A GNSS spoofing detection and anti-spoofing navigation method based on a hybrid estimation model

Abstract

The application relates to the field of GNSS spoofing detection and anti-spoofing navigation, in particular to a GNSS spoofing detection and anti-spoofing navigation method based on a hybrid estimation model. The scheme comprises the following steps: constructing a hybrid estimation model, wherein the hybrid estimation model is composed of a time series convolutional neural network and an adaptive Kalman filter; training and predicting the hybrid estimation model; during the training process, taking the INS / GNSS integrated navigation result under the normal condition of satellite signals as an output label, and taking a position residual as a loss value to train the hybrid estimation model; during the prediction process, inputting the output value of an inertial navigation system into the trained hybrid estimation model; performing spoofing detection according to the Euclidean distance in a detection window; when the spoofing is successfully detected, shielding the global navigation satellite system signal, and recapturing a correct positioning signal. The application is suitable for the field of spoofing detection and anti-spoofing navigation.

Description

Technical Field

[0001] This invention relates to the field of GNSS spoofing detection and anti-spoofing navigation, and specifically to a GNSS spoofing detection and anti-spoofing navigation method based on a hybrid estimation model. Background Technology

[0002] Existing INS-assisted spoofing detection methods include:

[0003] Deception detection methods based on fusion anomalies use changes in innovation, residuals, and state variable statistical properties as detection indicators. Detection methods based on information consistency detect deception by comparing differences between velocity / position solutions or raw INS measurements (acceleration / angular velocity) between systems.

[0004] However, existing INS (Inertial Navigation System)-assisted GNSS (Global Navigation Satellite System) spoofing detection methods have the following main problems:

[0005] Deception detection methods based on fusion anomalies mainly rely on the statistical characteristics of filter innovation, residuals, and state variables for identification. However, similar methods are also used to identify sensor faults and measurement anomalies, making it impossible to strictly distinguish the specific source of the fault.

[0006] Methods based on navigation information consistency compare the raw INS data or navigation solutions with GNSS positioning results. However, INS errors can diverge rapidly within a short period, severely impacting the accuracy and reliability of deception detection methods. Therefore, such methods are difficult to apply in scenarios such as civilian vehicles and low-cost drones equipped with low-cost INS systems.

[0007] Introducing external sensors such as odometers, vision, and radar to improve the divergence of INS errors can indeed enhance detection performance, but this incurs additional costs and computational overhead.

[0008] Existing detection methods based on information consistency are ineffective at detecting small-scale, slow-drift spoofing. However, if such spoofing is not detected in the first instance, the spoofing signal may have already affected the results of the integrated navigation, causing the detection algorithm to fail.

[0009] Most existing methods focus only on deception detection itself, but lack a systematic solution to the emergency navigation problem during satellite signal reconstruction after successful deception detection. Summary of the Invention

[0010] The purpose of this invention is to overcome the shortcomings of the prior art and provide a GNSS deception detection and anti-deception navigation method based on a hybrid estimation model, which solves the problem that traditional low-cost INS errors diverge rapidly and are difficult to provide accurate deception detection references.

[0011] The present invention achieves the above objectives by adopting the following technical solution: The present invention provides a GNSS spoofing detection and anti-spoofing navigation method based on a hybrid estimation model, comprising the following steps:

[0012] S1. Construct a hybrid estimation model, which consists of a time-series convolutional neural network and an adaptive Kalman filter;

[0013] The inputs to the time-series convolutional neural network are: the specific force measurement value output by the accelerometer, the angular acceleration information output by the gyroscope, and the attitude measurement value and velocity measurement value in the navigation system calculated by the inertial navigation system.

[0014] The output of the time-series convolutional neural network is the increment of the position measurement value of the Global Navigation Satellite System, which is used to establish a mapping relationship between the output of the inertial navigation system and the position result of the Global Navigation Satellite System;

[0015] The adaptive Kalman filter employs an interactive adaptive Kalman filter to address the mismatch problem of measurement noise parameters and to fuse and estimate the output of the time-series convolutional neural network with the solution results of the inertial navigation system.

[0016] S2. Train and predict the mixture estimation model;

[0017] During training, the INS / GNSS integrated navigation results under normal satellite signal conditions are used as output labels, and the position residual is used as the loss value to train the hybrid estimation model.

[0018] During the prediction process, the output value of the inertial navigation system is input into the trained hybrid estimation model to obtain the predicted GNSS measurement value, which is used for subsequent comparison.

[0019] The hybrid estimation model utilizes its fitting ability to suppress the divergence of INS navigation errors, thereby providing a reference benchmark for the detection algorithm without adding additional sensors.

[0020] S3. Based on the predicted GNSS measurement values ​​output by the hybrid estimation model and the positioning results of the global navigation satellite system, calculate the Euclidean distance within the detection window;

[0021] The Euclidean distance within each window is converted into an adjustment parameter, the value of which is directly related to the value of the Euclidean distance.

[0022] When the value of the adjustment parameter exceeds the set value, continuous detection begins, and the adjustment parameters of subsequent windows are continuously accumulated until the accumulated adjustment parameter value exceeds the threshold and a deception alarm is given, or the accumulation stops when the value of the adjustment parameter drops below the set value.

[0023] S4. Once the spoofing is successfully detected, the system blocks the Global Navigation Satellite System signal and reacquires the correct positioning signal;

[0024] During satellite navigation signal reconstruction, the predicted output of the neural network in the hybrid estimation model and the measured values ​​of the inertial navigation system are used to perform combined navigation based on the adaptive Kalman filter to suppress the divergence of the inertial navigation system error until the satellite navigation system returns to normal.

[0025] Furthermore, the calculation process of the interactive adaptive Kalman filter is as follows:

[0026] Define filter model The filter model M contains r parallel sub-filters, each corresponding to different noise parameters. Each sub-filter independently performs state estimation, corresponding to the rth sub-filter. There are several models, and the state estimate and covariance matrix of the filter model at time k are defined as follows: and Then the model interaction state estimate and covariance matrix at time k-1 are expressed as:

[0027] ;

[0028] ;

[0029] in Indicates the first The sub-filter model for the first The conditional probability of each model. represent Deviation between the model state and the interaction state at any given time:

[0030] ;

[0031] The contribution of each sub-filter state in the interaction can be represented by normalized weights:

[0032] ;

[0033] In the formula, Represents the correlation coefficient. Representation Model exist The probability at any given moment;

[0034] The residual estimates of the interactive adaptive Kalman filter follow a Gaussian white noise sequence with zero mean and constant covariance;

[0035] Residuals of the filter model and its covariance matrix Represented as:

[0036] ;

[0037] ;

[0038] in Indicates the measured value. Represents the observation matrix. The covariance represents the measurement noise;

[0039] The posterior probability value of the filter model is expressed as:

[0040] ;

[0041] in represents the likelihood function before normalization, which follows a Gaussian likelihood distribution;

[0042] After the prediction update, the fused output of the filter model is represented as:

[0043] ;

[0044] .

[0045] The beneficial effects of this invention are as follows:

[0046] This invention designs a hybrid estimation model based on deep learning and adaptive filtering techniques, which solves the problem of rapid error divergence in traditional low-cost INS and difficulty in providing accurate deception detection references without adding additional sensors.

[0047] This invention uses the high-precision trajectory of the hybrid estimation model as a reference, compares the Euclidean distance between the real signal and the predicted value, and introduces a time-series window correlation mechanism to improve the sensitivity to small-scale, slow-change deception, while reducing false alarms caused by signal quality fluctuations.

[0048] When GNSS is determined to be spoofed and unusable, this invention uses a hybrid estimation optimization method to constrain INS errors, preventing a complete degradation to pure INS navigation and the resulting rapid divergence in positioning errors. This solves the emergency navigation problem during GNSS signal reconstruction after successful spoofing detection, a problem often overlooked by other detection methods. Attached Figure Description

[0049] Figure 1This is a flowchart of a GNSS spoofing detection and anti-spoofing navigation method based on a hybrid estimation model provided in an embodiment of the present invention;

[0050] Figure 2 This is a schematic diagram of the input time series data provided in an embodiment of the present invention;

[0051] Figure 3 This is a schematic diagram of the training process of the hybrid estimation model provided in an embodiment of the present invention;

[0052] Figure 4 This is a flowchart of deception detection by calculating Euclidean distance provided in an embodiment of the present invention;

[0053] Figure 5 This is a diagram illustrating the post-detection deception processing steps provided in an embodiment of the present invention.

[0054] Figure 6 This is a schematic diagram of the time-series convolutional neural network structure provided in an embodiment of the present invention. Detailed Implementation

[0055] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings.

[0056] This invention provides a GNSS spoofing detection and anti-spoofing navigation method based on a hybrid estimation model, such as... Figure 1 As shown, it includes:

[0057] S1. Construct a hybrid estimation model, which consists of a time-series convolutional neural network and an adaptive Kalman filter.

[0058] Time series convolutional neural network structure as follows Figure 6 As shown, the inputs to the time series convolutional neural network are: the specific force measurement value output by the accelerometer, the angular acceleration information output by the gyroscope, and the attitude measurement value and velocity measurement value in the navigation system calculated by the inertial navigation system.

[0059] The output of the time-series convolutional neural network is the increment of the Global Navigation Satellite System (GNSS) position measurement, used to establish a mapping relationship between the inertial navigation system output and the GNSS position result; the input time-series data is as follows: Figure 2 As shown.

[0060] An interactive adaptive Kalman filter is used to handle the mismatch problem of measurement noise parameters and to fuse the output of the time series convolutional neural network with the solution results of the inertial navigation system for estimation.

[0061] The calculation process of the interactive adaptive Kalman filter is as follows:

[0062] Define filter model The filter model M contains r parallel sub-filters, each corresponding to different noise parameters. Each sub-filter independently performs state estimation, corresponding to the rth sub-filter. There are several models, and the state estimate and covariance matrix of the filter model at time k are defined as follows: and Then the model interaction state estimate and covariance matrix at time k-1 are expressed as:

[0063] ;

[0064] ;

[0065] in Indicates the first The sub-filter model for the first The conditional probability of each model. represent Deviation between the model state and the interaction state at any given time:

[0066] ;

[0067] The contribution of each sub-filter state in the interaction can be represented by normalized weights:

[0068] ;

[0069] In the formula, Represents the correlation coefficient. Representation Model exist The probability at any given moment;

[0070] The residual estimates of the interactive adaptive Kalman filter follow a Gaussian white noise sequence with zero mean and constant covariance;

[0071] Residuals of the filter model and its covariance matrix It can be represented as:

[0072] ;

[0073] ;

[0074] in Indicates the measured value. Represents the observation matrix. This represents the covariance of the measurement noise.

[0075] The posterior probability value of the filter model is expressed as:

[0076] ;

[0077] in This represents the likelihood function before normalization, which follows a Gaussian likelihood distribution.

[0078] After the prediction update, the fused output of the filter model is represented as:

[0079] ;

[0080] .

[0081] S2. Train and predict the mixture estimation model;

[0082] like Figure 3 As shown, during the training process, the INS / GNSS integrated navigation results under normal satellite signal conditions are used as the output label, and the position residual is used as the loss value to train the hybrid estimation model;

[0083] During the prediction process, the output value of the inertial navigation system is input into the trained hybrid estimation model to obtain the predicted GNSS measurement value, which is used for subsequent comparison.

[0084] The hybrid estimation model uses its fitting ability to suppress the divergence of INS navigation errors, thereby providing a reference benchmark for the detection algorithm without adding additional sensors.

[0085] Figure 3 middle, and These represent the measurement outputs of the accelerometer and gyroscope, respectively. This represents the pose result output by INS. This indicates the speed value output by the INS navigation system. The pseudo GNSS measurement increments predicted by the neural network model are summed to obtain the pseudo GNSS measurement values. The position prediction output value of the hybrid optimization model is obtained after integrated navigation. .

[0086] S3. Based on the GNSS measurement values ​​output by the hybrid estimation model and the positioning results of the global navigation satellite system, calculate the Euclidean distance within the detection window;

[0087] like Figure 4 As shown, the Euclidean distance within each window is converted into an adjustment parameter, the value of which is directly related to the value of the Euclidean distance.

[0088] When the value of the adjusted parameter exceeds the set value, continuous detection begins, and the adjusted parameters in subsequent windows are continuously accumulated until the accumulated value of the adjusted parameter exceeds the threshold, at which point a false alarm is issued, or the accumulation stops when the value of the adjusted parameter drops below the set value.

[0089] S4. Once the spoofing is successfully detected, the system blocks the Global Navigation Satellite System signal and reacquires the correct positioning signal;

[0090] Once spoofing is successfully detected, the system will block GNSS satellite signals and reacquire the correct positioning signal. During GNSS reconstruction, due to the lack of measurement information for correction, the positioning error of the low-cost inertial navigation system (INS) will diverge rapidly, making navigation difficult. To address this issue, this invention uses the predicted output of a neural network and the INS measurements based on an adaptive Kalman algorithm for combined navigation to suppress the rapid divergence of INS errors until the satellite navigation system returns to normal. This function meets the emergency navigation needs in the event of satellite navigation failure after successful GNSS spoofing detection, thus better addressing the potential hazards caused by spoofing. The processing procedure is as follows: Figure 5 As shown. Figure 5 middle, , , Represents the attitude, velocity, and position results output by the inertial navigation system. , and The state error of the inertial navigation system estimated by the filter is used to obtain the navigation output of the integrated navigation system after compensation.

[0091] The above description is merely a preferred embodiment of the present invention. It should be understood that the present invention is not limited to the forms disclosed herein and should not be construed as excluding other embodiments. It can be used in various other combinations, modifications, and environments, and can be altered within the scope of the concept described herein through the above teachings or related technologies or knowledge. Modifications and variations made by those skilled in the art that do not depart from the spirit and scope of the present invention should be within the protection scope of the appended claims.

Claims

1. A GNSS spoofing detection and anti-spoofing navigation method based on a hybrid estimation model, characterized in that, Includes the following steps: S1. Construct a hybrid estimation model, which consists of a time-series convolutional neural network and an adaptive Kalman filter; The inputs to the time-series convolutional neural network are: the specific force measurement value output by the accelerometer, the angular acceleration information output by the gyroscope, and the attitude measurement value and velocity measurement value in the navigation system calculated by the inertial navigation system. The output of the time-series convolutional neural network is the increment of the position measurement value of the Global Navigation Satellite System, which is used to establish a mapping relationship between the output of the inertial navigation system and the position result of the Global Navigation Satellite System; The adaptive Kalman filter employs an interactive adaptive Kalman filter to address the mismatch problem of measurement noise parameters and to fuse and estimate the output of the time-series convolutional neural network with the solution results of the inertial navigation system. S2. Train and predict the mixture estimation model; During training, the INS / GNSS integrated navigation results under normal satellite signal conditions are used as output labels, and the position residual is used as the loss value to train the hybrid estimation model. During the prediction process, the output value of the inertial navigation system is input into the trained hybrid estimation model to obtain the predicted GNSS measurement value, which is used for subsequent comparison. The hybrid estimation model utilizes its fitting ability to suppress the divergence of INS navigation errors, thereby providing a reference benchmark for the detection algorithm without adding additional sensors. S3. Based on the GNSS measurement values ​​output by the hybrid estimation model and the positioning results of the global navigation satellite system, calculate the Euclidean distance within the detection window; The Euclidean distance within each window is converted into an adjustment parameter, the value of which is directly related to the value of the Euclidean distance. When the value of the adjustment parameter exceeds the set value, continuous detection begins, and the adjustment parameters of subsequent windows are continuously accumulated until the accumulated adjustment parameter value exceeds the threshold and a deception alarm is given, or the accumulation stops when the value of the adjustment parameter drops below the set value. S4. Once the spoofing is successfully detected, the system blocks the Global Navigation Satellite System signal and reacquires the correct positioning signal; During satellite navigation signal reconstruction, the predicted output of the neural network in the hybrid estimation model and the measured values ​​of the inertial navigation system are used to perform combined navigation based on the adaptive Kalman filter to suppress the divergence of the inertial navigation system error until the satellite navigation system returns to normal.

2. The GNSS spoofing detection and anti-spoofing navigation method based on a hybrid estimation model according to claim 1, characterized in that, The calculation process of the interactive adaptive Kalman filter is as follows: Define filter model The filter model M contains r parallel sub-filters, each corresponding to different noise parameters. Each sub-filter independently performs state estimation, corresponding to the rth sub-filter. There are several models, and the state estimate and covariance matrix of the filter model at time k are defined as follows: and Then the model interaction state estimate and covariance matrix at time k-1 are expressed as: ； ； in Indicates the first The sub-filter model for the first The conditional probability of each model. represent Deviation between the model state and the interaction state at any given time: ； The contribution of each sub-filter state in the interaction can be represented by normalized weights: ； In the formula, Represents the correlation coefficient. Representation Model exist The probability at any given moment; The residual estimates of the interactive adaptive Kalman filter follow a Gaussian white noise sequence with zero mean and constant covariance; Residuals of the filter model and its covariance matrix Represented as: ； ； in Indicates the measured value. Represents the observation matrix. The covariance represents the measurement noise; The posterior probability value of the filter model is expressed as: ； in represents the likelihood function before normalization, which follows a Gaussian likelihood distribution; After the prediction update, the fused output of the filter model is represented as: ； 。

Images