal power C and the noise power spectrum density N0, and is usually expressed in logarithmic form, that is:

[0107]

[0108] Where, P C represents the power of the carrier signal, P N is the noise power in the passband.

[0109] Under normal circumstances, the signal power transmitted by the spoofing interference device will be higher than the power of the actual satellite signal. When the GPS receiver is attacked by spoofing signals, it will cause changes in C / N0. Therefore, C / N0 can be used as a feature quantity for spoofing interference detection. The carrier-to-noise power spectrum density of the receiver at a given time t is represented as (C / N0) t .

[0110] (2) Signal correlation peak feature

[0111] The signal correlation peak feature is mainly extracted based on signal quality monitoring (SQM) and used to describe the morphological changes of the satellite signal correlation function when affected by spoofing interference. When the receiver is attacked by spoofing interference, the superposition and interference effects between the spoofing signal and the real satellite signal will cause the distortion of the correlation peak of the tracking phase, and such distortion can be identified by analyzing the output values of the correlator. Specifically, by obtaining the output values of the Early, Prompt and Late correlators in the tracking loop, the changes in a certain time window are calculated to detect whether the correlation function is distorted and to identify potential spoofing attack behavior. The specific measurement indicators are:

[0112] ① Delta indicator

[0113] The Delta indicator is used to measure the power difference between the Early and Late correlators, reflecting the symmetry change of the correlation function. Its calculation formula is:

[0114]

[0115] Where I E (t), I L (t) and I P (t) represent the in-phase component outputs of the Early, Late and Prompt correlators, respectively.

[0116] Under the condition of no spoofing interference, the correlation function should be approximately symmetrical, and the difference between the Early and Late outputs is close to zero. Spoofing interference will cause peak shift, resulting in a significant deviation of this value, which can be used for spoofing detection.

[0117] ② Ratio indicator

[0118] The Ratio indicator measures the difference between the average values of the Prompt and Early / Late correlators, reflecting the sharpness or distortion of the correlation peak. Its calculation formula can be expressed as:

[0119]

[0120] Under the condition of no spoofing interference, the correlation function has a clear peak at the Prompt position, and the Early and Late outputs are relatively low, showing a typical sharp correlation morphology. In the presence of spoofing interference, the superposition effect of the interference signal and the real signal will cause the correlation function peak to be widened or even appear a "flat" region, resulting in an increase in the Early and Late outputs, thus causing changes in the Ratio indicator, reflecting the decrease in the sharpness of the correlation peak. Therefore, the abnormal change of Ratio can be used as a sensitive identification feature of spoofing attack.

[0121] ③ELP(Early-Late Phase, ELP) index

[0122] ELP is a signal quality monitoring (SQM) index based on carrier phase information, which is used to detect the phase consistency of BDS receiving signals in the correlation structure. The index extracts the instantaneous phase difference by performing arctangent operation on the ratio of in-phase (I) and quadrature (Q) components of the Early and Late correlator outputs, and then identifies whether the correlation peak is disturbed by spoofing signals. Its definition formula is:

[0123]

[0124] where Q E (t), Q L (t) are the quadrature component outputs of the Early and Late correlators, respectively.

[0125] In the normal receiving environment without interference, the complex vector corresponding to the Early and Late correlator outputs should have a symmetric structure, and the ELP index should maintain a stable level. When the BDS signal is subjected to spoofing attack, the superposition effect between the spoofing signal and the real signal will destroy the original symmetry, causing a significant phase shift between the Early and Late codes, and thus making the ELP value deviate significantly from the normal range. Therefore, ELP can be widely applied in spoofing detection tasks as one of the core features to measure the consistency of signal structure.

[0126] ④Q-channel index

[0127] The energy of satellite navigation signals is mainly concentrated in the in-phase channel (I-channel), and the quadrature channel (Q-channel) output is mostly noise. When there is spoofing interference, there is a carrier phase difference between the spoofing signal and the real signal, and abnormal energy will be generated in the quadrature component, which can be used to monitor spoofing attacks. Based on this, the SQM index of the Q-channel is defined as:

[0128]

[0129] m SQM The index identifies the injection of spoofing signals by measuring the abnormal energy of the quadrature channel. If the quadrature channel energy is abnormally high, it means that the signal modulation structure or phase consistency is destroyed. Compared with the traditional Delta and Ratio indexes based on the I-channel, m SQM is more sensitive to weak phase disturbances, so it is used as a detection feature of spoofing interference.

[0130] The SQM method based on instantaneous features (such as Delta, Ratio, ELP, etc.) only relies on the correlator output value at a certain moment for judgment, and such method is easily affected by instantaneous noise interference when facing gradual or slowly changing deception attacks, and it is difficult to effectively capture the change trend of signal features over time. In order to improve the sensitivity and robustness of deception signal detection, a sliding statistical mechanism is introduced, and the mean and variance of feature values in a sliding window are calculated to extract the fluctuation characteristics and evolution trend of time series features. Figure 2 A schematic diagram for data processing using a sliding window mechanism provided by an embodiment of the present application is shown in Figure 2 On the basis of setting a fixed window length, the sliding mean and sliding variance of different types of feature values are calculated respectively to enhance the detection ability of weak interference or gradual evolution attacks. The calculation formulas of the sliding mean and the sliding variance are as follows:

[0131]

[0132] Wherein, μ x (n) represents the mean of feature x at the n th sliding time step, σ Delta (n) represents the variance of feature x at the n th sliding time step, n represents the number of sliding windows, L represents the size of the sliding window, and x(i) represents the feature value at the i th moment. The mean of the Delta index, the Ratio index, the ELP index and the Q channel index at the n th sliding time step is represented as μ Ratio (n), μ ELP (n), μ The variance of the Delta index, the Ratio index, the ELP index and the Q channel index at the n th sliding time step is represented as σ In this embodiment, L = 100, which corresponds to a time length of 100 ms contained in each sliding time step.

[0133] (3) Signal tracking stability feature

[0134] In the tracking loop processing process of a software receiver, the code tracking loop (DLL, Delay Lock Loop) and the carrier tracking loop (PLL, Phase Lock Loop) are the key links to realize continuous signal tracking. Through feedback control mechanism, they constantly calibrate the pseudo code sequence and carrier signal generated locally, so as to keep them synchronized with the received satellite signal. Among them, the discriminator (Discriminator) as the core module in the loop, its output directly reflects the tracking error of code phase and carrier phase, so it can be used to quantify the locking stability of the receiver. The output of the code loop discriminator represents the code phase tracking error, which can be calculated by the following formula:

[0135]

[0136] where I E and I L represent the in-phase component output of the early and late correlators, respectively, and Q E and Q L represent the quadrature component output of the early and late correlators, respectively.

[0137] The output of the carrier loop discriminator represents the carrier phase tracking error, which can be calculated by:

[0138]

[0139] where Q P and I P represent the in-phase and quadrature component output of the prompt correlator, respectively.

[0140] When the receiver is under the spoofing jamming attack, the external false signal will disturb the receiver's locking ability to the real signal, and destroy the dynamic balance of the DLL and the PLL, which is manifested as significant fluctuations or trend deviations in the outputs of the two discriminators. Therefore, the outputs of the two discriminators can be used as effective stability features for spoofing jamming detection. At the same time, in order to enhance the sensitivity to slowly varying interference and the robustness against instantaneous noise, a sliding statistical mechanism is introduced to calculate the mean and variance of the sliding window. The mean of the code phase tracking error and the carrier phase tracking error at the nth sliding time step are denoted as μ dllDiscr (n) and μ pllDiscr (n), and the variances are denoted as and

[0141] Finally, the three types of statistical features, including the signal power feature, the signal correlation peak feature, and the signal tracking stability feature, contain 19 specific feature indicators in total. In order to realize the unified processing of instantaneous features and sliding window statistical features in the time series modeling process, the sampling frequency of the instantaneous features is set to be consistent with the sliding window update interval , so as to ensure that different features can be aligned and fused in a unified time scale. In this embodiment, f2=10 Hz is obtained by calculation. Based on this, the signal statistical feature vector fv at time step t is represented as:

[0142]

[0143] The extracted features of various types are aligned in the time dimension, and a feature time series graph is constructed according to the preset spoofing detection frequency f3 (usually f2>f3). The feature time series graph F T at time T is represented as:

[0144]

[0145] wherein the size of the feature time series graph is (f2 / f3) x 19, (f2 / f3) represents the number of time points contained in the feature time series graph, and 19 represents the feature dimension. In this embodiment, f3 = 1 Hz, and it can be obtained that the size of the feature time series graph F T is 10 x 19.

[0146] In step S3, for the Beidou navigation spoofing interference detection task in the rail transit train scenario, an unsupervised spoofing interference detection method based on an improved diffusion model is designed, the diffusion model is trained by using the unlabeled "zero spoofing" training feature sample set containing only normal samples, so that the model can learn and reconstruct the feature time series graph of the normal navigation signal.

[0147] Based on the feature evolution law of the spoofing interference signal in the satellite positioning of the rail transit train and the time sequence-feature two-dimensional structure of the feature time series graph constructed in step S2, the diffusion model structure is adaptively improved. A diffusion reconstruction network suitable for the unsupervised spoofing interference detection task in the rail transit field is designed, so that the diffusion reconstruction network can accurately learn the distribution of the "zero spoofing" feature time series graph and realize high-fidelity reconstruction, suppress the fitting ability of the diffusion reconstruction network to the spoofing samples, and further produce a differential response to the abnormal feature time series graph. A model architecture diagram for realizing the unsupervised spoofing interference detection of the "zero spoofing" sample based on the diffusion model provided in the embodiment of the present application is shown in FIG. 8. In this embodiment, the noise prediction network in the diffusion model is optimized and improved according to the characteristics of the extracted satellite signal feature time series graph. Figure 3

[0148] A noise prediction network structure diagram designed for the Beidou satellite positioning spoofing interference detection provided in the embodiment of the present application is shown in FIG. 9. Figure 4 ​As shown, the U-Net is used as the basic architecture, and several key modules are introduced. First, considering that traditional convolutional layers may have information loss or insufficient expression ability when processing complex time series data, the residual network (ResNet) module is introduced in the encoder and decoder part of the U-Net to replace the traditional convolutional block. ResNet effectively alleviates the gradient vanishing problem of deep networks through residual connections, enhances the transmission of information in deep networks, and significantly improves the feature extraction ability and learning depth of the network. Second, to further improve the sensitivity of the network to key features, an efficient channel attention (ECA) module is introduced. This module dynamically adjusts the weights of different channels, allowing the network to focus on important information when processing complex features, thereby strengthening the learning ability of key features and improving the learning efficiency of the network. In addition, to better capture the time-varying characteristics of the feature time series graph, a one-dimensional time convolution down / up sampling module is designed and embedded in the Encoder and Decoder for time dimension feature learning. This design enhances the network's modeling ability in the time dimension, thereby improving the accuracy and effectiveness of noise prediction. The down-sampling module of the Encoder in the network is connected to the corresponding up-sampling module in the Decoder through skip connections to preserve shallow feature information. This mechanism effectively avoids information loss during the reconstruction process and improves the convergence speed and overall performance of the network.

[0149] S3.2, using the unlabeled "zero fraud" training feature sample set constructed in step S2 to train the improved diffusion model, the training process includes the following steps:

[0150] (1) Initialize the training data:

[0151] Standardize the training feature sample set to ensure uniform input dimensions and adapt to the model structure;

[0152] (2) Define the key hyperparameters for model training:

[0153] Including the maximum time step T of the diffusion process, the noise scheduling function (such as linear / cosine, etc.), the learning rate, the optimizer type (such as Adam), the number of training epochs, the batch size, etc.

[0154] In this embodiment, the maximum time step T is set to 1000, the Adam optimizer is used, the learning rate is set to 2.5×10 -5 , the number of training epochs is set to 300, and the batch size is set to 256. The noise scheduling function is set to linear scaling, represented as:

[0155]

[0156] wherein, denotes and uniformly generates T numbers between and t denotes the noise intensity at the t-th time step, start and end are the starting and ending values of the noise intensity, respectively, taking 0.0015 and 0.0195, respectively.

[0157] (3) Perform the model training process:

[0158] The training process of the diffusion model is a Markov modeling process aiming at denoising, which aims to learn how to recover the original clean sample from the noisy data. Its training process mainly includes two stages:

[0159] ① Forward diffusion process:

[0160] In the forward process, a fixed noise-added Markov chain is used to add Gaussian noise to the "zero deception" feature time series x0 sampled from the unlabeled "zero deception" training feature sample set, to generate a series of noisy feature time series x1, x2,..., x T The noise-added process can be represented as:

[0161]

[0162] wherein, denotes the cumulative signal preservation coefficient of the previous t steps, s = 1 - β s , and ε denotes the standard Gaussian noise, x t denotes the noisy feature time series generated at the t-th step.

[0163] ② Reverse denoising process:

[0164] The main purpose of the reverse process is to recover the original "zero deception" feature time series x0 from the noisy feature time series x t . This process is based on the modeling of the forward diffusion chain and relies on a noise prediction network (usually adopting a U-Net structure) to fit the mapping relationship between the noisy feature time series x t and the original noise ε, to accurately estimate the noise component ε in each noisy feature time series.

[0165] The modeling target of the noise prediction network ε θ is to accurately estimate the noise ε in the noisy feature time series x t at any time step t. The most commonly used loss function in the network training process is the mean square error loss:

[0166]

[0167] wherein the time step t is usually randomly and uniformly sampled from [1, T]. The network parameters θ are iteratively optimized by gradient descent algorithm to minimize the loss function, improving the network's accurate fitting ability to noise under different noise levels.

[0168] After training, the diffusion model reconstructs and generates the feature time series based on the reverse sampling strategy:

[0169]

[0170] wherein x t represents the noise-added feature time series at time step t, ε θ (x t , t) represents the noise component predicted by the model, σ t represents the noise amplitude sampled at time step t, x t-1 represents the denoised feature time series at the previous time step. This process is iteratively performed in the full time domain from t = T → 0, and finally the complete "zero deception" feature time series x0 is obtained.

[0171] Through the above training, the diffusion model can gradually reconstruct the original input "zero deception" feature time series x0 from the noise-added feature time series x t with any noise intensity.

[0172] (4) Verify the model performance:

[0173] Use the "zero deception" verification feature sample set to evaluate the model performance, calculate the reconstruction error evaluation indicators such as mean square error MSE and structural similarity SSIM between the input feature time series and the reconstructed feature time series, and evaluate the learning ability and generalization effect of the model on normal samples.

[0174] S3.3, In the training process, save the model parameters with the minimum loss function value, and use them as the final detection model parameters for subsequent experimental verification and online detection and identification of deception interference during train operation.

[0175] Step S4, input the "zero deception" verification feature sample set and the test feature sample set into the trained diffusion model to obtain the model reconstruction graph, use the mean square error and the structural similarity index to quantify the model reconstruction error, analyze the deviation degree of the test feature sample reconstruction error relative to the "zero deception" verification feature sample reconstruction error distribution, and judge whether the test feature sample is abnormal, realize the detection and identification of deception signals, including:

[0176] S4.1, test verification is carried out by using the model saved in step S3.3, and a model reconstruction diagram is generated, specifically:

[0177] (1) Feature time sequence diagram initialization

[0178] The "zero deception" verification feature sample set obtained in step S2 and the test feature sample set are standardized to ensure that the size, channel number and time sequence structure of the input data are consistent with the training model, so as to adapt to the network input interface.

[0179] (2) Feature time sequence diagram noise processing

[0180] According to the preset time step, noise is added to the original feature time sequence diagram x0 to construct a noisy feature time sequence diagram x t , and the calculation formula is:

[0181]

[0182] , wherein represents the cumulative signal retention coefficient of the previous t steps, ε represents the standard Gaussian noise, and x t represents the noisy feature time sequence diagram generated at the t step. In this embodiment, 10 time steps are used to obtain the noisy feature time sequence diagram, and the time steps are t = [10, 110, 210, 310, 410, 510, 610, 710, 810, 910].

[0183] (3) Predicting noise

[0184] The trained noise prediction network (such as U-Net) is called to take the noisy feature time sequence diagram x t and the corresponding time step t as input, and predict the noise component contained in the current noisy feature time sequence diagram:

[0185]

[0186] (4) Step-by-step denoising reconstruction

[0187] Based on the predicted noise information, the denoising restoration process is performed step by step according to the reverse sampling path set by the scheduler, and finally the denoised sample estimate

[0188]

[0189] , wherein σ t is the residual noise intensity of each step, In this embodiment, the PLMS accelerated sampling method is used for denoising reconstruction, and denoising operation is performed every 10 time steps. Through multi-step prediction and correction, the denoising process is accelerated, making the generation process more efficient while maintaining high generation quality.

[0190] (5) Reconstruction Error Calculation

[0191] The reconstruction error between the original feature time series x0and the reconstructed feature time series is calculated respectively to measure the fitting ability of the model. Typical error indicators include:

[0192] ① Mean Squared Error (MSE, Mean Squared Error):

[0193] Mean Squared Error is a basic indicator to measure the numerical difference between the reconstruction result of the model and the original input, which is represented as the average of the squared error between the predicted value and the true value. Its mathematical expression is:

[0194]

[0195] where D represents the total number of pixels in the sample, represents the value of the jth pixel point of the original feature time series, represents the value of the jth pixel point of the reconstructed feature time series. The smaller the MSE value, the closer the reconstruction result of the model to the original input, i.e. the higher the reconstruction accuracy.

[0196] ② Structural Similarity (SSIM, Structural Similarity Index):

[0197] SSIM is an indicator to measure image quality, aiming to simulate the perception of human visual system on structural similarity, measuring the similarity of two images in structure, brightness and contrast, and is widely used to evaluate image reconstruction quality. It is defined as:

[0198]

[0199] where and are the average brightness of images x0and , respectively, and are the variances of images x0and , respectively, is the covariance of images x0and , C1 and C2 are constants for stable calculation, usually set as C1=(K1L) 2 and C2=(K2L) 2 , where L is the dynamic range of pixel value. The calculated SSIM ranges from 0 to 1, and when the two images are exactly the same, the SSIM value reaches the maximum value 1. The smaller the SSIM value, the worse the reconstruction quality.

[0200] MSE is used to measure the overall error of the image at the pixel level, and SSIM measures the structural fidelity of the image. The two are used together to comprehensively reflect the reconstruction ability of the model from the two dimensions of error amplitude and structural similarity. In the embodiment, the two reconstruction error indicators of MSE and SSIM are calculated respectively for subsequent detection and identification.

[0201] Based on this, the reconstruction error of the model for the "zero deception" verification feature sample and the test feature sample is obtained.

[0202] S4.2, by analyzing the deviation of the reconstruction error of the test feature sample from the distribution of the reconstruction error of the "zero deception" verification feature sample, it is judged whether the test feature sample has feature anomaly, and the detection and identification of the deception interference signal are realized, specifically:

[0203] (1) Calculate the statistical characteristics of the reconstruction error of the "zero deception" verification feature sample

[0204] Based on the reconstruction error of the "zero deception" verification feature sample calculated in step S4.1, the statistical quantity of the reconstruction error of the entire "zero deception" verification feature sample is calculated, and the reconstruction error of the i-th sample in the "zero deception" verification feature sample set is represented as The mean μ val and the standard deviation σ val of the reconstruction error of the entire "zero deception" verification feature sample set are:

[0205]

[0206] Where N val represents the number of "zero deception" verification feature samples, and the mean and standard deviation are used as reference standards for calculating Z-score in subsequent anomaly detection.

[0207] (2) Calculate the Z-score of the reconstruction error of the test feature data set

[0208] Based on the mean μ val and the standard deviation σ val of the reconstruction error of the "zero deception" verification feature sample, the reconstruction error of the test feature sample is standardized, and the reconstruction error of the j-th sample in the test feature sample set is represented as The corresponding Z-score is:

[0209]

[0210] Where Z (j)The Z-score reflects the jth test feature sample, which reflects the deviation of its reconstruction error from the "zero deception" verification feature sample reconstruction error distribution. The larger the Z-score, the more likely the sample is a sample interfered by deception. In this embodiment, the average of the Z-scores calculated by setting 10 different time steps t = [10, 110, 210, 310, 410, 510, 610, 710, 810, 910] is used for subsequent detection performance index calculation.

[0211] (3) Constructing ROC curve and calculating AUC value

[0212] Taking the Z-score as the abnormal detection score index, combined with the real label of the sample in the test set (0 represents normal, 1 represents deception), the ROC curve (Receiver Operating Characteristic Curve) is constructed. By traversing different Z-score thresholds, the false positive rate (FPR) and the true positive rate (TPR) under the threshold are calculated:

[0213]

[0214] Finally, the area under the ROC curve (Area Under Curve, AUC) is obtained by integral method or numerical calculation method:

[0215]

[0216] The AUC value reflects the discrimination ability of the detection model for "normal / deception" samples. The closer the AUC is to 1, the stronger the discrimination ability of the model based on reconstruction error; the AUC close to 0.5 indicates that the model has almost no discrimination ability for "normal / deception" samples.

[0217] In this embodiment, we use the Z-scores of the mean square error (MSE) and the structural similarity (SSIM) two error indicators to evaluate the performance of the detection model, Figure 5 The ROC curve diagram of the deception interference test result based on different reconstruction error indicators provided by the embodiment of the present application is as follows: Figure 5The ROC curve based on different error indicators is shown, from which it can be observed that the model performs better in recognition ability when detecting the combination of the two error indicators, and the AUC (area under the curve) value can reach 99.78%. The AUC values of different error indicators in the experiment are shown in Table 1. The experimental results show that compared with the method of using SSIM or MSE alone, the AUC value is increased by 3.05% and 0.12% respectively when using the combination of MSE and SSIM, which proves the effectiveness of the combined index method in improving the detection accuracy.

[0218] Table 1 AUC values obtained by different reconstruction error indicators

[0219] reconstruction error metric AUC SSIM 96.83％ MSE 99.66％ MSE+SSIM 99.78％

[0220] (4) Determine the optimal threshold and calculate the accuracy, precision, recall and F1 score

[0221] To further quantify the performance of the model in classifying test samples based on reconstruction error, the quantile method is used to determine the optimal threshold τ * , and then the classification performance indicators are calculated, the specific steps are as follows:

[0222] ① Select the threshold determination method to determine the optimal threshold τ * , common methods include maximizing Youden index, quantile method, etc. In this embodiment, the quantile method is used to set the threshold. Specifically, the Z-score of the verification feature sample set is calculated based on the reconstruction error, and the 95% quantile value of the Z-score distribution is taken as the optimal threshold τ * .

[0223] Based on this, the optimal thresholds based on SSIM, MSE and MSE+SSIM are 1.3341, 0.2849 and 1.4102 respectively.

[0224] ② Divide the predicted label (0 represents normal, 1 represents fraud) according to the optimal threshold τ * :

[0225]

[0226] Wherein, is the predicted label of the jth sample by the model.

[0227] ③ Calculate the detection performance indicators

[0228] After obtaining the predicted label, the following indicators can be calculated based on the statistics of the confusion matrix:

[0229] (a) Accuracy (Accuracy): The proportion of the model's overall correct prediction

[0230]

[0231] (b) Precision: the proportion of true anomalies among all predicted anomalies

[0232]

[0233] (c) Recall: the proportion of correctly identified anomalies among all anomalies

[0234]

[0235] (d) F1 Score: the harmonic mean of precision and recall, measuring the overall performance

[0236]

[0237] where TP represents the number of samples that are actually fraudulent and predicted to be fraudulent; TN represents the number of samples that are actually normal and predicted to be normal; FP represents the number of samples that are actually normal but predicted to be fraudulent; and FN represents the number of samples that are actually fraudulent but predicted to be normal.

[0238] In this embodiment, the values of accuracy, precision, recall, and F1 score based on SSIM, MSE, and MSE+SSIM are shown in Table 2. Figure 6 The performance comparison chart of fraud interference testing based on different reconstruction error indicators provided by the embodiment of the present application.

[0239] Table 2 Detection performance obtained by different reconstruction error indicators

[0240] reconstruction error metric Accuracy Precision Recall F1 Score SSIM 91.83％ 97.29％ 91.67％ 94.39％ MSE 96.58％ 96.23％ 99.33％ 97.76％ MSE+SSIM 97.83％ 97.60％ 99.56％ 98.57％

[0241] The experimental results show that whether SSIM or MSE is used alone or both are combined for detection, the model has the ability to detect fraud interference to some extent. However, compared with the method of using SSIM or MSE alone, the combination of MSE and SSIM for detection can improve the accuracy by 6.53% and 1.29%, respectively, the precision by 0.32% and 1.42%, respectively, the recall by 8.61% and 0.23%, respectively, and the F1 score by 4.43% and 0.83%, respectively, proving that the combination of error indicators for fraud detection is a more reliable detection scheme, suitable for unsupervised fraud interference detection tasks, and can effectively improve the detection accuracy.

[0242] Step S5, during the train in-transit operation, real-time receiving the Beidou satellite digital intermediate frequency signal, extracting signal correlation statistical features and constructing satellite signal feature time sequence diagram, calling the trained diffusion model and obtaining the model reconstruction diagram, calculating the mean square error, structural similarity and other reconstruction error indicators, and judging whether the train is attacked by deception based on the size of the reconstruction error indicator value.

[0243] During the train in-transit operation, real-time receiving the Beidou satellite digital intermediate frequency signal, and sending the collected satellite positioning digital intermediate frequency signal into a software receiver for processing; according to the processing flow of the acquisition and tracking loop, extracting three types of statistical features including signal power, signal peak value, signal tracking stability, and constructing the feature time sequence diagram according to step S2 to obtain the feature time sequence diagram F T at time T, which is represented as:

[0244] The feature time sequence diagram F T is input to the detection model saved in step S3, the model reconstruction diagram is generated according to step S4, and the mean square error, structural similarity and other reconstruction error indicators are calculated; taking the "zero deception" verification feature sample set as the benchmark, the Z-score indicator of the current reconstruction error is calculated, and the Z-score value is compared with the best threshold value τ * obtained in step S4; when the Z-score value is greater than the best threshold value τ * , it is judged that the train is attacked by deception at the current time; when the Z-score value is not greater than the best threshold value τ * , it is judged that the train is not attacked by deception at the current time. Thus, whether the train is attacked by deception at the current time is judged, and the unsupervised deception detection of the train satellite positioning is realized.

[0245] In summary, the present application designs a kind of unsupervised deception detection method for train satellite positioning based on zero deception sample, which is used for the identification of deception interference signal. This method mainly extracts satellite signal related statistical features, constructs satellite signal feature time sequence diagram with time as dimension, captures the dynamic changes of signal, and forms a no-label feature sample set. A kind of unsupervised deception interference detection model based on improved diffusion model is designed, and the model is trained based on the no-label "zero deception" training feature sample set, so that the model can accurately learn the distribution law of no-interference sample and realize high-fidelity reconstruction of normal signal. For abnormal sample, the model can effectively identify its deviation from "zero deception" normal sample through differential response, thereby suppressing the fitting ability of deception sample, realizing deception detection based on reconstruction error. This method can effectively overcome the dependence of traditional supervised learning method on a large number of labeled data, and significantly improve the train satellite.

[0246] The application provides a train Beidou positioning unsupervised deception detection method based on zero deception samples. The method fully utilizes Beidou satellite navigation positioning digital intermediate frequency signals, extracts three types of statistical features including signal power, signal correlation peak and signal tracking stability, constructs a satellite signal feature time sequence diagram with time as a dimension, adopts a label-free 'zero deception' verification feature sample set for diffusion model training, enables the model to learn and reconstruct the feature time sequence diagram of the 'zero deception' navigation signal, and produces a differential response to the feature time sequence diagram of the deceptive satellite signal, so that real-time detection of the deceptive signal during train operation is realized.

[0247] The application can be applied to the application of different types of satellite navigation systems under observable conditions, is universal for different types of satellite navigation deception interference modes, various railway line network conditions, different forms of surrounding environment conditions and different network scales, and has significant engineering application value.

[0248] Those skilled in the art can understand that the drawings are only schematic diagrams of an embodiment, and the modules or flows in the drawings are not necessarily essential for implementing the application.

[0249] From the above description of the embodiments, those skilled in the art can clearly understand that the application can be implemented by means of software and a necessary general hardware platform. Based on such understanding, the technical solutions of the application can be embodied in the form of a software product, which can be stored in a storage medium, such as a ROM / RAM, a magnetic disk, an optical disk, etc., and includes a plurality of instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in various embodiments or some parts of the embodiments.

[0250] Each of the embodiments in the specification is described in a progressive manner, and the same or similar parts of each embodiment can be referred to each other, and each embodiment mainly describes the differences from other embodiments. Especially, since the device or system embodiment is basically similar to the method embodiment, the description is relatively simple, and the related parts can be referred to the part of the method embodiment. The device and system embodiments described above are only schematic, and the units described as separate components can be or can not be physically separated, and the components displayed as units can be or can not be physical units, that is, they can be located in one place, or can be distributed on multiple network units. According to the actual needs, part or all of the modules can be selected to achieve the purpose of the embodiment. Those skilled in the art can understand and implement without creative labor.

[0251] The above description is only preferred specific embodiments of the present application, but the protection scope of the present application is not limited thereto, any person skilled in the art can easily think of changes or replacements within the technical range disclosed by the present application, which should be covered in the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.

Claims

1. A train Beidou positioning unsupervised fraud detection method based on zero fraud samples, characterized in that, The application relates to a method for detecting spoofing interference in train satellite positioning. The method comprises the following steps: Collecting a digital intermediate frequency signal of train satellite positioning, constructing a "zero spoofing sample" training data set and a "zero spoofing sample" verification data set which do not contain any spoofing interference, and generating a test feature data set containing spoofing interference by injecting a spoofing signal; Extracting satellite signal related statistical features, constructing a satellite signal feature time sequence graph in the time dimension, and generating a no-label "zero spoofing" training feature sample set, a no-label "zero spoofing" verification feature sample set and a test feature sample set by using the satellite signal feature time sequence graph; Constructing a diffusion reconstruction network suitable for an unsupervised spoofing interference detection task in the field of rail transit, training the diffusion reconstruction network by using the no-label "zero spoofing" training feature sample set, verifying the diffusion reconstruction network by using the verification feature sample set and the test feature sample set, quantifying the reconstruction error of the diffusion reconstruction network by using a mean square error and a structural similarity index, and obtaining a trained diffusion reconstruction network; 2. The method of claim 1, wherein, During the in-route running of a train, a satellite digital intermediate frequency signal is received in real time, satellite signal related statistical features are extracted and a signal feature time sequence graph is generated, the signal feature time sequence graph is input into the trained diffusion reconstruction network, a model reconstruction graph is output by the diffusion reconstruction network, a reconstruction error index value is calculated, and whether the train is subjected to spoofing attack is judged based on the size of the reconstruction error index value. The method for collecting a digital intermediate frequency signal of train satellite positioning, constructing a "zero spoofing sample" training data set and a "zero spoofing sample" verification data set which do not contain any spoofing interference, and generating a test feature data set containing spoofing interference comprises the following steps: During the running of a train on a specific line, positioning data are collected by using a train position reference system, a train running track file is generated, a satellite navigation signal simulation and collection environment is built, the train running track file is injected into a simulation system, a train running scene is reproduced, satellite positioning digital intermediate frequency signals of a certain time length are collected, and a "zero spoofing sample" training data set and a "zero spoofing sample" verification data set which do not contain any spoofing interference are generated; 3. The method of claim 1, wherein, During the reproduction of the train running scene, spoofing interference signals are injected into the "zero spoofing sample" training data set and the "zero spoofing sample" verification data set in a predetermined time window, satellite positioning digital intermediate frequency signals of a corresponding period are collected, and a test feature data set containing spoofing interference is generated. The method for extracting satellite signal related statistical features and constructing a satellite signal feature time sequence graph in the time dimension, and generating a no-label "zero spoofing" training feature sample set comprises the following steps: The collected satellite positioning digital intermediate frequency signal of the train is sent into a software receiver for processing, signal power features, signal correlation peak features and signal tracking stability features of the satellite positioning digital intermediate frequency signal are extracted in the frequency f1 according to the processing procedure of a capture and tracking loop: (1) the carrier noise power spectral density C / N0 is adopted as the signal power feature, and the signal power feature reflects the ratio between a demodulation signal power C and a noise power spectral density N0: where P C represents the power of the carrier signal, P N is the noise power within the passband; (2) The signal correlation peak features include a Delta index, a Ratio index, an ELP index, and a Q channel index; The calculation formula of the Delta index is: where I E (t),I L (t),I P (t) are the in-phase component outputs of the early, late, and prompt correlators, respectively. The calculation formula of the Ratio index is: The calculation formula of the ELP index is: where Q E (t),Q L (t) are the early and late correlator quadrature component outputs, respectively. The calculation formula of the Q channel index is: m SQM The index can identify the injection of fraudulent signals by measuring the normal traffic channel abnormal energy. If the normal traffic channel energy is abnormally high, it means that the signal modulation structure or phase consistency is destroyed. On the basis of setting a fixed window length, the sliding mean and the sliding variance of different types of feature quantities are calculated, and the calculation formulas of the sliding mean and the sliding variance are respectively: where μ x (n) represents the mean value of feature x at the n-th sliding time step, represents the variance of feature x at the n-th sliding time step, n represents the number of sliding windows, L represents the size of sliding window, x(i) represents the feature value at the i-th time, the mean values of the Delta indicator, the Ratio indicator, the ELP indicator and the Q-channel indicator at the n-th sliding time step are respectively represented as μ Delta (n), μ Ratio (n), μ ELP (n) and the variances at the n-th sliding time step are respectively represented as and (3) The signal tracking stability features include a code tracking loop and a carrier tracking loop, the output of the code loop discriminator represents a code phase tracking error, which is calculated by the following formula: where I E and I L represent the in-phase component outputs of the early and late correlators, respectively, and Q E and Q L represent the quadrature component outputs of the early and late correlators, respectively. The output of the carrier loop discriminator represents a carrier phase tracking error, which is calculated by the following formula: where I P and Q P represent the in-phase and quadrature component outputs of the prompt code correlator, respectively; The code phase tracking error and carrier phase tracking error mean at the n th sliding time step are denoted as μ dllDiscr (n) and μ pllDiscr (n), respectively, and the variances are denoted as and The extracted signal power features, signal correlation peak features, and signal tracking stability features include 19 specific feature indicators in total, and the sampling frequency of the instantaneous features is set to be consistent with the sliding window update interval The signal statistical feature vector fv at time step t is denoted as: The extracted features of various types are aligned in the time dimension, and a feature time sequence diagram is constructed according to a preset fraud detection frequency f3. The feature time sequence diagram F at time T is as follows: T is represented as: Wherein, the size of the satellite signal feature time series diagram is (f2 / f3) x 19, (f2 / f3) represents the number of time points contained in the feature time series diagram, and 19 represents the feature dimension. The unlabeled "zero deception" training feature sample set is generated by using the satellite signal feature time series diagram.

4. The method of claim 1, wherein, The diffusion reconstruction network suitable for the unsupervised deception jamming detection task in the field of rail transit is constructed, and the diffusion reconstruction network is trained by using the unlabeled "zero deception" training feature sample set, including: Based on the time series-feature two-dimensional structure of the satellite signal feature time series diagram, the noise prediction network in the diffusion model structure is optimized and improved. The residual network ResNet module is introduced into the encoder and decoder parts of the U-Net to replace the convolution block. The channel attention ECA module is introduced, and a one-dimensional time convolution down / up sampling module is designed and embedded into the encoder and decoder for feature learning in the time dimension. The down sampling module of the encoder in the network is connected to the corresponding up sampling module in the decoder through a jump connection to construct the diffusion reconstruction network suitable for the unsupervised deception jamming detection task in the field of rail transit. The diffusion reconstruction network is trained by using the unlabeled "zero deception" training feature sample set, and the training process includes the following steps: (1) Initialize the training data: The unlabeled "zero deception" training feature sample set is standardized; (2) Define the key hyperparameters of model training: Including the maximum time step T of the diffusion process, the noise scheduling function, the learning rate, the optimizer type, the number of training rounds epochs, and the batch size; The noise scheduling function is set to linear scaling, which is represented as: wherein, represents and T number of uniform generation between β t represents the noise intensity at the t-th time step, β start and β end are the starting value and the terminal value of the noise intensity, respectively; (3) The training process of the diffusion reconstruction network is a Markov modeling process with the goal of denoising. The training process includes a forward diffusion process and a reverse denoising process. After the training is completed, the diffusion reconstruction network realizes the reconstruction and generation of the feature time series diagram based on the reverse sampling strategy: where x t denotes the noisy feature time series at time step t, ε θ (x t ,t) denotes the model-predicted noise component, σ t denotes the noise amplitude sampled at this time step t, x t-1 denotes the de-noised feature time series at the previous time step, and this process is iteratively performed over the full time domain t = T → 0 to reconstruct the "zero-spoofing" feature time series 5. The method of claim 1, wherein, The diffusion reconstruction network is trained by using the unlabeled "zero deception" training feature sample set, and the training process includes the following steps: reconstruction error between the original feature time series x0 and the reconstructed feature time series The reconstruction error includes MSE and SSIM, and the mathematical expression of MSE is: wherein D represents the total number of pixels in the sample, represents the value of the jth pixel in the original feature time series, represents the value of the jth pixel in the reconstructed feature time series. The mathematical expression of SSIM is: wherein and is the average luminance of the image x0and and is the variance of the image x0and is the covariance of the image x0and C1and C2are constants used for stabilization of the calculation, C1= (K1L) 2 and C2= (K2L) 2 where L is the dynamic range of the pixel values;​​ The reconstruction error of the i-th sample in the set of unlabeled "zero-fraud" verification feature samples is denoted as The mean μ and the standard deviation σ of the reconstruction errors of the entire set of unlabeled "zero-fraud" verification feature samples are respectively: val val ​​ where N val represents the "zero spoofing" verification feature sample number; According to the calculation process of the reconstruction error of the i-th sample in the no-label "zero fraud" verification feature sample set, the reconstruction error of the j-th test feature sample in the test feature data set is calculated as The corresponding Z-score is: The corresponding Z-score is: wherein Z (j) Z-score reflecting the jth test feature sample; Taking Z-score as the abnormal detection score index, combining the real labels of the samples in the test feature data set, constructing the ROC curve, and calculating the false positive rate FPR and the true positive rate TPR under different Z-score thresholds by traversing the thresholds: An area AUC value under the ROC curve is obtained by integral method or numerical calculation method: When the difference between the AUC value and the threshold value 1 is less than a set numerical range, it is judged that the training process of the diffusion reconstruction network is ended, and a trained diffusion reconstruction network is obtained.

6. The method of claim 1, wherein, In the process of train running, satellite digital intermediate frequency signals are received in real time, signal feature time sequence diagrams are generated by extracting satellite signal related statistical features, the signal feature time sequence diagrams are input into the trained diffusion reconstruction network, the diffusion reconstruction network outputs a model reconstruction diagram, a reconstruction error index value is calculated, and whether the train is subjected to a fraudulent attack is judged based on the size of the reconstruction error index value, including: In the process of train running, satellite digital intermediate frequency signals are received in real time, signal feature time sequence diagrams are generated by extracting satellite signal related statistical features, the signal feature time sequence diagrams are input into the trained diffusion reconstruction network, the diffusion reconstruction network outputs a model reconstruction diagram, a reconstruction error index value is calculated, and whether the train is subjected to a fraudulent attack is judged based on the size of the reconstruction error index value, including: During the train running, the satellite digital intermediate frequency signal is received in real time, the satellite positioning digital intermediate frequency signal is sent into the software receiver for processing, the signal power characteristics, signal peak value characteristics and signal tracking stability characteristics of the satellite digital intermediate frequency signal are extracted, and the signal characteristic time sequence diagram F of the moment T is obtained T , which is expressed as: The signal feature time sequence diagram F T is input to the trained diffusion reconstruction network, and a reconstruction error index including a mean square error and a structural similarity is calculated. Taking the "zero deception" verification feature sample set as a benchmark, a Z-score index of the current reconstruction error is calculated, and the Z-score value is compared with a preselected threshold τ * . When the Z-score value is greater than the threshold τ * , it is judged that the train is subjected to a deception attack at the current time; and when the Z-score value is not greater than the optimal threshold τ * , it is judged that the train is not subjected to a deception attack at the current time.

Images