A combined navigation adaptive attitude correction method, device and medium

Abstract

This invention relates to the field of integrated navigation technology, and more particularly to an adaptive attitude correction method, device, and medium for integrated navigation. The method includes model training and real-time navigation correction. In model training, a CNN-GRU hybrid neural network is trained in a supervised manner using binary attitude quality labels generated from multi-scenario measured data to obtain a CNN-GRU model, which outputs an observation confidence score. In real-time navigation, the CNN-GRU model reads the continuous EKF innovation sequence and GNSS signal data at the current moment in real time and outputs an observation confidence score. Based on the observation confidence score, an adaptive dual-track correction mechanism is constructed and executed to achieve adaptive attitude correction for integrated navigation. This invention can establish a high-precision attitude reference within the available GNSS window, cut off abnormal observation contamination paths, and significantly suppress error divergence in subsequent pure inertial navigation stages.

Description

Technical Field

[0001] This invention relates to the field of integrated navigation technology, and in particular to an adaptive attitude correction method, device and medium for integrated navigation. Background Technology

[0002] In high-precision navigation applications under complex dynamic environments, INS / GNSS integrated navigation, by fusing the short-term dynamic performance of inertial sensors with the long-term absolute stability of satellite signals, has become the mainstream solution for attitude estimation. However, in typical complex environments such as urban canyons, long tunnels, or areas with electromagnetic interference, GNSS signals are highly susceptible to multipath effects, blockage, or interference, leading to degradation or even complete interruption (i.e., GNSS denial environments). In such cases, high-precision INS / GNSS integrated navigation systems are often forced to switch to a pure inertial navigation mode lacking external observation information for correction. In pure inertial navigation mode, the system faces a fatal divergence problem: the inherent sensor errors of INS (especially gyroscope and accelerometer zero bias) accumulate over time. More critically, the attitude angle estimation error, as a fundamental component of inertial calculation, experiences a severe amplification effect from even small initial deviations during force integration, leading to a rapid accumulation and even divergence in the system's position and velocity calculations. Therefore, the ability to maintain high-precision and robust attitude estimation under integrated navigation directly determines the continuous availability and overall navigation accuracy of the pure inertial navigation system.

[0003] Currently, to suppress error divergence in the pure inertial navigation phase, existing integrated navigation schemes often employ extended Kalman filters (EKFs) to attempt attitude reference correction during the integrated navigation phase. However, when vehicles and other carriers are in dynamic environments, actual GNSS observation noise exhibits strong time-varying and nonlinear characteristics. Traditional methods often rely on fixed thresholds for attitude feedback decisions or to eliminate abnormal observations. This threshold-based decision-making lacks robustness in complex and variable signal scenarios, easily leading to misjudgments or omissions. This can cause degraded or abnormal GNSS observation data to severely contaminate the system's original attitude reference through the cross-coupling effect of filters.

[0004] On the other hand, some auxiliary methods that have introduced shallow neural networks in recent years often only focus on fitting a single observation sequence, failing to fully combine filtered innovation sequences with multi-source data such as the underlying GNSS signal state for in-depth temporal correlation analysis. These methods lack the ability to dynamically adjust the EKF observation noise covariance matrix, making it impossible to accurately assess observation quality in complex signal environments, and consequently making it difficult to establish and maintain a high-purity attitude reference during the integrated navigation phase.

[0005] In summary, existing technologies face technical bottlenecks in complex dynamic scenarios, such as inaccurate observation quality assessment, poor adaptability of fixed thresholds, and susceptibility to cross-contamination of attitude references. There is an urgent need for a combined navigation adaptive attitude correction method that can robustly establish high-precision attitude references within the available GNSS window. Summary of the Invention

[0006] To address the aforementioned technical problems, this invention proposes an adaptive attitude correction method, device, and medium for integrated navigation. This method establishes a high-precision attitude reference within the available GNSS window, cuts off abnormal observation contamination paths, and significantly suppresses error divergence in the subsequent pure inertial navigation stage.

[0007] In a first aspect, the present invention provides an adaptive attitude correction method for integrated navigation, comprising model training and real-time correction, wherein: S1, Model Training: S11. Extract the position information sequence, velocity information sequence, and global navigation satellite system signal status data from the extended Kalman filter during integrated navigation to construct input feature samples; S12. Generate binary pose quality labels using multi-scenario measured data, and perform supervised training on a hybrid neural network containing convolutional neural networks and gated recurrent units using the input feature samples and corresponding labels to obtain the trained CNN-GRU model; the CNN-GRU model outputs continuous posterior probability values ​​as observation confidence scores; S2, Real-time calibration: S21. During integrated navigation, the position information, velocity information and current GNSS signal status data calculated by extended Kalman filter are acquired in real time to form a multi-source time-series feature sequence. S22. Input the multi-source time-series feature sequence into the trained CNN-GRU model to obtain the observation confidence score at the current time. S23. Compare the observed confidence score with a preset threshold and execute an adaptive dual-track correction mechanism; S24. If the observation confidence score is higher than or equal to the preset threshold, then maintain the nominal observation noise covariance matrix and perform attitude closed-loop feedback correction. S25. If the observation confidence score is lower than the preset threshold, then pause the attitude feedback correction and dynamically adjust the observation noise covariance matrix of the extended Kalman filter according to the observation confidence score.

[0008] Optionally, the position information is the result of the difference between the GNSS position and the INS position after lever compensation, and the velocity information is the difference between the GNSS velocity and the INS velocity.

[0009] Optionally, the CNN-GRU model extracts local spatial features through a convolutional neural network, captures temporal dependencies through a gated recurrent unit, and outputs the observation confidence score using a sigmoid function.

[0010] Optionally, the multi-source time-series feature sequence includes a time window sequence consisting of position information, velocity information, and GNSS signal status for the current time and N times prior, where N is a positive integer.

[0011] Optionally, the pause attitude feedback correction includes: In the observation update of the extended Kalman filter, the estimated attitude error is not fed back to the attitude correction stage of the inertial navigation solution; only the position and velocity states are updated and fed back.

[0012] Optionally, the observation noise covariance matrix of the extended Kalman filter can be dynamically adjusted based on the observation confidence score, including: Calculate the dilated observation noise covariance matrix The expression is as follows: ; in, , is the nominal observation noise covariance matrix. and These represent the error variances of the three-dimensional position and velocity, respectively. This represents the inflation coefficient, which is a value greater than 1 and related to the observation confidence score. negative correlation This represents the isolation sensitivity hyperparameter, used to control the nonlinearity of the expansion coefficient.

[0013] Optionally, the extended Kalman filter employs a loosely coupled model based on error states, whose state vector includes three-dimensional position error, three-dimensional velocity error, three-dimensional attitude error, accelerometer zero bias, and gyroscope zero bias; the observation vector includes the position information and the velocity information.

[0014] In a second aspect, the present invention also provides a combined navigation adaptive attitude correction device, including a processor, a memory, an inertial measurement unit, and a GNSS receiver; The memory stores computer programs; When the processor executes the computer program, it implements the steps of the integrated navigation adaptive attitude correction method described above. The processor acquires data from the inertial measurement unit and the GNSS receiver to execute the integrated navigation adaptive attitude correction method.

[0015] In a third aspect, the present invention also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the above-described integrated navigation adaptive attitude correction method.

[0016] The application of the technical solution of the present invention has at least the following beneficial effects: This invention provides a method, device, and medium for adaptive attitude correction in integrated navigation. The method utilizes a hybrid neural network composed of a convolutional neural network (CNN) and a gated recurrent unit (GRU) network to extract EKF innovation sequences and GNSS signal features, and to evaluate measurement confidence in real time. It then performs attitude closed-loop feedback correction based on an established dual-track correction mechanism, or dynamically adjusts the EKF observation noise covariance matrix. This method can establish a high-precision attitude reference within the available GNSS window, cut off abnormal observation contamination paths, and significantly suppress error divergence in subsequent pure inertial navigation stages. Attached Figure Description

[0017] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the structures shown in these drawings without creative effort.

[0018] Figure 1 This is a flowchart illustrating the steps of the integrated navigation adaptive attitude correction method in a preferred embodiment of the present invention. Figure 2 This is a training / validation loss curve for model training in a preferred embodiment of the present invention; Figure 3 This is a training / validation accuracy curve of the model training in a preferred embodiment of the present invention; Figure 4 This is a confusion matrix diagram used to verify the test set of the model training in a preferred embodiment of the present invention. Figure 5 The diagram shows the navigation error of different methods under static navigation, where (a) represents the latitude error, (b) represents the longitude error, (c) represents the overall position error, (d) represents the north velocity error, (e) represents the east velocity error, and (f) represents the overall velocity error. Figure 6 The diagram shows attitude data after entering pure inertial navigation under different methods under static navigation. In the diagram, (a) represents the roll angle, (b) represents the pitch angle, and (c) represents the yaw angle. Figure 7The diagram shows the navigation error of different methods under dynamic navigation, where (a) represents the latitude error, (b) represents the longitude error, (c) represents the overall position error, (d) represents the north velocity error, (e) represents the east velocity error, and (f) represents the overall velocity error. Detailed Implementation

[0019] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0020] Example 1: like Figure 1 As shown, this embodiment provides a combined navigation adaptive attitude correction method, including the following steps (S1 model training and S2 real-time correction): S1. Model Training: The algorithm extracts the position and velocity information sequences from the extended Kalman filter and the signal state data from the Global Navigation Satellite System during integrated navigation as input features for the network model. The CNN-GRU model is then trained in a supervised manner using binary attitude quality labels generated from multi-scenario measured data. After sufficient fitting, the model's output layer no longer performs rigid binary classification decisions but instead outputs a continuous posterior probability value, i.e., a measurement confidence score.

[0021] In this embodiment, the model training process specifically includes: S11. Extract the position information sequence, velocity information sequence, and global navigation satellite system signal status data of the extended Kalman filter during the integrated navigation process to construct input feature samples; the position information is the result of the difference between GNSS position and INS position after lever compensation, and the velocity information is the difference between GNSS velocity and INS velocity.

[0022] S12. Generate binary pose quality labels using multi-scenario measured data, and perform supervised training on a hybrid neural network containing a convolutional neural network (CNN) and a gated recurrent unit (GRU) using the input feature samples and corresponding labels to obtain a trained CNN-GRU model; the CNN-GRU model outputs continuous posterior probability values ​​as observation confidence scores.

[0023] Furthermore, the CNN-GRU model extracts local spatial features through a convolutional neural network, captures temporal dependencies through a gated recurrent unit, and outputs the observation confidence score using a sigmoid function.

[0024] S2, Real-time calibration: S21. During the integrated navigation process, the position information, velocity information, and current GNSS signal status data calculated by the extended Kalman filter are acquired in real time to form a multi-source time-series feature sequence. The multi-source time-series feature sequence includes a time window sequence composed of the position information, velocity information, and GNSS signal status of the current time and N previous times, where N is a positive integer.

[0025] It should be noted that the extended Kalman filter adopts a loosely coupled model based on error states, and its state vector includes three-dimensional position error, three-dimensional velocity error, three-dimensional attitude error, accelerometer zero bias and gyroscope zero bias; the observation vector includes the position information and the velocity information.

[0026] In this embodiment, the Northeast Earth (NED) is selected as the navigation reference frame. (system), adopted The angular error model describes the INS error, and the combined zero bias error constitutes the INS error equation, which is expressed as: ; in, Indicates disturbance error; , and These represent the error vectors for position, velocity, and attitude, respectively. express The system is relative to the Earth system ECEF system ( The rotation vector of the system; This represents the Earth's angular velocity of rotation; Indicates the relative inertial frame of reference The rotation vector of the system; Represents the carrier coordinate system ( (system) to Rotation matrix of the system; This represents the measured value of the specific force under the load system; This indicates that the accelerometer has zero bias. This indicates that the gyroscope has zero bias; This indicates the error in the gravity model.

[0027] The INS error equation shows that position and velocity errors are coupled, and attitude errors... This can lead to errors in force projection, resulting in velocity errors, and the continuity equation of INS It can be represented as: ; in, Represents the system state transition matrix. Indicates system noise. The INS error state vector is represented as follows: .

[0028] Furthermore, the continuous observation equation of the extended Kalman filter... Z It can be represented as: ; in, Represents the observation vector; Represents the observation matrix; This indicates observation noise.

[0029] This embodiment uses the INS measurement center as a reference to compensate for the GNSS observation arm, and then uses the GNSS position and velocity for observation. The observation vector expression is as follows: ; ;

[0030] in, Indicates location information; Indicates speed information; and These represent the location results from GNSS and INS, respectively. express The rotational angular velocity of the system; Indicates GNSS in The pole arm is attached; and These represent the velocity results for GNSS and INS, respectively. This represents the n-dimensional identity matrix.

[0031] As can be seen from the observation vector expression, the innovation represents the error between the observed value and the predicted value. Statistical analysis of the innovation over continuous time can determine the quality of the observation information.

[0032] In this embodiment, state estimation can be achieved by iteratively calculating the state equation and observation equation using the EKF algorithm. The extended Kalman filter process includes two steps: state prediction and observation update, wherein: The expression for state prediction is as follows: ; ; The expression for observation update is as follows: ; ; ; in, Represents the state estimation vector; Represents the state transition matrix; Indicates the time when the system is running; Represents the process covariance matrix; Represents the EKF gain matrix; Represents the observation matrix; Represents the process noise covariance matrix; Represents the observation noise covariance matrix; Represents the actual observed vector; Represents the identity matrix; the subscripts in the above formula Indicates the current Time, subscript -1 means The moment before the previous moment.

[0033] S22. Input the multi-source time-series feature sequence into the trained CNN-GRU model to obtain the observation confidence score at the current time. S23. Compare the observed confidence score with a preset threshold and execute an adaptive dual-track correction mechanism; S24. If the observation confidence score is higher than or equal to the preset threshold, then maintain the nominal observation noise covariance matrix and perform attitude closed-loop feedback correction. S25. If the observation confidence score is lower than the preset threshold, then pause the attitude feedback correction and dynamically adjust the observation noise covariance matrix of the extended Kalman filter according to the observation confidence score.

[0034] In this embodiment, the pause attitude feedback correction includes: In the observation update of the extended Kalman filter, the estimated attitude error is not fed back to the attitude correction stage of the inertial navigation solution; only the position and velocity states are updated and fed back.

[0035] Furthermore, the observation noise covariance matrix of the extended Kalman filter is dynamically adjusted based on the observation confidence score, including: Calculate the dilated observation noise covariance matrix The expression is as follows: ; in, , is the nominal observation noise covariance matrix. and These represent the error variances of the three-dimensional position and velocity, respectively. This represents the inflation coefficient, which is a value greater than 1 and related to the observation confidence score. negative correlation This represents the isolation sensitivity hyperparameter, used to control the degree of nonlinearity of the expansion coefficient; This indicates a preset threshold.

[0036] In this embodiment, the execution logic of the adaptive dual-track correction mechanism is as follows: when At that time, the GNSS signal is determined to be superior, and the original observation noise covariance matrix is ​​maintained. And perform direct feedback correction of attitude; conversely, when If the GNSS signal is deemed contaminated, attitude feedback correction is paused, and the expanded observation noise covariance matrix is ​​used. Dynamic adjustments are made. Due to low confidence levels... Adaptive amplification, measurement update gain of Kalman filter This is significantly weakened, thus effectively cutting off the error propagation path of the cross-coupling contaminated attitude reference of abnormal observations.

[0037] This embodiment provides an adaptive attitude correction method for integrated navigation. During the integrated navigation phase, a CNN-GRU model is introduced to evaluate the GNSS observation confidence in real time, and based on this, adaptive attitude error feedback correction or dynamic adjustment of the EKF observation noise covariance matrix is ​​performed. This allows for the establishment of a high-precision attitude reference within the available GNSS window. Compared to traditional methods that use fixed thresholds to judge GNSS signal quality, the method in this embodiment utilizes a CNN-GRU model to extract innovation sequences and low-level GNSS features (such as PDOP value and satellite count), outputting continuous confidence scores to achieve soft decision-making and improve robustness in complex environments.

[0038] To verify the beneficial effects of the method in this embodiment, the following experimental tests were conducted: Experimental Equipment: Data was collected on a test platform equipped with a vehicle-mounted navigation system in various urban road scenarios (urban canyons, open areas, and tunnels). A laser-based INS was used and mounted on the centerline at the rear of the vehicle. Since the attitude accuracy of the laser INS used in this embodiment is significantly higher than that of a dual-antenna GNSS for a considerable period, a single-antenna GNSS solution was adopted, with the antenna mounted on the centerline at the rear of the vehicle. The INS data output frequency was 1000 Hz, the GNSS module data output frequency was 1 Hz, and soft synchronization was achieved using the satellite navigation pulses per second (PPS).

[0039] The vehicle-mounted test compared the navigation results of the following three methods: Method 1: Traditional EKF integrated navigation algorithm without attitude correction; Method 2: A posture feedback method based on time consistency heuristic rules; The method in this embodiment is: an adaptive pose correction method for EKF assisted by CNN-GRU.

[0040] Model training: Training label creation: Truth label creation is carried out using a time-consistency-based heuristic label. This heuristic rule has been rigorously verified in static and multi-scenario sports car tests. The results show that strictly following this rule to trigger attitude closed-loop feedback can effectively ensure the purity and accuracy of the attitude reference and improve the calculation accuracy of the system in the subsequent pure inertial navigation stage.

[0041] Truth value label extraction rules: Under integrated navigation, a sliding observation time window with a duration of 10 seconds is constructed backward from the current time. If, within this time window, the absolute values ​​of position information, absolute values ​​of velocity information, and GNSS signal (PDOP value, positioning solution status indication, number of visible satellites) of the integrated navigation system in the north, east, and ground dimensions are continuous and strictly meet the set threshold, the label value is 1; otherwise, it is 0.

[0042] The training dataset contains 15,000 sample points, stored chronologically at a GNSS receiving frequency of 1 Hz. Data collection scenarios include typical GNSS application scenarios such as urban canyons, tree-lined roads, and underpasses. A sliding window method was used to collect EKF position, velocity, and GNSS signal data (number of satellites, positioning mode, DPOP) at corresponding times, which were then used as input to the neural network in a 20×9 shape. Based on a time-order slicing method, the dataset was divided into training and validation sets at a ratio of 80% and 20%, respectively. The test dataset included 2,657 sample points not in the training set. The model was evaluated on the test set, completely independent of the training and validation processes.

[0043] like Figure 2 The training / validation loss curves shown indicate that the training loss steadily decreases from an initial 0.68 to below 0.25, while the validation loss remains consistently around 0.30. Figure 3 The training / validation accuracy curves shown indicate that the training accuracy rapidly increased from 56% to 90.3%, and the validation accuracy steadily increased from 47% to 87.5% and eventually stabilized, demonstrating that the model proposed in this embodiment has excellent fitting ability and generalization performance.

[0044] like Figure 4 The test set validation confusion matrix is ​​shown in Table 1. The statistical evaluation metrics for the test set are also shown in Table 1. The results of the test set validation evaluation metrics table indicate that the model proposed in this embodiment can accurately distinguish the purity of observation information and has a strong ability to distinguish between positive and negative samples. In summary, the CNN-GRU model constructed in this embodiment exhibits high accuracy, high robustness, and good probabilistic calibration characteristics in the observation information reliability identification task, and can meet the needs of practical navigation systems.

[0045] Table 1. Statistical Table of Test Set Validation Evaluation Metrics

[0046] AUC is a core metric used in machine learning to measure the performance of binary classification models.

[0047] After the model training is completed, a static navigation experiment will be conducted: The test vehicle was placed in an open, flat area, and the total duration of the static navigation test was 5800 seconds. The test procedure was as follows: initial alignment was completed at 400 seconds and the vehicle entered the combined navigation mode, after which it remained stationary for 1800 seconds; at 2200 seconds, the vehicle switched to pure inertial navigation mode to perform undamped navigation calculations until the test ended.

[0048] like Figure 5 The navigation error results shown are for three methods. Navigation error statistics are shown in Table 2.

[0049] Table 2. Navigation Error Statistics for Different Methods under Static Navigation

[0050] Experimental results show that, in pure inertial navigation mode, the root mean square error (RMSE) of the position error in this embodiment is 0.55 nautical miles (n mile), the RMSE of the velocity error is 0.836 m / s, the maximum position error is 0.868 n mile, and the maximum velocity error is 1.943 m / s. Compared with Method 1, the position RMSE of this embodiment is reduced by 31.8%, and the velocity RMSE is reduced by 23.8%; compared with Method 2, the position RMSE is reduced by 26.9%, and the velocity RMSE is reduced by 17.6%. Regarding the maximum error, compared with Method 1, the maximum position error of this embodiment is reduced by 28.5%, and the maximum velocity error is reduced by 15.3%; compared with Method 2, the maximum position error is reduced by 24.3%, and the maximum velocity error is reduced by 11.4%.

[0051] like Figure 6 The attitude data after entering pure inertia under different methods under static navigation are shown in Table 3. The standard deviation is used for quantitative analysis.

[0052] Table 3. Attitude data statistics after entering pure inertial navigation using different methods under static navigation.

[0053] Compared to Method 1, the method in this embodiment reduces the standard deviation (σ) of roll angle and yaw angle by 14.3% and 7.8%, respectively, while maintaining the same pitch angle as Method 2. Compared to Method 1, Method 2 reduces roll angle by 5.0%, yaw angle by 1.3%, and pitch angle by 1.2%. Considering the total standard deviation across all three axes, the method in this embodiment reduces the standard deviation by 6.5% compared to Method 1, which is superior to Method 2's 2.2%. These results demonstrate that the method in this embodiment effectively improves the accuracy of the attitude reference during the integrated navigation phase and significantly enhances the accuracy of pure inertial navigation.

[0054] Further dynamic navigation tests will be conducted: Initial alignment is completed in 400 seconds and the system enters integrated navigation mode, which continues for 1800 seconds. At 2200 seconds, the system switches to pure inertial navigation mode to perform undamped navigation calculations until the end.

[0055] like Figure 7 The results of pure inertial navigation errors are shown in Table 4. Navigation error statistics are also shown in Table 4.

[0056] Table 4. Navigation Error Statistics for Different Methods under Dynamic Navigation

[0057] Experimental results show that, compared with Method 1, the position RMSE is reduced by 20.0% and the velocity RMSE by 29.8%; compared with Method 2, the position RMSE is reduced by 16.4% and the velocity RMSE by 21.8%. Regarding the maximum error, the method in this embodiment reduces the maximum position error by 30.9% compared with Method 1 and by 23.3% compared with Method 2. These results verify the significant ability of the method in this embodiment to suppress the divergence of pure inertial navigation errors in dynamic and complex environments.

[0058] In addition, this embodiment also provides a combined navigation adaptive attitude correction device, including a processor, a memory, an inertial measurement unit and a GNSS receiver; The memory stores computer programs; When the processor executes the computer program, it implements the steps of the integrated navigation adaptive attitude correction method described above. The processor acquires data from the inertial measurement unit and the GNSS receiver to execute the integrated navigation adaptive attitude correction method.

[0059] It should be noted that computer program code for performing the operations of this application can be written in one or more programming languages ​​or a combination thereof. These programming languages ​​include object-oriented programming languages—such as Java, Smalltalk, and C++—as well as conventional procedural programming languages—such as the "C" language or similar programming languages. The program code can be executed entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving remote computers, the remote computer can be connected to the user's computer via any type of network—including a Local Area Network (LAN) or a Wide Area Network (WAN)—or can be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0060] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of this application. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, can be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.

[0061] In addition, this embodiment also provides a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the steps of the error hiding method described above.

[0062] The readable storage medium provided in this application is a computer-readable storage medium that stores computer-readable program instructions (i.e., a computer program) for performing the above-described error hiding method. Compared with the prior art, the beneficial effects of the computer-readable storage medium provided in this application are the same as the beneficial effects of the error hiding method provided in the above embodiments, and will not be repeated here.

[0063] The above description is only a preferred embodiment of the present invention and does not limit the scope of the present invention. All equivalent structural transformations made under the inventive concept of the present invention using the contents of the present invention specification and drawings, or direct / indirect applications in other related technical fields, are included within the protection scope of the present invention.

Claims

1. A combined navigation adaptive attitude correction method, characterized in that, This includes model training and real-time calibration, where: S1, Model Training: S11. Extract the position information sequence, velocity information sequence, and global navigation satellite system signal status data from the extended Kalman filter during integrated navigation to construct input feature samples; S12. Generate binary pose quality labels using multi-scenario measured data, and perform supervised training on a hybrid neural network containing convolutional neural networks and gated recurrent units using the input feature samples and corresponding labels to obtain the trained CNN-GRU model; the CNN-GRU model outputs continuous posterior probability values ​​as observation confidence scores; S2, Real-time calibration: S21. During integrated navigation, the position information, velocity information and current GNSS signal status data calculated by extended Kalman filter are acquired in real time to form a multi-source time-series feature sequence. S22. Input the multi-source time-series feature sequence into the trained CNN-GRU model to obtain the observation confidence score at the current time. S23. Compare the observed confidence score with a preset threshold and execute an adaptive dual-track correction mechanism; S24. If the observation confidence score is higher than or equal to the preset threshold, then maintain the nominal observation noise covariance matrix and perform attitude closed-loop feedback correction. S25. If the observation confidence score is lower than the preset threshold, then pause the attitude feedback correction and dynamically adjust the observation noise covariance matrix of the extended Kalman filter according to the observation confidence score.

2. The combined navigation adaptive attitude correction method according to claim 1, characterized in that, The position information is the result of the difference between the GNSS position and the INS position after lever arm compensation, and the velocity information is the difference between the GNSS velocity and the INS velocity.

3. The combined navigation adaptive attitude correction method of claim 1, wherein, The CNN-GRU model extracts local spatial features through a convolutional neural network, captures temporal dependencies through a gated recurrent unit, and outputs the observation confidence score using the sigmoid function.

4. The combined navigation adaptive attitude correction method of claim 1, wherein, The multi-source time-series feature sequence includes a time window sequence consisting of position information, velocity information, and GNSS signal status for the current time and N times prior, where N is a positive integer.

5. The combined navigation adaptive attitude correction method of claim 1, wherein, The pause attitude feedback correction includes: In the observation update of the extended Kalman filter, the estimated attitude error is not fed back to the attitude correction stage of the inertial navigation solution; only the position and velocity states are updated and fed back.

6. The combined navigation adaptive attitude correction method of claim 1, wherein, The observation noise covariance matrix of the extended Kalman filter is dynamically adjusted based on the observation confidence score, including: Computing the inflated observation noise covariance matrix The expression is as follows: ； wherein, is a nominal observation noise covariance matrix, and denote the error variances for three-dimensional position and velocity, respectively; denotes an inflation coefficient, the inflation coefficient being a value greater than 1 and being a function of the observation confidence score negatively correlated, denotes an isolation sensitivity hyperparameter for controlling the degree of nonlinearity of the inflation coefficient.

7. The integrated navigation adaptive attitude correction method according to claim 1, characterized in that, The extended Kalman filter employs a loosely coupled model based on error states, whose state vector includes three-dimensional position error, three-dimensional velocity error, three-dimensional attitude error, accelerometer zero bias, and gyroscope zero bias; the observation vector includes the position information and the velocity information.

8. A combined navigation adaptive attitude correction device, characterized in that, Includes a processor, memory, inertial measurement unit, and GNSS receiver; The memory stores computer programs; When the processor executes the computer program, it implements the steps of the integrated navigation adaptive attitude correction method according to any one of claims 1 to 7; The processor acquires data from the inertial measurement unit and the GNSS receiver to execute the integrated navigation adaptive attitude correction method.

9. A computer-readable storage medium, characterized in that, It stores a computer program that, when executed by a processor, implements the steps of the integrated navigation adaptive attitude correction method as described in any one of claims 1 to 7.

Images