GNSS / INS integrated navigation method based on generalized maximum correlation entropy under short GNSS outage
Through the combination of generalized maximum correlation entropy Kalman filtering and CNN-GRU network, the stability problem of GNSS/INS combined navigation system in GNSS interrupts and outlier abnormalities is solved, and the system's difference resistance and interrupt prediction are realized to ensure that The stability and robustness of navigation performance are achieved.
Patent Information
- Application Number
- CN202410412134.8
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2024-04-08
- Publication Date
- 2024-11-19
- Estimated Expiration
- 2044-04-08
AI Technical Summary
When GNSS/INS combined navigation system encounters GNSS observation interrupts and outlier abnormalities, it is difficult to maintain stable performance. The existing methods fail to effectively combine differences resistance and interrupt prediction, resulting in a degradation in navigation performance.
The generalized maximum correlation entropy Kalman filtering and CNN-GRU neural network are used to establish a GNSS/INS tight combined navigation system, providing resistance to differences through generalized correlation entropy Kalman filtering, and using the CNN-GRU network to perform GNSS pseudorange observation prediction. Reconstruct the pseudo-range observation of the lost lock satellite to ensure the stability of the system when the GNSS is short-term interruption.
It realizes the integration of the resistance to GNSS and interrupt prediction when short-term interruption of GNSS, provides higher quality data sample sequence and navigation information fusion, improves the robustness and stability of the system, and can be used when GNSS signal is lost. Continue to effectively navigate.
Smart Images

Figure CN118226489B_ABST
Abstract
Description
Technical Field
[0001] The present invention belongs to the field of integrated navigation, and in particular relates to a GNSS / INS integrated navigation method based on generalized maximum correlation entropy under short-term GNSS interruption. Background Art
[0002] In the field of navigation, multi-source fusion navigation solutions that utilize both inertial navigation system (INS) and global navigation satellite system (GNSS) positioning information can leverage the complementarity between each information source and overcome the limitations of navigation using a single source. This is a current trend in navigation technology, specifically multi-source information fusion navigation and positioning technology using multiple sensors. However, integrated GNSS / INS navigation systems rely on continuous, high-quality GNSS observations to suppress the error divergence of inertial navigation. As an exogenous navigation system, GNSS is susceptible to environmental influences, leading to outlier anomalies in GNSS observations and even widespread receiver lock loss, resulting in interruptions in GNSS observations and an inability to obtain sufficient GNSS observations. These phenomena can degrade the overall performance of the integrated navigation system. Furthermore, in various complex application scenarios, such as urban canyons, tunnels, forests, and environments with strong electromagnetic interference, GNSS observations containing outlier errors and loss of lock on observable satellite signals can often be correlated. Interference can prevent the receiver from correctly demodulating and tracking observable satellite signals, thus affecting the observed data. Therefore, a robust and stable integrated navigation system must consider these issues.
[0003] Current research mostly focuses on one of these issues, without addressing their interrelationships. For example, algorithms currently use time series or neural network methods to predict GNSS outages, but these require large, high-quality datasets during training. Furthermore, if filtering algorithms are not robust enough to handle the degradation of navigation performance caused by outlier-induced observations before a GNSS outage, contaminated data can significantly impact training results and the accuracy of GNSS outage predictions.
[0004] Specifically, when outliers appear in GNSS observations, current research focuses on improving the Kalman filter. For example, by introducing scaling factors and fading factors, these factors can be used to assign smaller weights to observations by appropriately adjusting the covariance of the observation channel where the outliers appear. However, setting appropriate scaling and fading factors is difficult. In fact, the reason why observations with outliers degrade the performance of traditional Kalman filters is that observations with increased outliers exhibit the statistical characteristics of a fat-tailed distribution. Traditional Kalman filters assume a Gaussian noise distribution and are derived as optimal estimators under the minimum mean square error criterion. Therefore, they are susceptible to the influence of outliers in the observation errors.
[0005] Most existing prediction methods during GNSS outages directly predict the carrier position change. This is because they are mostly aimed at single-sensor GNSS positioning scenarios or loosely coupled GNSS / INS systems. For tight combinations, directly predicting the user position or position change means that the tight combination filtering system is interrupted and some of the GNSS observations that can still be observed are not utilized. Summary of the Invention
[0006] In view of the above analysis, the present invention aims to disclose a GNSS / INS integrated navigation method based on generalized maximum correlation entropy under short GNSS outages, which can not only provide robustness when the GNSS system observations are interfered by outliers, but also provide observation prediction to maintain system performance stability when the situation deteriorates further, that is, when the observable satellite signals are lost and sufficient GNSS observations cannot be obtained.
[0007] The present invention discloses a GNSS / INS integrated navigation method based on generalized maximum correlation entropy under GNSS short-term interruption, comprising:
[0008] Step S1, establishing the discrete state equation and observation equation of the GNSS / INS tightly integrated navigation system INS error; using the generalized correlation entropy Kalman filter to perform INS recursive error estimation and recursive error correction;
[0009] Step S2: Building an observation prediction network; using observation data collected by the tightly integrated navigation system during normal GNSS positioning, including INS sensor data, corrected state estimation data, and satellite position change data, to train the observation prediction network online to predict satellite pseudorange observations;
[0010] Step S3: When a satellite lock interruption occurs and the satellite pseudorange observations are insufficient, the trained observation prediction network is used to reconstruct the pseudorange observations of the most recently lost satellite, and the pseudorange observations of the satellites that can still be observed are input into the filter together, and the GNSS / INS integrated navigation is continued.
[0011] Furthermore, during the INS recursive error estimation process, the state vector of the established GNSS / INS tightly integrated navigation system is:
[0012]
[0013] in, φ, b g 、b a 、s g and s 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 results in the ECEF coordinate system;
[0014] The observed quantity is:
[0015]
[0016] Among them, ρ GNSSi is the pseudorange observed for the i-th observable satellite at time k, is the three-dimensional coordinate of the position of the i-th satellite, is the three-dimensional coordinate of the position recursively calculated by INS at time k; i = 1,…,m, where m is the number of observable satellites.
[0017] Furthermore, the process of using generalized correlation entropy Kalman filtering to estimate the INS recursive error includes:
[0018] 1) Perform one-step state prediction to obtain one-step estimation at time k The covariance matrix P(k|k-1) is:
[0019] 2) One-step estimation based on k moments The covariance matrix P(k|k-1) of and the covariance matrix R(k) of the observation noise are decomposed by Cholesky;
[0020] 3) According to the Cholesky decomposition results, one-step estimation based on k moments Perform iterative calculation of the filter residual e(k, t); t represents the number of iterations, and the initial value is 1;
[0021] 4) Calculate the generalized correlation entropy weighted value of each observation channel of t iterations based on the filtering residual of t iterations to form the generalized correlation entropy weighted matrix
[0022] 5) Using the generalized correlation entropy weighted matrix Calculate the Kalman gain K(k, t);
[0023] 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 re-execute; otherwise, obtain the best estimate of the state quantity k at the current moment
[0024] Furthermore, the weight matrix of t iterations at time k is
[0025]
[0026] Among them, h i (e i (k, t)) is the kernel function of the i-th channel,
[0027] Kernel function h i (e i (k, t)) is expressed as:
[0028] h i (e i (k, t)) = |e i (k, t)| α-2 exp(-β -α |e i (k, t) α |);
[0029] The shape parameter of the generalized Gaussian kernel function is α=2, and the kernel bandwidth is β=0.1.
[0030] Furthermore, the estimated error value is used to perform feedback correction on the recursive carrier motion state.
[0031] Use the best estimate of the state quantity at time k in φ corrects the position of the INS recursive state speed Direction cosine matrix Get the current position after estimated error correction speed Direction cosine matrix
[0032]
[0033] Furthermore, in step S2, the observation prediction network is constructed as a CNN-GRU neural network, including one CNN network and one GRU network; wherein,
[0034] The CNN network is used to reconstruct the features of the input data of l observable satellites to obtain the feature data of each observable satellite with significant similarity;
[0035] The output ends of the CNN network are respectively connected to the input ends of 1 GRU network; 1 GRU network corresponds to 1 observable satellite;
[0036] Each GRU network is used to process the feature data of the corresponding observable satellite output by the CNN network and infer the pseudorange change of the observable satellite at the next moment.
[0037] Furthermore, the l visible satellites are the l visible satellites with the largest elevation angles.
[0038] Furthermore, the process of online training of the observation prediction network includes:
[0039] 1) Using the observation data collected from N consecutive moments back to the GNSS normal positioning moment k, including INS sensor data, corrected state estimation data, and the position change and pseudorange change of the i-th observable satellite, a training sample X(k, i|N) of the i-th visible satellite is constructed, where i = 1, ..., l;
[0040] 2) Splice the training samples of l visible satellites into a sample input data matrix by row;
[0041] 3) Starting from the first row of the sample input data matrix, the data is input into the CNN network of the CNN-GRU neural network row by row. After each input of X(k, i|N), the CNN network outputs a feature matrix X′(k, i|N);
[0042] 4) Output the feature matrix X′(k, i|N) to the GRU network corresponding to the observable satellite i, and infer the pseudorange change Δρ of the observable satellite i at the next moment k+1,i .
[0043] Furthermore, when constructing the training sample, the data collected at the normal GNSS positioning time k includes:
[0044] INS sensor outputs incremental data S k =[Δθ x , Δθ y , Δθ z , Δv x , Δv y , Δv z ]; where Δθ x , Δθ y , Δθ z and Δv x , Δv y , Δv z are the three-axis angle and velocity increments of the INS sensor from time k to time k+1;
[0045] The estimated values of the carrier position and speed after correction obtained in step S1
[0046] Carrier location The line of sight vector L with observable satellite i k,i :
[0047]
[0048] in, is the position vector of observable satellite i at time k, ||·| |2 is the vector 2-norm;
[0049] The change vector AP(k,i) of the position of observable satellite i:
[0050]
[0051] is the position vector of observable satellite i at time k+1; are the changes in the X, Y, and Z axis coordinates of the observable satellite i respectively;
[0052] In addition, the pseudorange change of the observable satellite i from time k-1 to time k must also be included:
[0053]
[0054] The training sample X(k, i|N) of observable satellite i collected from N consecutive moments back to time k is:
[0055]
[0056] Furthermore, in step S3, when the observable satellites of the GNSS / INS tightly integrated navigation system lose lock with the observable satellite i at time k+1, the prediction process includes:
[0057] 1) Detect the most recently lost observable satellite i and trace back the navigation data of the N most recent moments at time k;
[0058] 2) Generate feature samples X(k, i|N) according to step S2 and input them into the observation prediction network completed by online training; use convolutional network f Conv (·) performs feature extraction and then inputs the GRU network corresponding to the observable satellite i Predict the pseudorange change Δρ k+1,i ;
[0059] 3) Calculate the predicted pseudorange is the pseudorange observed at time k;
[0060] get The next step is to enter step S1 to replace ρ GNSSi Continue to calculate Z k , thus, after the observable satellite i loses lock, the filtering process of step S1 is continued.
[0061] The present invention can achieve the following beneficial effects:
[0062] 1. The present invention integrates the robustness function and the interruption prediction function into the same system, and cooperates with each other: when affected by outliers, the robustness of the generalized maximum correlation entropy filter provides a higher quality and more consistent data sample sequence for the training of the prediction network; the observation value predicted by the prediction network can be input into the robustness filter for information fusion to provide navigation information.
[0063] 2. This paper uses a CNN-GRU neural network to predict GNSS outage observations. The CNN converts raw data, such as sensor output data sequences, historical GNSS pseudorange change sequences, and observable satellite position change sequences, into feature sequences. The GRU network then predicts the interrupted GNSS pseudorange observations based on the temporal correlation between these features. The GRU network's memory capacity allows it to extract information from sequences spanning longer time periods.
[0064] 3. The GNSS outage prediction method targets pseudorange observation changes, rather than directly predicting the user's position or position change. Furthermore, the predicted pseudoranges can be input into a filter together with the pseudoranges of the remaining observable satellites, fully utilizing all available information. BRIEF DESCRIPTION OF THE DRAWINGS
[0065] The accompanying drawings are only used for the purpose of illustrating specific embodiments and are not to be considered as limiting the present invention. Throughout the drawings, the same reference symbols denote the same components.
[0066] Figure 1 Flowchart of a GNSS / INS integrated navigation method based on generalized maximum correlation entropy under short GNSS outage in an embodiment of the present invention;
[0067] Figure 2 This is a structural diagram of the CNN-GRU neural network in an embodiment of the present invention. DETAILED DESCRIPTION
[0068] The preferred embodiments of the present invention will be described in detail below with reference to the accompanying drawings, which 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.
[0069] One embodiment of the present invention discloses a GNSS / INS integrated navigation method based on generalized maximum correlation entropy under short GNSS outage, such as Figure 1 As shown, including:
[0070] Step S1, establishing the discrete state equation and observation equation of the GNSS / INS tightly integrated navigation system INS error; using the generalized correlation entropy Kalman filter to perform INS recursive error estimation and recursive error correction;
[0071] Step S2: Building an observation prediction network; using observation data collected by the tightly integrated navigation system during normal GNSS positioning, including INS sensor data, corrected state estimation data, and satellite position change data, to train the observation prediction network online to predict satellite pseudorange observations;
[0072] Step S3: When a satellite lock interruption occurs and the satellite pseudorange observations are insufficient, the trained observation prediction network is used to reconstruct the pseudorange observations of the most recently lost satellite, and the pseudorange observations of the satellites that can still be observed are input into the filter together, and the GNSS / INS integrated navigation is continued.
[0073] Specifically, in step S1, the state vector of the established GNSS / INS tightly integrated navigation system is:
[0074]
[0075] in, φ, b g 、b a 、s g and s 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 results in the ECEF coordinate system;
[0076] According to the error perturbation theory, the discrete state equation and observation equation of INS error are established as follows:
[0077] X k =Φ k|k-1 X k-1 +ω k-1 ;
[0078] Z k =H k X k +v k ;
[0079] Among them, 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;
[0080] Z k is the observed value at time k, v k is the observation noise at time k; H k is the observation matrix at time k.
[0081] Among them, the state transfer matrix Φ k|k-1 for:
[0082]
[0083] 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 、T ba 、T sg 、T sa Gyro bias error, accelerometer bias error, gyro scale factor error, and accelerometer scale factor error are modeled as related time parameters of the first-order Markov process;
[0084] Observation quantity Z k for:
[0085]
[0086] Among them, ρ GNSSi is the pseudorange observed for the i-th observable satellite at time k, ρ INSi is the geometric distance calculated based on the position delivered by the INS at time k and the position of the i-th observable satellite; i = 1, ..., m; m is the number of observable satellites;
[0087] The observation matrix H at time k k , the optimal estimate of the known k-1 time state and are calculated from the satellite position at time k solved from the ephemeris solution and are known quantities.
[0088] The relationship between the satellite pseudorange 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:
[0089]
[0090] Among them, ρ GNSSi is the pseudorange observed for the i-th observable satellite at time k, is the three-dimensional coordinate of the position of the i-th satellite, is the three-dimensional coordinate of the position recursively calculated by INS at time k; i = 1,…,m, where m is the number of observable satellites.
[0091] Specifically, in step S1, the process of using the generalized correlation entropy Kalman filter to perform INS recursive error estimation includes:
[0092] 1) Perform one-step state prediction to obtain one-step estimation at time k The covariance matrix P(k|k-1) is:
[0093]
[0094] is the optimal estimate of the error state at time k-1, For use The one-step estimate of the error state at time k calculated by the state transfer matrix Φk|k-1, 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;
[0095] Here and P(k|k-1) will be used later to calculate the optimal estimate of the error state at time k
[0096] When the system starts, it is assumed that the first moment k=1. At this time, the k-1 moment corresponds to the 0 moment. The state at the 0 moment 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 physical system used and the prior knowledge.
[0097] 2) One-step estimation based on k moments The covariance matrix P(k|k-1) of and the covariance matrix R(k) of the observation noise are decomposed by Cholesky;
[0098]
[0099] R(k) is the covariance matrix of the observation noise in the observation equation.
[0100] 3) According to the Cholesky decomposition results, one-step estimation based on k moments Perform iterative calculation of the filtering residual e(k, t);
[0101] e(k,t)=D(k)-W(k)x(k,t);
[0102] in, z(k) is the observation at time k; n is the dimension of the state vector;
[0103] H k is the observation matrix at time k;
[0104] 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 the initial value is 1.
[0105] 4) Calculate the generalized correlation entropy weighted value of each observation channel of t iterations based on the filtering residual of t iterations to form the generalized correlation entropy weighted matrix
[0106] The weight matrix of t iterations at time k
[0107]
[0108] Among them, hi(ei(k, t)) is the i-th channel kernel function,
[0109] The kernel function hi(ei(k, t)) is expressed as:
[0110] h i (e i (k, t)) = |e i (k, t)| α-2 exp(-β -α |e i (k, t) α |);
[0111] β is the kernel bandwidth of the generalized Gaussian kernel function, α is the shape parameter, when the filter's 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 to errors.
[0112] 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 hi(e i The decay of (k, t)) is first-order in the interval with small residuals, while the decay is high-order and decays rapidly to 0 in the interval with large residuals. 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.
[0113] 5) Using the generalized correlation entropy weighted matrix Calculate the Kalman gain K(k, t);
[0114]
[0115] in, is the observation error covariance matrix of each observation channel weighted by relevant entropy;
[0116] 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 re-execute; otherwise, obtain the best estimate of the state quantity k at the current moment
[0117] Specifically, It is the iteration value obtained at the t+1th iteration, which is calculated from the result of the tth iteration.
[0118] At this time, if:
[0119]
[0120] 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.
[0121] The estimated error value is used to perform feedback correction on the carrier motion state.
[0122] Use the best estimate of the state quantity at time k in φ corrects the position of the INS recursive state speed Direction cosine matrix Get the current position after estimated error correction speed Direction cosine matrix
[0123]
[0124] 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 recursive 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.
[0125] Specifically, the observation prediction network built in step S2 is a CNN-GRU neural network, such as Figure 2 As shown, it includes 1 CNN network and ,GRU networks; among them,
[0126] The CNN network is used to reconstruct the features of the input data of l observable satellites to obtain the feature data of each observable satellite with significant similarity;
[0127] The output ends of the CNN network are respectively connected to the input ends of 1 GRU network; 1 GRU network corresponds to 1 observable satellite;
[0128] Each GRU network is used to process the feature data of the corresponding observable satellite output by the CNN network and infer the pseudorange change of the observable satellite at the next moment.
[0129] Since, for different observable satellites, the relative motion characteristics of the carrier are similar when sufficient feature parameters are provided; therefore, in the solution of this embodiment, a network structure of one CNN network connected to two GRU networks is adopted; wherein, the CNN feature reconstruction network should show significant similarity under the condition of proper fitting, so when constructing the CNN-GRU pseudorange prediction network for different observable satellites, the feature extraction part can use the same CNN convolutional network f Conv (·);
[0130] Preferably, a dropout layer is added to the CNN convolutional network to randomly inactivate some neurons to suppress overfitting of individual observable satellite features;
[0131] In the CNN-GRU network, the parameter scale of CNN is relatively large. Sharing the CNN network can also effectively control the network training efficiency, reduce deployment costs, and improve the real-time prediction performance.
[0132] Furthermore, since pseudorange variation is not a time-independent state variable, it is not only related to the current input feature parameters but can also be derived from the time dimension. Therefore, the GRU network is introduced. It is a recurrent neural network that excels at analyzing time series data and making inferences and predictions about unknown data. As a variant of the LSTM, the GRU network has a stronger memory capacity and can accept larger input sequences. Furthermore, it simplifies the LSTM model, requiring fewer parameters to train and achieving faster training speed, making it suitable for larger-scale deployments. This further reduces network size and deployment costs for scenarios requiring differentiated predictions for multiple observable satellites.
[0133] In addition, since the time series of different observable satellites are independent, the fitted GRU network is specific and cannot be shared; therefore, there is a GRU network corresponding to each observable satellite; and the respective network parameters need to be trained during training.
[0134] More specifically, considering that satellites observable at high elevation angles have higher observation quality and that high elevation angles can alleviate multipath fading and shadowing issues, given limited platform resources, the outage prediction network must prioritize ensuring that short, unexpected outages of high-quality observable satellites are promptly controlled to prevent serious impacts on system performance. Based on this consideration, it is not necessary to allocate GRU networks to all observable satellites. Instead, GRU networks should be allocated and trained online only for satellites with elevation angles that meet the requirements. The network training process for these observable satellites should be continuous until a certain observable satellite becomes unobservable for an extended period (>5 minutes), at which point the corresponding network is abandoned, initialized, and reassigned.
[0135] Preferably, the l observable satellites are the l observable satellites with the largest elevation angles, and l ≥ 4. The network training process for these observable satellites should be kept consistent until an observable satellite cannot be observed for a long time (> 5 minutes), then the corresponding network is abandoned and initialized and reallocated.
[0136] For an observable satellite i, the elevation angle El at time k k , i is calculated as:
[0137]
[0138] in Refers to the estimated value of the carrier position The line of sight vector L with observable satellite i k,i The third coordinate element of the projection vector in the ENU coordinate system (n system).
[0139] After establishing a model and generating training samples, the network is iteratively trained using the back-propagation algorithm. As the navigation system runs, new data is continuously received, the sample sequence is updated, and the weights are continuously updated to optimize network performance.
[0140] Specifically, Figure 2 In the CNN-GRU neural network, the network parameters are:
[0141] parameter Setting value Conv1: 32@3×1 Conv2: 16@4×1 Flatten: Output dimension 8 Pooling Type: Maxpool Pooling window size 2 CNN activation function Relu CNN Dropout inactivation rate 0.25 GRU activation function Hyperbolic tangent Number of GRU hidden layer units 64 Optimizer methods Adam algorithm Loss Function Mean square error Fully connected layer Output dimension 1
[0142] Specifically, the process of online training of the observation prediction network includes:
[0143] 1) Using the observation data collected from N consecutive moments back to the GNSS normal positioning moment k, including INS sensor data, corrected state estimation data, and the position change and pseudorange change of the i-th observable satellite, a training sample X(k, i|N) of the i-th visible satellite is constructed, where i = 1, ..., l;
[0144] Specifically, at the time k when the GNSS is performing normal positioning, the information fusion filtering has been completed, and the INS sensor has completed the data output at the time k+1. Then, the collected data includes:
[0145] At time k+1, the INS sensor outputs incremental data S k =[Δθ x , Δθ y , Δθ z , Δv x , Δv y , Δv z ]; where Δθ x , Δθ y , Δθ z and Δv x , Δv y , Δv z are the three-axis angle and velocity increments of the INS sensor from time k to time k+1;
[0146] The estimated values of the carrier position and velocity after correction obtained after step S1 at time k
[0147] Carrier position at time k The line of sight vector L with observable satellite i k,i :
[0148]
[0149] in, is the position vector of observable satellite i at time k, ||·| |2 is the vector 2-norm;
[0150] The change vector of the observable satellite i’s position at time k, AP(k,i):
[0151]
[0152] is the position vector of observable satellite i at time k+1; are the changes in the X, Y, and Z axis coordinates of the observable satellite i respectively;
[0153] In addition, the pseudo-range change of the observable satellite i from time k-1 to time k must also be included:
[0154]
[0155] The training sample X(k, i|N) of observable satellite i collected from N consecutive moments back to time k is:
[0156]
[0157] 2) Splice the training samples of l visible satellites into a sample input data matrix by row;
[0158] 3) Starting from the first row of the sample input data matrix, the data is input into the CNN network of the CNN-GRU neural network row by row. After each input of X(k, i|N), the CNN network outputs a feature matrix X′(k, i|N);
[0159] In this embodiment, the convolutional neural network re-extracts features from each moment sample in X(k, i|N) (i.e., each row of X(k, i|N)). Here, a multi-feature time series is used to train the network to predict the upcoming pseudorange change, because The change characteristics can be seen not only from Δρ i It is obtained from the sequence itself and also exists in L k,i There are complex connections between the feature parameters, and the feature extraction capability of convolutional neural networks is used for pre-processing.
[0160] The CNN network is responsible for fitting the complex nonlinear relationship between features and pseudorange changes to form the extracted feature sequence;
[0161] 4) Output the feature matrix X′(k, i|N) to the GRU network corresponding to the observable satellite i, and infer the pseudorange change Δρ of the observable satellite i at the next moment k+1,i ;
[0162] The GRU network corresponding to the observable satellite i is responsible for inferring the pseudorange change Δρ to the observable satellite i at the next moment k+1,i :
[0163]
[0164] Specifically, when the observable satellites of the GNSS / INS tightly integrated navigation system lose lock with observable satellite i at time k+1, the prediction process includes:
[0165] 1) Detect the most recently lost observable satellite i and trace back the navigation data of the N most recent moments at time k;
[0166] 2) Generate feature samples X(k, i|N) according to step S2 and input them into the observation prediction network completed by online training; use convolutional network f Conv (·) performs feature extraction and then inputs the GRU network corresponding to the observable satellite i Predict the pseudorange change Δρ k+1,i ;
[0167] 3) Calculate the predicted pseudorange is the pseudorange observed at time k;
[0168] get The next step is to enter step S1 to replace ρ GNSSi Continue to calculate Z k , thus, after the observable satellite i loses lock, the filtering process of step S1 is continued.
[0169] To sum up, the present invention of the embodiment of the present invention integrates the robustness function and the interruption prediction function into the same system, and cooperates with each other: the robustness of the generalized maximum correlation entropy filter when affected by outliers provides a higher quality and more consistent data sample sequence for the training of the prediction network; the observation quantity prediction value output by the prediction network can be input into the robustness filter for information fusion to provide navigation information.
[0170] A CNN-GRU neural network is used to predict GNSS outage observations. The CNN converts raw data, such as sensor output data sequences, historical GNSS pseudorange change sequences, and observable satellite position change sequences, into feature sequences. The GRU network then predicts the interrupted GNSS pseudorange observations based on the temporal correlation between these features. The GRU network's memory capacity allows it to extract information from sequences spanning longer time periods.
[0171] GNSS outage prediction targets the change in pseudorange observations, rather than directly predicting the user's position or position change. Furthermore, the predicted pseudoranges can be input into a filter together with the pseudoranges of the remaining observable satellites, fully utilizing all available information.
[0172] To sum up, the above is only a preferred specific embodiment of the present invention, but the scope of protection of the present invention is not limited to this. Any changes or replacements that can be easily thought of by any technician familiar with this technical field within the technical scope disclosed by the present invention should be covered by the scope of protection of the present invention.
Claims
1. A GNSS / INS integrated navigation method based on generalized maximum correlation entropy under GNSS short-term interruption, 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 correlation entropy Kalman filter is used for INS recursive error estimation and recursive error correction; Step S2, building an observation prediction network; Using the observation data collected by the tightly integrated navigation system during normal GNSS positioning, including INS sensor data, corrected state estimation data, and satellite position change data, the observation prediction network is trained online to predict the satellite pseudorange observations. Step S3: When a satellite lock interruption occurs and the satellite pseudorange observations are insufficient, the pseudorange observations of the most recently lost satellite are reconstructed using the trained observation prediction network, and the pseudorange observations of the still observable satellites are input into the filter together to continue the GNSS / INS integrated navigation.
2. The GNSS / INS integrated navigation method based on generalized maximum correlation entropy under GNSS short-term interruption according to claim 1, characterized in that: During the INS recursive error estimation process, the state vector of the established GNSS / INS tightly integrated navigation system is: 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; The observed quantity is: Among them, ρ GNSSi is the pseudorange observed for the i-th observable satellite at time k, is the three-dimensional coordinate of the position of the i-th satellite, is the three-dimensional coordinate of the position recursively derived by INS at time k; i = 1,…,m, where m is the number of observable satellites.
3. The GNSS / INS integrated navigation method based on generalized maximum correlation entropy under GNSS short-term interruption according to claim 2, characterized in that: The process of using generalized correlation entropy Kalman filtering to estimate INS recursive error includes: 1) Perform one-step state prediction to obtain one-step estimation at time k The covariance matrix P(k|k-1) is: 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); t represents the number of iterations, and the initial value is 1; 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 weighting 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.
4. The GNSS / INS integrated navigation method based on generalized maximum correlation entropy under GNSS short-term interruption according to claim 3, characterized in that: The weight matrix for t iterations at time k Among them, h i (e i (k, t)) is the kernel function of the i-th channel, Kernel function h i (e i (k, t)) is expressed as: h i (e i (k,t))=|e i (k,t)| α-2 exp(-β -α |e i (k,t) α |); The shape parameter of the generalized Gaussian kernel function is α=2, and the kernel bandwidth is β=0.
1.
5. The GNSS / INS integrated navigation method based on generalized maximum correlation entropy under GNSS short-term interruption according to claim 3, characterized in that: The estimated error value is used to feedback correct the recursive carrier motion state. Use the best estimate of the state at time k In φ corrects the position of the INS recursive state speed Direction cosine matrix Get the current position after the estimated error correction speed Direction cosine matrix 6. The GNSS / INS integrated navigation method based on generalized maximum correlation entropy under GNSS short-term interruption according to any one of claims 1 to 5, characterized in that: In step S2, the observation prediction network is constructed as a CNN-GRU neural network, including 1 CNN network and 1 GRU network; wherein, The CNN network is used to reconstruct the features of the input data of l observable satellites to obtain the feature data of each observable satellite with significant similarity; The output ends of the CNN network are respectively connected to the input ends of l GRU networks; one GRU network corresponds to one observable satellite; Each GRU network is used to process the feature data of the corresponding observable satellite output by the CNN network, and infer the pseudorange change of the observable satellite at the next moment.
7. The GNSS / INS integrated navigation method based on generalized maximum correlation entropy under GNSS short-term interruption according to claim 6, characterized in that: The l visible satellites are the l visible satellites with the largest elevation angles.
8. The GNSS / INS integrated navigation method based on generalized maximum correlation entropy under GNSS short-term interruption according to claim 7, characterized in that: The process of online training of observation prediction network includes: 1) Using the observation data collected from N consecutive moments back to the normal GNSS positioning moment k, including INS sensor data, corrected state estimation data, and the position change and pseudorange change of the i-th observable satellite, a training sample X(k, i|N) of the i-th visible satellite is constructed, i=1,...,l; 2) Concatenate the training samples of visible satellites into a sample input data matrix row by row; 3) Starting from the first row of the sample input data matrix, the data is input into the CNN network of the CNN-GRU neural network row by row. After each input of X(k, i|N), the CNN network outputs a feature matrix X′(k, i|N); 4) Output the feature matrix X′(k, i|N) to the GRU network corresponding to the observable satellite i, and infer the pseudo-range change ΔΔρ of the observable satellite i at the next moment k+1,i .
9. The GNSS / INS integrated navigation method based on generalized maximum correlation entropy under GNSS short-term interruption according to claim 8, characterized in that: When constructing training samples, the collected data of the GNSS normal positioning time k includes: INS sensor outputs incremental data S k =[Δθ x , Δθ y , Δθ z , Δv x , Δv y , Δv z ]; where Δθ x , Δθ y , Δθ z and Δv x , Δv y , Δv z are the three-axis angle and velocity increments of the INS sensor from time k to time k+1; The estimated values of the carrier position and speed after correction obtained in step S1 Carrier location The line of sight vector L with observable satellite i k,i : in, is the position vector of observable satellite i at time k, ||·||2 is the vector 2-norm; The change vector AP(k, i) of the observable satellite i position: is the position vector of observable satellite i at time k+1; are the changes in the X, Y, and Z axis coordinates of the observable satellite i respectively; In addition, the pseudo-range change of observable satellite i from time k-1 to time k must also be included: The training sample X(k, i|N) of observable satellite i collected from N consecutive moments back to time k is:
10. The GNSS / INS integrated navigation method based on generalized maximum correlation entropy under GNSS short-term interruption according to claim 9, characterized in that: In step S3, when the observable satellites of the GNSS / INS tightly integrated navigation system lose lock on the observable satellite i at time k+1, the prediction process includes: 1) Detect the observable satellite i that has lost lock recently, and trace back the navigation data of the N most recent moments at time k; 2) Generate feature samples X(k, i|N) according to step S2 and input them into the observation prediction network completed by online training; use convolutional network f Conv (·) Feature extraction is performed and then input into the GRU network corresponding to the observable satellite i Predict the pseudorange change Δρ k+1,i ; 3) Calculate the predicted pseudorange is the pseudorange observed at time k; The next step is to replace ρ in step S1 GNSSi Continue to calculate Z k , thus, after the observable satellite i loses lock, the filtering process of step S1 continues.
