A Tightly Integrated Navigation Fault Detection Method Based on Mutual Information

Abstract

The present invention relates to a mutual information-based tight integrated navigation fault detection method, comprising: establishing a discrete state equation and an observation equation of INS error of a GNSS / INS tight integrated navigation system; using a generalized maximum correlation entropy Kalman filter to perform INS recursive error estimation and recursive error correction; collecting a pseudorange time series and a pseudorange rate time series in filtered new information data of a single satellite for normalization; establishing a standard distribution time series of pseudorange and pseudorange rate, respectively performing mutual information operations with the normalized pseudorange time series and pseudorange rate time series to obtain a pseudorange MI matrix and a pseudorange rate MI matrix; establishing a data anomaly detection model based on a residual attention network, taking the pseudorange MI matrix and the pseudorange rate MI matrix with normal or abnormal labels as inputs for data normal or abnormal classification training; using the trained data anomaly detection model to detect single satellite faults in the positioning process of the integrated navigation system. The present invention improves the fault detection accuracy.

Description

Technical Field

[0001] The invention relates to the technical field of integrated navigation, and in particular to a tight integrated navigation fault detection method based on mutual information. Background Art

[0002] With the development of technologies such as drones and autonomous driving, the demand for accuracy and reliability of navigation systems continues to increase. GPS / INS tightly integrated navigation has received more and more attention due to its high positioning accuracy and stability. In order to ensure the positioning accuracy, integrity and reliability of the system, timely fault detection and troubleshooting are of great significance.

[0003] The traditional methods for integrated navigation fault detection are mainly model methods, such as residual chi-square test, autonomous integrity monitoring extrapolation method, etc. The above model-based fault detection methods will be restricted in terms of accuracy and real-time performance of fault identification due to the limitations of the model itself. Summary of the invention

[0004] In view of the above analysis, the present invention aims to disclose a tightly integrated navigation fault detection method based on mutual information, and to improve the detection accuracy of satellite faults in the tightly integrated navigation system.

[0005] The present invention discloses a tight combination navigation fault detection method based on mutual information, comprising:

[0006] Step S1, establishing the discrete state equation and observation equation of the INS error of the GNSS / INS tightly integrated navigation system; using the generalized maximum relevant entropy Kalman filter to perform INS recursive error estimation and recursive error correction;

[0007] Step S2, collecting the pseudorange time series and pseudorange rate time series in the filtered new information data of a single satellite and normalizing them;

[0008] Step S3, establishing a standard distribution time series of pseudorange and pseudorange rate, and performing mutual information operation with the normalized pseudorange time series and pseudorange rate time series to obtain a pseudorange MI matrix and a pseudorange rate MI matrix;

[0009] Step S4, establishing a data anomaly detection model based on the residual attention network, taking the pseudo-range MI matrix and the pseudo-range rate MI matrix with normal or abnormal labels as input to perform data normal or abnormal classification training;

[0010] Step S5: Use the trained data anomaly detection model to detect single-satellite faults during the positioning process of the integrated navigation system.

[0011] Furthermore, the new information data γ at time k is obtained by using the generalized maximum correlation entropy Kalman filtering process k for:

[0012]

[0013] ρ INS , are the pseudorange and pseudorange rate calculated by INS respectively; ρ GNSS , are the pseudorange and pseudorange rate calculated by a single GNSS satellite respectively; H k is the observation matrix; is the one-step estimate of the error state;

[0014] Take a period of continuous length l of new information data to form a vector Γ=[γ1,...,γ i ,...,γ l ];

[0015] Extract the first and second rows of the vector Γ to obtain the pseudorange time series Γ1′ and the pseudorange rate time series Γ2′ respectively;

[0016] After normalization, the normalized pseudorange time series is obtained and normalized pseudorange rate time series

[0017] Furthermore, from the normalized, zero-mean Gaussian-distributed, randomly generated time series with the same length as the pseudorange time series and pseudorange rate time series, the time series distribution that is closest to the actual satellite data when it is normal is selected through experimental verification as the standard distribution time series of pseudorange and pseudorange rate.

[0018] Further, the data anomaly detection model based on the residual attention network includes: a first input end, a first convolutional layer, a first attention mechanism and a residual learning module, a second input end, a second convolutional layer, a second attention mechanism and a residual learning module, a fully connected layer and an output end;

[0019] The first input terminal, the first convolutional layer, the first attention mechanism and the residual learning module are sequentially connected, and the second input terminal, the second convolutional layer, the second attention mechanism and the residual learning module are sequentially connected;

[0020] The output ends of the first attention mechanism and residual learning module and the second attention mechanism and residual learning module are connected to a fully connected layer; the fully connected layer is connected to the output end;

[0021] The first input terminal is connected to the pseudo-range MI matrix with normal or abnormal labels; the second input terminal is connected to the pseudo-range rate MI matrix with normal or abnormal labels; and the output terminal outputs the classification result of the data.

[0022] Furthermore, the first attention mechanism and residual learning module and the second attention mechanism and residual learning module have the same network structure, both comprising a first full residual module, an attention module, and a second full residual module connected sequentially;

[0023] Among them, the first full residual module extracts high-level features from the input features; and provides them as input to the attention module, and the attention module develops specific low-level features as input to the second residual block to extract and output high-level features.

[0024] Furthermore, in the attention module, the trunk branch T(x) contains two stacked residual blocks and performs feature extraction; the mask branch M(x) is a bottom-up structure consisting of downsampling using maximum pooling and upsampling using bilinear interpolation; to achieve fast feedforward and feedback attention actions.

[0025] Furthermore, the first full residual module and the second full residual module have the same structure, both of which are network structures obtained by superimposing a ReLU activation layer, a batch normalization layer and a convolutional layer twice.

[0026] Furthermore, during the model training process of step S4,

[0027] The input is the pseudorange MI matrix and pseudorange rate MI matrix with normal or abnormal labels; the normal data label is set to 0, and the abnormal data label is set to 1;

[0028] The loss function is a binary classification cross entropy function, and the model parameters are optimized by gradient descent method;

[0029] Set the number of training cycles to M times, cycle through the optimal parameters of the training model, calculate the loss and accuracy, and finally save the model with the least loss and the highest accuracy as the trained model.

[0030] Furthermore, in the detection of a single-star fault in step S5, the following steps are included:

[0031] 1) During the test, a fixed time series length is used, and the pseudo-range time series of a single satellite is updated with a step size of 1, so that the end time of the pseudo-range time series is always the vector Γ at the current time t t =[γ1,...,γ i ,...,γ l ];

[0032] 2) Extract the pseudorange time series and pseudorange rate time series at the current time t and normalize them to obtain the normalized pseudorange time series at the current time t and pseudorange time series

[0033] 3) Use the standard distribution time series of pseudorange and pseudorange rate to compare with the normalized pseudorange time series and pseudorange rate time series Perform mutual information calculation to obtain the pseudorange MI matrix and pseudorange rate MI matrix at the current time t;

[0034] 4) The pseudorange MI matrix and pseudorange rate MI matrix at the current time t are input into the trained data anomaly detection model based on residual attention network for single satellite fault detection.

[0035] Furthermore, in step S1, the process of using generalized correlation entropy Kalman filtering to perform INS recursive error estimation includes:

[0036] 1) Perform one-step state prediction to obtain one-step estimation at time k The covariance matrix P(k|k-1) is:

[0037] 2) One-step estimation based on k moment The covariance matrix P(k|k-1) of and the covariance matrix R(k) of the observation noise are decomposed by Cholesky;

[0038] 3) Based on the Cholesky decomposition results, one-step estimation based on k moments Perform iterative calculation of the filter residual e(k,t);

[0039] 4) According to the filtering residual of t iterations, the generalized correlation entropy weighted value of each observation channel of t iterations is calculated to form a generalized correlation entropy weighted matrix

[0040] 5) Using the generalized correlation entropy weighted matrix Calculate the Kalman gain K(k,t);

[0041] 6) According to the Kalman gain K(k,t), the state is estimated to obtain the t+1th iteration value When the difference between the t+1th iteration value and the tth iteration value is greater than the threshold, return to step 3) and execute again; otherwise, the best estimated value of the state quantity k at the current moment is obtained.

[0042] The invention can achieve one of the following beneficial effects:

[0043] The method disclosed in the present invention combines pseudorange and pseudorange rate time series to capture abnormal characteristics of signals from two perspectives: slow change and sudden change, thereby improving the accuracy of fault detection.

[0044] A data method based on mutual information transformation is used to reveal the characteristics of each abnormal type, which helps to improve classification performance;

[0045] The data anomaly detection model based on residual attention network can reduce gradient vanishing and improve prediction performance. Compared with other commonly used methods, the trained model has better classification performance;

[0046] And the generalized maximum relevant entropy Kalman filter is used to improve the input data and reduce the non-Gaussian noise in the data, so as to distinguish faults from non-faults more accurately.

[0047] In the present invention, each satellite independently performs a fault determination process, and satellites determined to be faulty can be excluded from the navigation system. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] The drawings are only for the purpose of illustrating particular embodiments and are not to be considered limiting of the present invention. Like reference symbols denote like components throughout the drawings.

[0049] Figure 1 is a flow chart of a tightly integrated navigation fault detection method based on mutual information in an embodiment of the present invention;

[0050] Figure 2 2 is a diagram showing the architecture of a data anomaly detection model based on a residual attention network in an embodiment of the present invention;

[0051] Figure 3 Schematic diagram of the attention mechanism and residual learning module in an embodiment of the present invention;

[0052] Figure 4 is a diagram of the attention module architecture in an embodiment of the present invention;

[0053] Figure 5 2 is a diagram of the architecture of the full residual module in an embodiment of the present invention. DETAILED DESCRIPTION

[0054] The preferred embodiments of the present invention will be described in detail below in conjunction with the accompanying drawings, wherein the accompanying drawings constitute a part of this application and are used to illustrate the principles of the present invention together with the embodiments of the present invention.

[0055] An embodiment of the present invention discloses a tight integration navigation fault detection method based on mutual information, such as Figure 1 As shown, including:

[0056] Step S1, establishing the discrete state equation and observation equation of the INS error of the GNSS / INS tightly integrated navigation system; using the generalized maximum relevant entropy Kalman filter to perform INS recursive error estimation and recursive error correction;

[0057] Step S2, collecting the pseudorange time series and pseudorange rate time series in the filtered new information data of a single satellite and normalizing them;

[0058] Step S3, establishing a standard distribution time series of pseudorange and pseudorange rate, and performing mutual information operation with the normalized pseudorange time series and pseudorange rate time series to obtain a pseudorange MI matrix and a pseudorange rate MI matrix;

[0059] Step S4, establishing a data anomaly detection model based on the residual attention network, taking the pseudo-range MI matrix and the pseudo-range rate MI matrix with normal or abnormal labels as input to perform data normal or abnormal classification training;

[0060] Step S5: Use the trained data anomaly detection model to detect single-satellite faults during the positioning process of the integrated navigation system.

[0061] Specifically, in step S1, the state vector of the established GNSS / INS tightly integrated navigation system is:

[0062]

[0063] in, φ, b g 、b a 、s g and a They are the position vector error, velocity error, misalignment angle, gyroscope bias, accelerometer bias, gyroscope scale factor error and accelerometer scale factor error of the INS state recursion result in the ECEF coordinate system;

[0064] According to the error disturbance theory, the discrete state equation of INS error is established as:

[0065] X k =Φ k|k-1 X k-1 +ω k-1

[0066] X k , X k-1 are the state vectors at time k and k-1 respectively; ω k-1 is the system noise at time k-1, Φ k|k-1 is the state transfer matrix;

[0067] Among them, the state transfer matrix Φ k|k-1 for:

[0068]

[0069] Where Δt is the discrete time interval of the system, is the Earth's rotation angular velocity vector, is the direction cosine matrix from the carrier coordinate system b to the ECEF coordinate system, f b is the specific force output by the carrier accelerometer, T bg , Tba 、T sg 、T sa The gyro bias error, accelerometer bias error, gyro scale factor error, and accelerometer scale factor error are modeled as the relevant time parameters of the first-order Markov process respectively;

[0070] According to the error perturbation theory, the discrete observation equation of INS error is established as:

[0071] Z k =H k X k +v k

[0072] Among them, Z k is the observed quantity at time k, ρ INS , are the pseudorange and pseudorange rate calculated by INS respectively; ρ GNSS , are the pseudorange and pseudorange rate calculated by a single GNSS satellite respectively; v k is the observation noise at time k; H k is the observation matrix at time k, and the optimal estimate is obtained from the known state at time k-1 and are calculated from the observable satellite positions at time k solved from the ephemeris solution and are known quantities.

[0073] The relationship between the observable satellite pseudo-range observation and the INS recursive state error (i.e., state quantity X) is established through the observation equation; that is, the observation at time k:

[0074] Specifically, in step S1, the process of using generalized correlation entropy Kalman filtering to perform INS recursive error estimation includes:

[0075] 1) Perform one-step state prediction to obtain one-step estimation at time k The covariance matrix P(k|k-1) is:

[0076]

[0077] is the optimal estimate of the error state at time k-1, For use and the state transfer matrix Φ k|k-1 The one-step estimate of the error state at time k is calculated, and P(k|k-1) is the one-step estimate The covariance matrix of , P(k-1|k-1) is the estimation error covariance matrix at time k-1, and Q(k-1) is the system noise covariance matrix at time k-1;.

[0078] Here and P(k|k-1) will be used to calculate the optimal estimate of the error state at time k. When the system starts, the first moment k=1 is considered, and the moment k-1 corresponds to the moment 0. The state at the moment 0 is the preset initial state. The determination of the initial state x(0) and the corresponding Q(0) and P(0|0) depends on the characteristics of the adopted physical system and prior knowledge.

[0079] 2) One-step estimation based on k moment The covariance matrix P(k|k-1) of and the covariance matrix R(k) of the observation noise are decomposed by Cholesky;

[0080]

[0081] R(k) is the covariance matrix of the observation noise in the observation equation.

[0082] 3) Based on the Cholesky decomposition results, one-step estimation based on k moments Perform iterative calculation of the filter residual e(k,t);

[0083] e(k,t)=D(k)-W(k)x(k,t);

[0084] in, z(k) is the observed value at time k;

[0085] H k is the observation matrix at time k;

[0086] When step 3) is executed for the first time at the same time, x(k, t) takes the value in step 1) Then the calculation is performed according to the iteration value; t represents the number of iterations, and t=1 at the initial time.

[0087] 4) According to the filtering residual of t iterations, the generalized correlation entropy weighted value of each observation channel of t iterations is calculated to form a generalized correlation entropy weighted matrix

[0088]

[0089] h i (e i (k,t))=|e i (k,t)| α-2 exp(-β -α |e i (k,t) α |)

[0090] β is the generalized Gaussian kernel function h of the i-th observation channel i (ei (k, t)) is the kernel bandwidth, α is the shape parameter, and when the filter robustness is concerned, h i (e i The α parameter in (k, t) is set to 2, which is the form of Gaussian distribution, and the β parameter is set to 0.1 to make it robust.

[0091] When using the optimal estimation filter under the maximum generalized correlation entropy criterion, the Kalman gain needs to consider the correlation entropy weighting based on the observation noise covariance. The weight h i (e i (k, t)) has a first-order decay in the interval with small residuals, and a high-order decay in the interval with large residuals and decays to 0 quickly. Compared with the traditional Kalman filter that only uses covariance as a fixed weight, GMCKF has excellent anti-error capability for observations with gross error interference.

[0092] 5) Using the generalized correlation entropy weighted matrix Calculate the Kalman gain K(k,t);

[0093]

[0094] in, is the observation error covariance matrix of each observation channel weighted by relevant entropy;

[0095] 6) According to the Kalman gain K(k,t), the state is estimated to obtain the t+1th iteration value When the difference between the t+1th iteration value and the tth iteration value is greater than the threshold, return to step 3) and execute again; otherwise, the best estimated value of the state quantity k at the current moment is obtained.

[0096] Specifically, It is the iteration value obtained at the t+1th iteration, which is calculated from the result of the tth iteration.

[0097] At this time, if:

[0098]

[0099] Jump to step 3) and re-execute, otherwise get the best estimate of the state quantity at the current moment ε is the set threshold.

[0100] The estimated error value is used to feedback correct the recursive carrier motion state.

[0101] Use the best estimate of the state at time k In φ correction pure inertial navigation state recursion state: position speed Direction cosine matrix Get the position at the current time (assuming it is time k) after the estimated error correction speed Direction cosine matrix

[0102]

[0103] Among them, the pure inertial navigation state recursive state: position speed Direction cosine matrix The conventional pure inertial navigation state recursion method can be used to obtain the state, and the use of the conventional pure inertial navigation state recursion method to obtain the recursive state does not affect the protection scope of the present invention.

[0104] In the preferred solution, the generalized Gaussian kernel function h of each observation channel is adapted to the observation noise that varies dynamically with the carrier motion. i (e i (k,t)) with shape parameter α i Dynamically adjust the generalized Gaussian kernel function h of each observation channel i (e i The shape characteristics of (k,t)) match the observation noise and autonomously adapt to the changes of observation noise.

[0105] Specifically, in step S2, the new information data γ obtained by the filtering process k for:

[0106]

[0107] ρ INS , are the pseudorange and pseudorange rate calculated by INS respectively; ρ GNSS , are the pseudorange and pseudorange rate calculated by a single GNSS satellite respectively; H k is the observation matrix; is a one-step estimate of the error state.

[0108] For a single observable satellite, take a continuous period of time with a length of l to form a vector Γ = [γ1,...,γ i ,...,γ l ]; Preferably, l=20.

[0109] Extracting the first and second rows of vector Γ yields:

[0110] Pseudorange time series Γ1′=[γ1[1],...,γi [1],...,γ l [1]];

[0111] Pseudorange rate time series Γ2′=[γ1[2],...,γ i [2],...,γ l [2]];

[0112] The pseudorange time series Γ1′ is normalized to obtain the normalized pseudorange time series

[0113]

[0114] in,

[0115] The pseudorange rate time series Γ2′ is normalized to obtain the normalized pseudorange rate time series

[0116]

[0117] in,

[0118] Optionally, in step S3, from a plurality of normalized time series that conform to the zero-mean Gaussian distribution and are randomly generated and have the same length as the pseudorange time series and the pseudorange rate time series, the time series distribution that is closest to the distribution when the actual satellite data is normal is selected through experimental verification as the standard distribution time series of the pseudorange and pseudorange rate.

[0119] Optionally, historically collected satellite data may be used to extract the pseudorange time series and pseudorange rate time series when the satellite is determined to be normal as the standard distribution time series of the pseudorange and pseudorange rate.

[0120] In another optional scheme, the above two methods for determining the standard distribution time series of pseudoranges and pseudorange rates can be combined together. For satellites that can be determined to be operating normally, historical data are used to establish the standard distribution time series of pseudoranges and pseudorange rates; otherwise, another method is used to establish the standard distribution time series of pseudoranges and pseudorange rates.

[0121] In step S3, the obtained pseudorange MI matrix or pseudorange rate MI matrix is ​​expressed as:

[0122]

[0123] in,

[0124] h is the scanning window, which is a key parameter affecting the estimation accuracy. A small scanning window will cause the result to contain considerable noise, while a large scanning window will lose some information.

[0125] The optimal length of the scanning window is h opt =[4 / (d+2)] 1 / (d+4) n -1 / (d+4) , where d represents the dimension of data, in this embodiment, d=2, n=20. When l=20, the pseudorange MI matrix or pseudorange rate MI matrix is ​​a 20×20 parameter matrix.

[0126] Specifically, Figure 2 As shown, the data anomaly detection model based on the residual attention network established in step S4 includes: a first input end, a first convolutional layer, a first attention mechanism and a residual learning module, a second input end, a second convolutional layer, a second attention mechanism and a residual learning module, a fully connected layer and an output end;

[0127] The first input terminal, the first convolutional layer, the first attention mechanism and the residual learning module are sequentially connected, and the second input terminal, the second convolutional layer, the second attention mechanism and the residual learning module are sequentially connected;

[0128] The output ends of the first attention mechanism and residual learning module and the second attention mechanism and residual learning module are connected to a fully connected layer; the fully connected layer is connected to the output end;

[0129] The first input terminal is connected to the pseudo-range MI matrix with normal or abnormal labels; the second input terminal is connected to the pseudo-range rate MI matrix with normal or abnormal labels; and the output terminal outputs the classification result of the data.

[0130] The introduction of full residual blocks and attention mechanisms in the model can reduce gradient vanishing and improve prediction performance. Figure 3 As shown, the first attention mechanism and residual learning module and the second attention mechanism and residual learning module have the same network structure, both of which include a first full residual module, an attention module, and a second full residual module connected sequentially;

[0131] Among them, the first full residual module extracts high-level features from the input features; and provides them as input to the attention module, and the attention module develops specific low-level features as input to the second residual block to extract and output high-level features.

[0132] By using the attention module and the full residual module alternately, the input data is divided into several high-level features and low-level features, which can be classified more effectively. After being processed by the attention module and the full residual module, the features of the two-channel input are obtained, the features are fused through the fully connected layer, and the classification results are obtained through the softmax function. In terms of processing methods, the two-channel input has the advantage of high detection accuracy compared to using pseudorange as input data for fault detection alone.

[0133] Preferably, Figure 4 As shown in the figure, in the attention module, the trunk branch T(x) contains two stacked residual blocks and performs feature extraction; the mask branch M(x) is a bottom-up structure consisting of downsampling using maximum pooling and upsampling using bilinear interpolation; to achieve fast feedforward and feedback attention actions.

[0134] like Figure 5 As shown in the figure, the structures of the first full residual module and the second full residual module are the same, both of which are network structures obtained by superimposing the ReLU activation layer, the batch normalization layer and the convolution layer twice. A batch normalization layer is added to the network, and an adaptive reparameterization algorithm is used to enhance the training stability. Before the convolution layer, the nonlinear ReLU activation function truncates all negative values. After the batch normalization layer, a superposition sequence of ReLU activation layers is added, and the classification accuracy is higher.

[0135] The following table shows the model parameters using the MI matrix (20×20) as an example. In the model architecture, the input is first processed by the convolutional layer, then the full residual block and the attention module are used to extract features, and then the maximum pooling layer is used to downsample the input by taking the maximum value on a square window (10×10). Finally, the fully connected layer and the softmax function are used to classify the output into 2 categories.

[0136] Table 1 Model parameters

[0137]

[0138] Specifically, during the model training process of step S4,

[0139] The input is the pseudorange MI matrix and the pseudorange rate MI matrix with normal or abnormal labels; the normal data label is set to 0, and the abnormal data label is set to 1;

[0140] The loss function is a binary classification cross entropy function, and the model parameters are optimized by gradient descent method;

[0141] Set the number of training cycles to M, cycle through the optimal parameters of the training model, calculate the loss and accuracy, and finally save the model with the least loss and the highest accuracy as the trained model. This enables it to more quickly and accurately determine normal data and abnormal data.

[0142] Preferably, M is 50.

[0143] Specifically, the detection of a single-star fault in step S5 includes:

[0144] 1) During the test, a fixed time series length is used, and the pseudo-range time series of a single satellite is updated with a step size of 1, so that the end time of the pseudo-range time series is always the vector Γ at the current time t t =[γ1,...,γ i ,...,γ l ];

[0145] 2) Extract the pseudorange time series and pseudorange rate time series at the current time t and normalize them to obtain the normalized pseudorange time series at the current time t and normalized pseudorange time series

[0146] Extract vector Γ t The first and second rows of the equation are:

[0147] Pseudorange time series Γ t1 ′=[γ1[1],...,γ i [1],...,γ l [1]];

[0148] Pseudorange rate time series Γ t2 ′=[γ1[2],...,γ i [2],...,γ l [2]];

[0149] For the pseudorange time series Γ t1 After normalization, the normalized pseudorange time series is obtained

[0150]

[0151] For the pseudorange rate time series Γ t2 After normalization, we can get the normalized pseudorange rate time series

[0152]

[0153] 3) Use the standard distribution time series of pseudorange and pseudorange rate to compare with the normalized pseudorange time series and pseudorange rate time series Perform mutual information calculation to obtain the pseudorange MI matrix and pseudorange rate MI matrix at the current time t;

[0154] 4) The pseudorange MI matrix and pseudorange rate MI matrix at the current time t are input into the trained data anomaly detection model based on residual attention network for single satellite fault detection.

[0155] In summary, the method disclosed in the embodiment of the present invention combines the pseudorange and pseudorange rate time series to capture the abnormal characteristics of the signal from the two perspectives of slow change and sudden change, thereby improving the accuracy of fault detection;

[0156] A data method based on mutual information transformation is used to reveal the characteristics of each abnormal type, which helps to improve classification performance;

[0157] The data anomaly detection model based on residual attention network can reduce gradient vanishing and improve prediction performance. Compared with other commonly used methods, the trained model has better classification performance;

[0158] And the generalized maximum relevant entropy Kalman filter is used to improve the input data and reduce the non-Gaussian noise in the data, so as to distinguish faults from non-faults more accurately.

[0159] In the embodiment of the present invention, each satellite performs a fault determination process independently, and satellites determined to be faulty can be excluded from the navigation system.

[0160] The above description is only a preferred specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any changes or substitutions that can be easily conceived by any technician familiar with the technical field within the technical scope disclosed by the present invention should be covered within the protection scope of the present invention.

Claims

1. A tightly integrated navigation fault detection method based on mutual information, characterized in that: include: Step S1, establishing the discrete state equation and observation equation of INS error of GNSS / INS tightly integrated navigation system; Generalized maximum correlation entropy Kalman filter is used for INS recursive error estimation and recursive error correction; Step S2, collecting the pseudorange time series and pseudorange rate time series in the filtered new information data of a single satellite and normalizing them; Step S3, establishing a standard distribution time series of pseudorange and pseudorange rate, and performing mutual information operation with the normalized pseudorange time series and pseudorange rate time series to obtain a pseudorange MI matrix and a pseudorange rate MI matrix; Step S4, establishing a data anomaly detection model based on the residual attention network, taking the pseudo-range MI matrix and the pseudo-range rate MI matrix with normal or abnormal labels as input to perform data normal or abnormal classification training; The data anomaly detection model based on the residual attention network includes: a first input end, a first convolutional layer, a first attention mechanism and a residual learning module, a second input end, a second convolutional layer, a second attention mechanism and a residual learning module, a fully connected layer and an output end; The first input terminal, the first convolutional layer, the first attention mechanism and the residual learning module are sequentially connected, and the second input terminal, the second convolutional layer, the second attention mechanism and the residual learning module are sequentially connected; The output ends of the first attention mechanism and residual learning module and the second attention mechanism and residual learning module are connected to a fully connected layer; the fully connected layer is connected to the output end; The first input terminal is connected to the pseudo-range MI matrix with normal or abnormal labels; the second input terminal is connected to the pseudo-range rate MI matrix with normal or abnormal labels; the output terminal outputs the classification result of the data; Step S5: Use the trained data anomaly detection model to detect single-satellite faults during the positioning process of the integrated navigation system.

2. The method for detecting tight integrated navigation faults based on mutual information according to claim 1, characterized in that: The new information data γ at time k obtained by using the generalized maximum relevant entropy Kalman filter process k for: ρ INS , are the pseudorange and pseudorange rate calculated by INS respectively; ρ GNSS , are the pseudorange and pseudorange rate calculated by a single GNSS satellite respectively; H k is the observation matrix; is the one-step estimate of the error state; Take a period of continuous length l of new information data to form a vector Γ=[γ1,...,γ i ,...,γ l ]; Extract the first and second rows of the vector Γ to obtain the pseudorange time series Γ1′ and the pseudorange rate time series Γ2′ respectively; After normalization, the normalized pseudorange time series is obtained and normalized pseudorange rate time series 3. The method for detecting tight integrated navigation faults based on mutual information according to claim 2, characterized in that: From the normalized, zero-mean Gaussian-distributed, randomly generated time series with the same length as the pseudorange time series and pseudorange rate time series, the time series distribution that is closest to the actual satellite data when it is normal is selected through experimental verification as the standard distribution time series of pseudorange and pseudorange rate.

4. The method for detecting tight integrated navigation faults based on mutual information according to claim 3 is characterized in that: The first attention mechanism and residual learning module and the second attention mechanism and residual learning module have the same network structure, both comprising a first full residual module, an attention module, and a second full residual module connected sequentially; Among them, the first full residual module extracts high-level features from the input features; and provides them as input to the attention module, and the attention module develops specific low-level features as input to the second residual block to extract and output high-level features.

5. The method for detecting tight integrated navigation faults based on mutual information according to claim 4, characterized in that: In the attention module, the backbone branch T(x) contains two stacked residual blocks and performs feature extraction; the mask branch M(x) is a bottom-up structure consisting of downsampling using max pooling and upsampling using bilinear interpolation to achieve fast feedforward and feedback attention actions.

6. The method for detecting tight integrated navigation faults based on mutual information according to claim 4, characterized in that: The first full residual module and the second full residual module have the same structure, both of which are network structures obtained by stacking a ReLU activation layer, a batch normalization layer and a convolutional layer twice.

7. The method for detecting tight integrated navigation faults based on mutual information according to claim 4, characterized in that: During the model training process of step S4, The input is the pseudorange MI matrix and pseudorange rate MI matrix with normal or abnormal labels; the normal data label is set to 0, and the abnormal data label is set to 1; The loss function is a binary classification cross entropy function, and the model parameters are optimized by gradient descent method; Set the number of training cycles to M times, cycle through the optimal parameters of the training model, calculate the loss and accuracy, and finally save the model with the least loss and the highest accuracy as the trained model.

8. The mutual information-based tightly integrated navigation fault detection method according to claim 7, characterized in that: The detection of a single-star fault in step S5 includes: 1) During the test, a fixed time series length is used, and the pseudo-range time series of a single satellite is updated with a step size of 1, so that the end time of the pseudo-range time series is always the vector Γ at the current time t t =[γ1,...,γ i ,...,γ l ]; 2) Extract the pseudorange time series and pseudorange rate time series at the current time t and normalize them to obtain the normalized pseudorange time series at the current time t and pseudorange time series 3) Use the standard distribution time series of pseudorange and pseudorange rate to compare with the normalized pseudorange time series and pseudorange rate time series Perform mutual information calculation to obtain the pseudorange MI matrix and pseudorange rate MI matrix at the current time t; 4) The pseudorange MI matrix and pseudorange rate MI matrix at the current time t are input into the trained data anomaly detection model based on residual attention network for single satellite fault detection.

9. The method for detecting tight integrated navigation faults based on mutual information according to claim 8, characterized in that: In step S1, the process of using generalized correlation entropy Kalman filtering to perform INS recursive error estimation includes: 1) Perform one-step state prediction to obtain one-step estimation at time k The covariance matrix P(k|k-1) of 2) One-step estimation based on k moment The covariance matrix P(k|k-1) of and the covariance matrix R(k) of the observation noise are decomposed by Cholesky; 3) Based on the Cholesky decomposition results, one-step estimation based on k moments Perform iterative calculation of the filter residual e(k,t); 4) According to the filtering residual of t iterations, the generalized correlation entropy weighted value of each observation channel of t iterations is calculated to form a generalized correlation entropy weighted matrix 5) Using the generalized correlation entropy weighted matrix Calculate the Kalman gain K(k,t); 6) According to the Kalman gain K(k,t), the state is estimated to obtain the t+1th iteration value When the difference between the t+1th iteration value and the tth iteration value is greater than the threshold, return to step 3) and execute again; otherwise, the best estimated value of the state quantity k at the current moment is obtained.

Images