GNSS real-time deception jamming detection method and system based on deep learning

Abstract

The invention discloses a GNSS real-time deception jamming detection method and system based on deep learning, and belongs to the field of satellite navigation signal processing, and the method comprises the steps: collecting Doppler frequency shift signals of a GNSS receiver in real time, and obtaining an initial training data set; an LSTM prediction model with an incremental learning mechanism is constructed, the LSTM prediction model comprises a double-layer LSTM network and a full connection layer, a gating mechanism is introduced into the double-layer LSTM network, and the initial training data set is adopted to add Gaussian white noise to the prediction model for model training; performing sliding window processing on the initial training data set, and extracting a multi-dimensional feature vector; inputting the trained prediction model for prediction to obtain a model prediction value; calculating a residual error between a model predicted value and a real value, calculating a standard deviation of the residual error after eliminating a preheating period, and dynamically setting a threshold value based on the standard deviation of the residual error; and when the absolute value of the residual exceeds a threshold value and the sampling point processed by the sliding window is located in a preset interference time window, judging that a signal interference event occurs.

Description

Technical Field

[0001] The present invention belongs to the field of satellite navigation signal processing, and particularly relates to a GNSS real-time spoofing interference detection method and system based on deep learning. Background Technique

[0002] GNSS (Global Navigation Satellite System) is the spatio-temporal benchmark of modern information society and has been widely used in key fields such as transportation, communication network, financial transactions, national defense and military. More than 80% of industries globally rely on the high-precision positioning and timing services provided by GNSS. However, there are some inherent defects in the transmission process of GNSS signals, such as extremely low ground power, with a typical signal strength of about -160 dBW, far lower than the ambient noise (-140 dBW). This defect leads to prominent vulnerability of GNSS signals and is extremely vulnerable to spoofing interference.

[0003] The current mainstream interference detection technologies mainly include SQM (Signal Quality Monitoring) and STAP (Space-Time Adaptive Processing) and other technologies. The former is based on single-dimensional indicators such as carrier-to-noise ratio and pseudorange residuals, but relies on empirical threshold setting and has a high false positive rate in complex electromagnetic environments. The latter requires the support of multi-antenna arrays, with high hardware costs and unable to detect concealed spoofing interference.

[0004] The detection method based on deep learning can well improve the deficiencies of traditional detection methods and can jointly analyze signal characteristics from multiple dimensions of time-frequency-space. Although some studies have introduced deep learning into GNSS interference detection, there are still the following key bottlenecks: Firstly, in the long-sequence prediction of traditional deep learning models, as time goes by, the prediction errors gradually accumulate, easily leading to subsequent prediction failures; Secondly, the static model has limitations. Fixed model parameters may not be able to adapt to the dynamic changes of signal characteristics, and the fixed threshold detection method has a high false alarm rate in complex environments and is difficult to distinguish real interference from noise fluctuations. Summary of the Invention

[0005] The purpose of the present invention is to overcome the problems of long-time sequence prediction attenuation and static threshold sensitivity, and proposes a GNSS real-time spoofing interference detection method and system based on deep learning, which can dynamically update model parameters, adapt to environmental changes, and accurately detect GNSS signal interference.

[0006] To achieve the above object, the present invention adopts the following technical solutions: In the first aspect, the present invention provides a GNSS real-time spoofing interference detection method based on deep learning, including the following steps: Collect the Doppler frequency shift signals of the GNSS receiver in real time to obtain the initial training dataset; Construct an LSTM prediction model with an incremental learning mechanism. Use the initial training dataset to add Gaussian white noise to the LSTM prediction model with an incremental learning mechanism for model training to obtain the trained LSTM prediction model with an incremental learning mechanism. The LSTM prediction model with an incremental learning mechanism includes a double-layer LSTM network and a fully connected layer, and the double-layer LSTM network introduces a gating mechanism; After performing sliding window processing on the initial training dataset, extract the multi-dimensional feature vectors; input the multi-dimensional feature vectors into the trained LSTM prediction model with an incremental learning mechanism for prediction to obtain the model prediction values; Calculate the residuals between the model prediction values and the true values of the Doppler frequency shift signals collected in real time. After excluding the warm-up period, calculate the standard deviation of the residuals, and dynamically set the threshold based on the standard deviation of the residuals; Calculate the absolute values of the residuals. When the absolute value of the residual exceeds the threshold and the sampling points of the sliding window processing are within the preset interference time window, it is determined that a signal interference event has occurred.

[0007] Furthermore, the double-layer LSTM network includes an input gate, a forget gate, cell state update, an output gate, and a hidden state, and the multi-dimensional feature vectors include instantaneous values, sliding means, standard deviations, linear trend coefficients, and fast Fourier transform frequency domain components.

[0008] Furthermore, the prediction in inputting the multi-dimensional feature vectors into the trained LSTM prediction model with an incremental learning mechanism for prediction specifically is: Randomly select some historical training data and mix it with the Doppler frequency shift signals collected in real time, and dynamically update the prediction model parameters using a hierarchical update strategy with weight freezing.

[0009] Furthermore, in dynamically updating the prediction model parameters using a hierarchical update strategy with weight freezing, it includes a gradient norm threshold and a hierarchical weight update amount constraint. The gradient norm threshold and the hierarchical weight update amount constraint include: The gradient norm threshold range is set using the gradient clipping method; The weight update amount of the double-layer LSTM network is within the first threshold range of the initial weights; The weight update amount of the fully connected layer is within the second threshold range of the initial weights.

[0010] Furthermore, the hierarchical update strategy with weight freezing adopts an incremental learning rule. The incremental learning rule specifically is:

[0011] The constraint conditions of the incremental learning rule are:

[0012]

[0013] Among them, is the updated parameter of the LSTM layer, is the learning rate of the LSTM layer, ▽ is the gradient mathematical symbol, is the parameter of the LSTM layer, is the loss function, is the updated parameter of the fully connected layer, is the learning rate of the fully connected layer, is the parameter of the fully connected layer, is the parameter value after initial training, is the parameter value of the LSTM layer after initial training, is the parameter value of the fully connected layer after initial training.

[0014] Furthermore, a dynamic threshold is set to introduce a moving average correction factor, specifically:

[0015] Among them, is the dynamic threshold at the current moment, is the smoothing coefficient, is the threshold at the previous moment, is the length of the sliding window, i is an intermediate variable, is the historical residual value, is the mean of the window residual.

[0016] Furthermore, the calculation formula of the historical residual value is specifically:

[0017] Among them, is the historical residual value, is the true value at the moment, is i the predicted value at the moment;

[0018] Among them, is the mean of the window residual, is the length of the sliding window.

[0019] In the second aspect, the present invention provides a GNSS real-time spoofing interference detection system based on deep learning, including: A training data collection module for collecting the Doppler frequency shift signals of a GNSS receiver in real time to obtain an initial training data set; A prediction model training module for constructing an LSTM prediction model with an incremental learning mechanism, using the initial training data set to add Gaussian white noise to the LSTM prediction model with an incremental learning mechanism for model training, and obtaining a trained LSTM prediction model with an incremental learning mechanism. The LSTM prediction model with an incremental learning mechanism includes a double-layer LSTM network and a fully connected layer, and the double-layer LSTM network introduces a gating mechanism; A prediction module for performing a sliding window process on the initial training data set and then extracting a multi-dimensional feature vector; inputting the multi-dimensional feature vector into the trained LSTM prediction model with an incremental learning mechanism for prediction to obtain a model prediction value; A dynamic threshold setting module for calculating the residual between the model prediction value and the true value of the Doppler frequency shift signal collected in real time. After excluding the warm-up period, calculating the standard deviation of the residual, and dynamically setting a threshold based on the standard deviation of the residual; A signal interference event determination module for calculating the absolute value of the residual. When the absolute value of the residual exceeds the threshold and the sampling points processed by the sliding window are within a preset interference time window, it is determined that a signal interference event has occurred.

[0020] In a third aspect, the present invention provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements a GNSS real-time spoofing interference detection method based on deep learning.

[0021] In a fourth aspect, the present invention provides a computer-readable storage medium storing a computer program. When the computer program is executed by a processor, it is a GNSS real-time spoofing interference detection method based on deep learning.

[0022] Compared with the prior art, the present invention has the following beneficial technical effects: The GNSS real-time spoofing interference detection method based on deep learning proposed by the present invention introduces an incremental learning mechanism, enabling the model to dynamically adjust parameters with new data and continuously optimize performance. The double-layer LSTM network with a gating mechanism can capture complex signal features and long-term dependencies, improving prediction accuracy. Gaussian white noise is added during training to enhance the model's stability against noise interference, and the hierarchical update strategy with weight freezing balances the influence of historical data and new data, avoiding drastic parameter fluctuations. Multidimensional feature vectors are extracted after sliding window processing to comprehensively describe the signal, improving detection sensitivity. The interference event is determined by combining the absolute value of the residual and the interference time window, reducing the interference of accidental factors. The sliding window processing adapts to the spatio-temporal changes of the signal, and real-time data fusion enables the model to quickly adapt to the new environment, achieving efficient prediction and calculation, meeting the requirements of real-time detection, and ensuring the stable operation of the GNSS system. The present invention combines LSTM (Long Short-Term Memory) with GNSS signal interference detection, and is applicable to real-time monitoring of spoofing interference events in the Doppler frequency shift signals of GNSS receivers. Innovatively introducing the incremental learning mechanism into the LSTM model training, it solves the problem of performance decay in long-time series prediction of traditional methods. Through multi-dimensional feature fusion and hierarchical learning rate setting, it achieves the technical effects of an interference detection response time less than 0.1 second and a detection accuracy greater than 98%, significantly improving the anti-interference ability of GNSS and realizing real-time spoofing interference detection of GNSS signals. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] The drawings described herein are for illustrative purposes only and are not intended to limit the scope of the disclosure of the present invention in any way. Additionally, the shapes and proportional dimensions of the components in the drawings are only schematic and are used to assist in understanding the present invention, rather than specifically limiting the shapes and proportional dimensions of the components of the present invention. In the drawings: Figure 1 is a flowchart of the GNSS real-time spoofing interference detection method based on deep learning of the present invention.

[0024] Figure 2 is a structural diagram of the GNSS real-time spoofing interference detection system based on deep learning of the present invention.

[0025] Figure 3 is a diagram of an electronic device for the GNSS real-time spoofing interference detection method based on deep learning of the present invention.

[0026] Figure 4 is an overall detection flowchart.

[0027] Figure 5 is a comparison diagram of input features and the original signal.

[0028] Figure 6 is a comparison diagram of traditional LSTM model-based signal detection and residuals.

[0029] Figure 7 This is the signal detection and residual comparison diagram of the LSTM prediction model with an incremental learning mechanism of the present invention.

[0030] Figure 8 This is the real-time detection window diagram. Detailed implementation manners

[0031] In order to enable those skilled in the art of this technology to better understand the solution of the present invention, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative work shall fall within the protection scope of the present invention.

[0032] Embodiment 1 Refer to Figure 1 , a GNSS real-time spoofing interference detection method based on deep learning, includes the following steps: Collect the Doppler frequency shift signals of the GNSS receiver in real time to obtain an initial training data set; Construct an LSTM prediction model with an incremental learning mechanism, use the initial training data set, add Gaussian white noise to the LSTM prediction model with an incremental learning mechanism for model training to obtain a trained LSTM prediction model with an incremental learning mechanism. The LSTM prediction model with an incremental learning mechanism includes a double-layer LSTM network and a fully connected layer, and the double-layer LSTM network introduces a gating mechanism; After performing a sliding window process on the initial training data set, extract multi-dimensional feature vectors; input the multi-dimensional feature vectors into the trained LSTM prediction model with an incremental learning mechanism for prediction to obtain model prediction values; Calculate the residuals between the model prediction values and the true values of the Doppler frequency shift signals collected in real time. After excluding the warm-up period, calculate the standard deviation of the residuals, and dynamically set a threshold based on the standard deviation of the residuals; Calculate the absolute value of the residuals. When the absolute value of the residuals exceeds the threshold and the sampling points processed by the sliding window are within a preset interference time window, it is determined that a signal interference event has occurred.

[0033] The LSTM prediction model with an incremental learning mechanism in this embodiment can dynamically adjust model parameters and continuously optimize model performance as new data arrives, thereby more accurately predicting the changing trend of Doppler shift signals and improving the modeling ability for normal signal patterns. It avoids the problem that the parameters of the traditional LSTM model are fixed after training and are difficult to adapt to the dynamic changes of the GNSS signal environment. The double-layer LSTM network increases the depth of the model, enabling it to learn more complex signal features. At the same time, the introduced gating mechanism can effectively control the flow of information, avoid the problem of gradient vanishing or explosion, further enhance the model's ability to capture long-term dependence relationships, and improve the accuracy of prediction. The hierarchical update strategy with weight freezing is used to dynamically update the prediction model parameters, which not only retains the model's learning results for historical data but also can timely absorb the features of new data, avoiding drastic fluctuations in model parameters and enhancing the model's stability under different data distributions. After performing a sliding window process on the initial training dataset, multi-dimensional feature vectors are extracted. These feature vectors contain various information of the Doppler shift signal at different time scales, such as mean, variance, spectral features, etc. The use of multi-dimensional features can more comprehensively describe signal features, enabling the model to capture more subtle signal changes, thereby improving the detection sensitivity to spoofing interference signals. The method of this embodiment not only considers whether the absolute value of the residual exceeds the threshold but also requires that the sampling point is within a preset interference time window to determine that a signal interference event has occurred. This comprehensive judgment mechanism avoids false detections caused by accidental noise or signal fluctuations and also reduces missed detections caused by short-duration interference signals, improving the reliability of detection. Performing a sliding window process on the initial training dataset can adapt to the changes of GNSS signals under different time and space conditions. The sliding window can capture the local features of the signal in different time periods, enabling the model to better adapt to the complex and changing GNSS signal environment. The LSTM prediction model can quickly predict the input multi-dimensional feature vectors, calculate the residual sequence between the model prediction value and the true value, and dynamically set the threshold. The entire detection process can be completed in a short time, meeting the requirements of real-time spoofing interference detection of GNSS signals, being able to detect and respond to interference events in a timely manner, and ensuring the normal operation of the GNSS system.

[0034] See Figure 4 , the overall detection flowchart of the GNSS real-time spoofing interference detection method based on deep learning. The specific steps include: Step 1. Real-time collect the Doppler shift signals of the GNSS receiver with a sampling rate of 1000 Hz.

[0035] Step 2. Perform a sliding window process on the signal. The window length is set to 15 - 30 sampling points, and multi-dimensional feature vectors including instantaneous values, sliding means, standard deviations, linear trend coefficients, and fast Fourier transform frequency domain components are extracted. Step 2 specifically includes: The linear trend coefficient is extracted by least - squares fitting, and the specific implementation is as follows:

[0036] In the formula, trend is the linear trend coefficient, is the sliding window length, is the th signal sampling value within the window, is the time index within the window, i is an intermediate variable.

[0037] Step 3. Construct a prediction model including a double - layer LSTM network and a fully - connected layer, where the number of nodes in the LSTM hidden layer is 128 - 512. Step 3 specifically includes: The double - layer LSTM network constructed in the model is specifically implemented as follows: The input gate is expressed as:

[0038] In the formula, is the input gate activation vector, is the sigmoid activation function, is the weight matrix of the input state, is the weight matrix of the hidden state, is the weight matrix of the cell state, is the feature vector of the current input (including instantaneous value, mean, standard deviation, etc.), is the cell state at the previous moment.

[0039] The forget gate is expressed as:

[0040] In the formula, is the forget gate, is the hidden state of the model at the previous moment, , are the weight matrices of the current output and the current state of the forget gate respectively, is the bias term of the forget gate.

[0041] Cell state update:

[0042] In the formula, represents the candidate cell state, is the weight matrix of the current input to the candidate cell state, is the weight matrix of the hidden state to the candidate cell state, is the bias term of the candidate cell state, is the hyperbolic tangent activation function.

[0043] Finally:

[0044] In the formula, is the cell state, is the forget gate, represents the Hadamard product, is the cell state at the previous moment, is the input gate activation vector, represents the candidate cell state.

[0045] The output gate is expressed as:

[0046] In the formula, is the output gate, is the sigmoid activation function, is the weight matrix of the current input to the output gate, is the feature vector of the current input, is the weight matrix from the hidden state to the output gate, is the hidden state of the model at the previous moment, is the weight matrix from the cell state to the output gate, is the cell state, is the bias term of the output gate.

[0047] The hidden state is expressed as:

[0048] In the formula, represents the hidden state, is the output gate, represents the Hadamard product, is the hyperbolic tangent activation function, is the cell state.

[0049] The double-layer LSTM network in step 3 is constructed by the above formula, with an input dimension of 5 (corresponding to 5 features) and a hidden layer dimension of 256.

[0050] Step 4. Train the model using the initial training dataset, with the number of training epochs being 8 - 15 and the initial learning rate set to 0.0005 - 0.005.

[0051] Step 5. When new monitoring data is received, randomly select 5%-15% of the historical training data and mix it with the new data, and update the model parameters using a hierarchical learning rate strategy, where the learning rate of the LSTM layer is 0.1-0.3 times that of the fully connected layer, and the number of training epochs is 3-8. The new monitoring data refers to data different from that in the model training process. Step 5 adopts a hierarchical update strategy with weight freezing, specifically including: the weight update amount of the LSTM layer is limited within ±10% of the initial weight; the weight update amount of the fully connected layer is limited within ±20% of the initial weight; the gradient clipping technique is adopted, and the gradient norm threshold is set to 1.0-2.0. The hierarchical learning rate is an optimization strategy, and weight freezing is a parameter constraint mechanism. The two work together to solve the model degradation problem in the dynamic environment of GNSS signals.

[0052] The hierarchical update incremental learning rule is as follows:

[0053] The constraint conditions are:

[0054]

[0055] In the formula, is the updated parameter of the LSTM layer, is the learning rate of the LSTM layer, ▽ is the gradient mathematical symbol, is the parameter of the LSTM layer, is the loss function, is the updated parameter of the fully connected layer, is the learning rate of the fully connected layer, is the parameter of the fully connected layer, is the parameter value after initial training, is the parameter value of the LSTM layer after initial training, is the parameter value of the fully connected layer after initial training.

[0056] Step 6. Calculate the residual sequence between the model prediction value and the true value, dynamically calculate the residual standard deviation after excluding the warm-up period of the first 50-100 sampling points, and set the threshold to 5-10 times the standard deviation. The true value is the new monitoring data received. Each time the model predicts the next data based on the data within the window and then compares it with the received true data (taking the difference) to determine whether it is interfered. The dynamic threshold calculation in Step 6 introduces a moving average correction factor, which can be automatically adjusted according to the actual change of the signal, reducing the false detection and missed detection rates. The specific calculation formula is:

[0057] In the formula, is the dynamic threshold at the current moment, is the threshold at the previous moment, is the smoothing coefficient, is the sliding window length, i is an intermediate variable, is the historical residual value, is the mean of the window residual; The calculation formula for the historical residual value is:

[0058] In the formula, is the true value at the is the predicted value at the

[0059] The calculation formula for the mean of the window residual is: .

[0060] Step 7. When the absolute value of the residual of the sampling point exceeds the threshold and is within the preset interference time window, it is determined that a signal interference event has occurred.

[0061] The LSTM model with an incremental learning mechanism proposed in this embodiment aims to solve the problem that the traditional LSTM model has inaccurate predictions in long time series.

[0062] The specific implementation flowchart is as shown in Figure 4 .

[0063] In the data acquisition and preprocessing stage, the original data is the intermediate frequency Doppler shift signal output by the GNSS receiver, the sampling rate is set to fs = 1000Hz, and the data format is a one-dimensional time series { }, where t = 1, 2... T. In addition, Gaussian white noise is added in the training stage to enhance the robustness of the model. The Gaussian white noise formula is:

[0064] In the formula: is the noise, is the noise intensity coefficient, is the standard deviation of the original signal, is the standard normal distribution.

[0065] After the model is started, it first checks whether there is a model that has been initially trained. If there is, after loading the saved model, the real-time data stream module is immediately started to process new data in real time; if there is no saved model, initial training is performed, and the model and parameters are saved after the training is completed.

[0066] See Figure 5 , the comparison chart of the input features and the original signal, Figure 5 where a inFigure 5 b is the current value feature comparison, Figure 5 c is the moving average feature comparison, Figure 5 d is the standard deviation feature comparison, Figure 5 e is the trend slope feature comparison, Figure 5 f is the FFT (Fast Fourier Transform) main frequency amplitude feature comparison. When training the model using the initial training dataset, the number of training rounds is 8 - 15 epochs, and the initial learning rate is set to 0.0005 - 0.005. First, multi-dimensional feature extraction is performed on the initial training data. The sliding window length is set to 15 - 30 sampling points, and multi-dimensional feature vectors including instantaneous values, moving averages, standard deviations, linear trend coefficients, and fast Fourier transform frequency domain components are extracted. The feature extraction formulas are as follows: The instantaneous value is expressed as:

[0067] The mean value is expressed as:

[0068] The standard deviation is expressed as:

[0069] The trend term is expressed as:

[0070] The frequency domain component is expressed as:

[0071] In the above formulas is the time index within the window, is the number of data within the window, is the value of the nth data within the window.

[0072] Multidimensional feature vectors of the real-time data stream are extracted in the data buffer in the above manner, and the extracted feature vectors are input into the LSTM prediction model with an incremental learning mechanism for prediction. The LSTM prediction model with an incremental learning mechanism mainly consists of an input layer (including a 5-dimensional feature vector), an LSTM layer (with 256 hidden nodes), and an output layer (outputting a one-dimensional predicted value ).

[0073] When new data is received, the following steps are executed to achieve dynamic model update: Step 1: Data mixing: Take all samples from the new dataset ; randomly select 10% of the samples from the historical data ; merge the datasets: .

[0074] Step 2: Hierarchical parameter update: Set the learning rate of the LSTM layer to 0.1 times that of the fully connected layer; The optimization objective is to minimize the mean squared error:

[0075] The parameter update formula is:

[0076]

[0077] In the formula, is the true value at time is the predicted value at time are the parameters of the LSTM layer, are the parameters of the fully connected layer, is the learning rate of the LSTM layer, is the learning rate of the fully connected layer, is the loss function, , and the gradient clipping technique is used to limit the magnitude of parameter changes (gradient norm threshold 1.0).

[0078] Dynamic threshold interference determination: After excluding the warm-up period of the first 50 - 100 sampling points, the first 50 - 100 sampling points of the warm-up period are excluded and do not participate in subsequent residual calculations or threshold setting. Calculate the residual standard deviation and dynamically set the threshold:

[0079]

[0080] In the formula: is the dynamic threshold, is the residual standard deviation, is the adjustment coefficient, is the number of samples in the warm-up period, is the preset length of the interference detection time window, is the prediction residual, is the mean of the window residuals, and the calculation formula is:

[0081] In the formula, m is the length of the sliding window.

[0082] Interference determination logic: When the following conditions are met simultaneously, an interference alarm is triggered: (1) The absolute value of the residual at the sampling point exceeds the threshold; (2) The abnormal point is within the preset interference time window; (3) The statistical difference degree D ≥ 2.5.

[0083] See Figure 6, Traditional signal detection and residual comparison graph based on the LSTM model. Comparison between the GNSS signal prediction results of the traditional LSTM model and the real signal. The results show that when the prediction time step exceeds the length range of the training signal, the trend of the predicted signal flattens out, deviating significantly from the dynamic changes of the real signal. As the time step increases, the residual gradually increases, especially at the end of the long sequence, where the peak residual reaches more than 5 times that of the training stage. The reason is that the traditional LSTM model has limited long-term dependence ability due to the lack of a dynamic update mechanism and cannot adapt to the non-stationary characteristics of GNSS signals. This embodiment verifies the limitations of the static model in real-time detection and provides a basis for introducing an incremental learning strategy.

[0084] See Figure 7 , Signal detection and residual comparison graph of the improved LSTM prediction model with an incremental learning mechanism. Demonstrates the detection effect of the improved LSTM prediction model with an incremental learning mechanism in this embodiment. Key observations are as follows: The predicted curve highly coincides with the time-frequency characteristics of the real signal. Especially at the signal mutation points, the model quickly tracks the changes. The residual always remains at a low level, and the sudden residual at the interference points is accurately captured. The model continuously adapts to the dynamic changes of the signal, solving the problem of long-sequence prediction attenuation in the traditional LSTM model.

[0085] See Figure 8 , Real-time detection window graph. Demonstrates the real-time determination process within the dynamic detection window. In the figure, the model-predicted signal significantly deviates from the real value within the interference window. The residual continuously exceeds the dynamic threshold during the interference period, and the system triggers a spoofing interference alarm.

[0086] For the verification of the embodiment effects, see Table 1: Table 1 Final Operating Result Evaluation Table

[0087] The F1 score in Table 1 is an index used in statistics to measure the accuracy of the model.

[0088] After the program runs, by changing the sliding window length, the performance of the detection model is tested. The sliding window length ranges from 20 to 200, and the scenarios in the TEXBAT (Texas Spoofing Test Battery) dataset are used for testing. The detection accuracy is above 98%, and the response time is within 0.1 seconds, which can meet the subsequent engineering applications. The TEXBAT dataset is a spoofing interference dataset with multiple application scenarios released by the University of Texas in the United States and is one of the most widely used and systematically structured GNSS spoofing test sets currently.

[0089] In this embodiment, interference detection is achieved by using an LSTM prediction model with an incremental learning mechanism. The LSTM model is good at processing and predicting time series data and can capture the dynamic changes and long-term dependencies of signals over time through its built-in memory mechanism. This makes the LSTM model perform better than the DNN (Deep Neural Network) model in tasks related to time series correlation or signal change patterns. In addition, the LSTM network introduces a gating mechanism, including an input gate, a forget gate, and an output gate, which can selectively remember or forget past information. This is very helpful for identifying important features in time series patterns and ignoring noise. In this embodiment, the residual sequence between the model prediction value and the true value of the Doppler frequency shift signal collected in real time is calculated. After excluding the warm-up period, the standard deviation of the residuals is calculated and the threshold is set dynamically, and a moving average correction factor is introduced in the threshold calculation. The dynamic threshold can be automatically adjusted according to the actual changes in the signal, avoiding the problems of false detection or missed detection that may occur with a fixed threshold when the signal fluctuates greatly, and improving the accuracy of detection. Gaussian white noise is added to the model during the training process to simulate various interference factors that may exist in the actual environment. This training method makes the model more stable in the face of noise interference, can better extract effective signal features from the noise, and improves the adaptability and robustness of the model to actual GNSS signals. Part of the historical training data is randomly selected and mixed with the Doppler frequency shift signal collected in real time for prediction, enabling the model to timely incorporate new real-time data into consideration, quickly adapt to changes in the signal environment, and improve the adaptability of the model in complex environments.

[0090] The LSTM model can be well applied to different types of time series data, has strong adaptability, can handle the dynamic changes and non-linear relationships in signals, and can effectively capture and understand complex time series data patterns, long-term dependencies, and time-varying characteristics of interference signals. Especially in this embodiment, a method combining incremental learning with the LSTM model is adopted, enabling the model to be updated online and perform real-time signal processing.

[0091] Embodiment 2 See Figure 2 , a GNSS real-time spoofing interference detection system based on deep learning, including: A training data acquisition module, configured to collect the Doppler frequency shift signal of the GNSS receiver in real time to obtain an initial training data set; A prediction model training module, configured to construct an LSTM prediction model with an incremental learning mechanism, use the initial training data set, add Gaussian white noise to the LSTM prediction model with an incremental learning mechanism for model training, and obtain a trained LSTM prediction model with an incremental learning mechanism. The LSTM prediction model with an incremental learning mechanism includes a double-layer LSTM network and a fully connected layer, and the double-layer LSTM network introduces a gating mechanism; A prediction module, which is used to perform sliding window processing on the initial training data set and extract multi-dimensional feature vectors; input the multi-dimensional feature vectors into a trained LSTM prediction model with an incremental learning mechanism for prediction to obtain model prediction values; A dynamic threshold setting module, which is used to calculate the residual between the model prediction value and the true value of the Doppler shift signal collected in real time. After excluding the warm-up period, calculate the standard deviation of the residual, and dynamically set the threshold based on the standard deviation of the residual; A signal interference event determination module, which is used to calculate the absolute value of the residual. When the absolute value of the residual exceeds the threshold and the sampling points processed by the sliding window are within a preset interference time window, it is determined that a signal interference event has occurred.

[0092] Embodiment III Refer to Figure 3 , an electronic device, including a memory, a processor, and a computer program stored in the memory and operable on the processor. When the processor executes the computer program, it implements a GNSS real-time spoofing interference detection method based on deep learning.

[0093] Embodiment IV A computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, it implements a GNSS real-time spoofing interference detection method based on deep learning.

[0094] Those skilled in the art should understand that the embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present invention can take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk memories, read-only optical discs, optical memories, etc.) containing computer-usable program code.

[0095] The present invention is described with reference to the flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to the embodiments of the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram can be implemented by computer program instructions, and the combination of the processes and / or blocks in the flowchart and / or block diagram can also be implemented by computer program instructions. These computer program instructions can be provided to the processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing devices to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing devices generate a device for realizing the functions specified in one process Figure 1 one process or multiple processes and / or blocks Figure 1 one block or multiple blocks.

[0096] These computer program instructions can also be stored in a computer-readable memory that can direct a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer-readable memory produce an article of manufacture including an instruction device that implements the functions specified in one or more of the processes and / or blocks Figure 1 one or more of the processes and / or blocks Figure 1 specified in one or more of the blocks or blocks.

[0097] These computer program instructions can also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer-implemented process, so that the instructions executed on the computer or other programmable apparatus provide steps for implementing the functions specified in one or more of the processes and / or blocks Figure 1 one or more of the processes and / or blocks Figure 1 specified in one or more of the blocks or blocks. Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit them. Although the present invention has been described in detail with reference to the above embodiments, those of ordinary skill in the art should understand that: modifications or equivalent replacements can still be made to the specific embodiments of the present invention, and any modification or equivalent replacement that does not depart from the spirit and scope of the present invention should be covered by the protection scope of the present invention.

Claims

1. A GNSS real-time spoofing interference detection method based on deep learning, characterized in that It includes the following steps: Collect the Doppler frequency shift signals of the GNSS receiver in real time to obtain the initial training dataset; Construct an LSTM prediction model with an incremental learning mechanism. Use the initial training dataset to add Gaussian white noise to the LSTM prediction model with an incremental learning mechanism for model training to obtain the trained LSTM prediction model with an incremental learning mechanism. The LSTM prediction model with an incremental learning mechanism includes a double-layer LSTM network and a fully connected layer, and the double-layer LSTM network introduces a gating mechanism; After performing sliding window processing on the initial training dataset, extract multi-dimensional feature vectors; Input the multi-dimensional feature vectors into the trained LSTM prediction model with an incremental learning mechanism for prediction to obtain the model prediction value; Calculate the residual between the model prediction value and the true value of the Doppler frequency shift signal collected in real time. After excluding the warm-up period, calculate the standard deviation of the residual, and dynamically set the threshold based on the standard deviation of the residual; Calculate the absolute value of the residual. When the absolute value of the residual exceeds the threshold and the sampling points of the sliding window processing are within the preset interference time window, it is determined that a signal interference event has occurred.

2. The GNSS real-time spoofing interference detection method based on deep learning according to claim 1, wherein The double-layer LSTM network includes an input gate, a forget gate, cell state update, an output gate, and a hidden state. The multi-dimensional feature vectors include instantaneous values, sliding means, standard deviations, linear trend coefficients, and fast Fourier transform frequency domain components.

3. The GNSS real-time spoofing interference detection method based on deep learning according to claim 1, characterized in that The prediction in the step of inputting the multi-dimensional feature vectors into the trained LSTM prediction model with an incremental learning mechanism for prediction specifically is: Randomly select some historical training data and mix it with the Doppler frequency shift signal collected in real time, and dynamically update the prediction model parameters using a hierarchical update strategy with frozen weights.

4. The GNSS real-time spoofing interference detection method based on deep learning according to claim 3, wherein In the step of dynamically updating the prediction model parameters using a hierarchical update strategy with frozen weights, it includes a gradient norm threshold and a hierarchical weight update amount constraint. The gradient norm threshold and the hierarchical weight update amount constraint include: The gradient norm threshold range is set using the gradient clipping method; The weight update amount of the double-layer LSTM network is within the first threshold range of the initial weights; The weight update amount of the fully connected layer is within the second threshold range of the initial weights.

5. The GNSS real-time spoofing interference detection method based on deep learning according to claim 3, characterized in that, The hierarchical update strategy with frozen weights uses an incremental learning rule. The incremental learning rule specifically is: The constraint conditions of the incremental learning rule are: Among them, are the updated LSTM layer parameters, is the learning rate of the LSTM layer, ▽ is the gradient mathematical symbol, are the LSTM layer parameters, is the loss function, are the updated fully connected layer parameters, is the learning rate of the fully connected layer, are the fully connected layer parameters, is the parameter value after initial training, is the parameter value of the LSTM layer after initial training, is the parameter value of the fully connected layer after initial training.

6. The GNSS real-time spoofing interference detection method based on deep learning according to claim 1, wherein The dynamically set threshold introduces a sliding average correction factor, specifically: Among them, is the dynamic threshold at the current moment, is the smoothing coefficient, is the threshold at the previous moment, is the length of the sliding window, i is an intermediate variable, is the historical residual value, is the mean of the window residual.

7. The GNSS real-time spoofing interference detection method based on deep learning according to claim 6, wherein The calculation formula of the historical residual value specifically is: Among them, is the historical residual value, is the true value at time is the predicted value at time i is an intermediate variable; The calculation formula of the window residual mean specifically is: Among them, is the window residual mean, is the sliding window length.

8. A GNSS real-time spoofing interference detection system based on deep learning, characterized in that, It includes: A training data collection module, which is used to collect the Doppler frequency shift signals of the GNSS receiver in real time to obtain the initial training dataset; A prediction model training module, which is used to construct an LSTM prediction model with an incremental learning mechanism. Use the initial training dataset to add Gaussian white noise to the LSTM prediction model with an incremental learning mechanism for model training to obtain the trained LSTM prediction model with an incremental learning mechanism. The LSTM prediction model with an incremental learning mechanism includes a double-layer LSTM network and a fully connected layer, and the double-layer LSTM network introduces a gating mechanism; A prediction module, which is used to perform sliding window processing on the initial training dataset and then extract multi-dimensional feature vectors; Input the multi-dimensional feature vector into the trained LSTM prediction model with an incremental learning mechanism for prediction to obtain the model prediction value; A dynamic threshold setting module, which is used to calculate the residual between the model prediction value and the true value of the Doppler frequency shift signal collected in real time. After excluding the warm-up period, calculate the standard deviation of the residual, and dynamically set the threshold based on the standard deviation of the residual; A signal interference event determination module, which is used to calculate the absolute value of the residual. When the absolute value of the residual exceeds the threshold and the sampling points processed by the sliding window are within the preset interference time window, it is determined that a signal interference event has occurred.

9. An electronic device, characterized in that, It includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the deep learning-based GNSS real-time spoofing interference detection method described in any one of claims 1-7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, it implements the deep learning-based GNSS real-time spoofing interference detection method described in any one of claims 1-7.

Images