Navigation enhancement method and system for integrated navigation during interruption

Abstract

The invention provides a navigation enhancement method and system for integrated navigation during interruption, and the method comprises the steps: introducing a time attention mechanism into an LSTM network to construct a TA-LSTM prediction model, and predicting a pseudo GNSS position increment to assist INS navigation. The model can autonomously extract key time sequence characteristics of historical time sequence information and dynamically adjust the attention of the key time sequence characteristics so as to improve the prediction precision of a long-time sequence. And secondly, integrating the RKF into the data fusion process of the pseudo GNSS and the INS, calculating a scale factor through a covariance matching mode, and carrying out adaptive expansion on a noise covariance matrix in the Kalman filter to reduce the influence of an accumulative error of model prediction noise on navigation performance. In the GNSS interruption period, the navigation performance of the integrated navigation system can be effectively improved, and the navigation performance is more excellent under long-time GNSS interruption.

Description

Technical Field

[0001] The present invention relates to the field of navigation technology, and particularly to a navigation enhancement method and system for integrated navigation during interruption. Background Art

[0002] The Global Navigation Satellite System / Inertial Navigation System (GNSS / INS) integrated navigation system is a classic integration mode. Among them, the GNSS has high continuous positioning accuracy, but is vulnerable to environmental factors. The INS navigates and locates through a recursive method, but there are cumulative errors. Therefore, by taking advantage of the GNSS and INS to complement each other, high-precision positioning can be achieved. The GNSS / INS integrated navigation system has been widely used in fields such as unmanned vehicles and robots.

[0003] Due to the vulnerability of GNSS signals, they are easily blocked by external environments such as tunnels, high-rise buildings, and canyons, resulting in interference or even interruption of GNSS signals. This will cause a significant decline in the navigation performance of the GNSS / INS integrated navigation system. When the number of satellites that can be captured is less than 4, the absolute positioning result of the GNSS receiver cannot be obtained, causing the GNSS / INS loose coupling system to directly degrade to a pure INS, resulting in the divergence of navigation errors over time. Although the GNSS / INS tight coupling system can assist the INS navigation when the number of visible satellites is less than 4 based on the raw observation data such as pseudorange and carrier phase provided by the GNSS receiver, when the GNSS signal is completely interrupted, the tight coupling system will also not be able to work properly. If the GNSS signal is interrupted for a long time without applying other methods to suppress the divergence of navigation errors, it will be fatal to the navigation device.

[0004] Currently, in the case of GNSS interruption, there are mainly three existing navigation enhancement methods. The first method is to adopt a GNSS / INS deep coupling system. By assisting the baseband tracking loop of the GNSS receiver with INS navigation information, the GNSS positioning result is obtained, and then combined with INS for integrated navigation. Higher navigation accuracy can be obtained during short-term GNSS interruption. This is a method of deep integration of satellite navigation and inertial navigation at the software and hardware levels, which requires readjusting the software and hardware structure of the GNSS receiver, with immature technology and high implementation costs. The second method is to add other sensors, such as odometers, magnetometers, lidar, and cameras, etc., to provide other measurement information for the integrated navigation system to constrain INS in order to improve navigation performance. However, this method undoubtedly increases the equipment cost and also makes the system more complex. The third method is to use artificial intelligence (AI) methods to assist INS navigation. By training the AI model with INS information to predict additional virtual measurement information, such as speed and position information, etc., stable and continuous pseudo-measurement information can be provided for the integrated navigation system during long-term GNSS interruption, and it does not change the GNSS / INS combination method and does not require additional hardware costs, which is a good and effective navigation enhancement method. Summary of the Invention

[0005] The present invention provides a navigation enhancement method and system for integrated navigation during interruption to solve the defects existing in the prior art.

[0006] In a first aspect, the present invention provides a navigation enhancement method for integrated navigation during interruption, including: Construct a TA-LSTM prediction model for predicting pseudo GNSS position information; When the GNSS signal exists and is reliable, perform GNSS / INS integrated navigation based on the Kalman filter and train the TA-LSTM prediction model; When the GNSS signal is interrupted, use the trained TA-LSTM prediction model to predict the pseudo GNSS position increment, obtain the pseudo GNSS position information of the current epoch through increment accumulation, and obtain the optimal filtering state based on the error state Kalman filter, and output the navigation position information.

[0007] According to the navigation enhancement method for integrated navigation during interruption provided by the present invention, constructing a TA-LSTM prediction model for predicting pseudo GNSS position information includes: Determine that the TA-LSTM prediction model includes an input layer, an LSTM layer, a dense layer, and an output layer; Input the feature sequence into the input layer, where represents the angular rate, and represents the specific force. denotes the INS speed, denotes the attitude information, which is normalized and then fed into the LSTM layer to capture temporal correlations. Calculate and allocate the temporal attention weights for the state vectors output by the LSTM layer, perform linear summation processing in the dense layer, and after denormalization, the prediction output is performed by the output layer.

[0008] According to a navigation enhancement method for integrated navigation during interruption provided by the present invention, which is normalized and then fed into the LSTM layer to capture temporal correlations, calculate and allocate the temporal attention weights for the state vectors output by the LSTM layer, perform linear summation processing in the dense layer, and after denormalization, the prediction output is performed by the output layer, including: Determine the number of layers and the number of hidden layer units of the LSTM, and add a Dropout layer for regularization; Calculate the correlation scores between the LSTM hidden state and each time step in the input feature sequence, normalize the attention scores for different time steps to obtain the attention weights. The attention weights represent the degree of association between the current hidden state and different time steps of the input feature sequence. At each time step, weight the LSTM hidden state and the corresponding attention weights to obtain the hidden state after weight allocation. The temporal attention mechanism is expressed as:

[0009] where, and are the temporal attention scores and weights respectively, and are the parameters of the attention layer, is the LSTM hidden layer state at the t-th time step, is the hidden state after attention weight allocation; Perform linear summation of the hidden state after attention weight allocation in the dense layer:

[0010] At the output layer, after activation function processing, the predicted values of the latitude increment, longitude increment, and altitude increment are obtained after denormalization; Adopt the Adam algorithm as the model optimizer, use the mean squared error as the loss function of the prediction model, and adjust the network weights inside the TA-LSTM model until the value of the loss function is minimized. The loss function is expressed as:

[0011] where, M is the number of samples, is the pseudo GNSS position increment obtained by model prediction, The true GNSS position increment in the collected dataset.

[0012] According to a navigation enhancement method for integrated navigation during interruption provided by the present invention, when the GNSS signal is present and reliable, GNSS / INS integrated navigation is performed based on a Kalman filter, and a TA-LSTM prediction model is trained, including: Input the position information obtained by INS mechanical arrangement and the position information solved by GNSS into the Kalman filter for data fusion to obtain the pose result of integrated navigation and perform error correction; Store the specific force provided by the IMU and the angular rate , as well as the INS velocity and attitude obtained by INS mechanical arrangement, and extract the stored data of multiple historical epochs and the current epoch GNSS increment data and input them into the TA-LSTM model for training.

[0013] According to a navigation enhancement method for integrated navigation during interruption provided by the present invention, input the position information obtained by INS mechanical arrangement and the position information solved by GNSS into the Kalman filter for data fusion to obtain the pose result of integrated navigation and perform error correction, including: Determine the system state vector as the position error, velocity error, attitude error related to the navigation state, and the zero biases of the three-axis gyroscopes and three-axis accelerometers related to the sensor errors, a total of 15 dimensions:

[0014] Among them, represents the error state vector, represents the position error vector, represents the velocity error vector, represents the attitude error vector, represents the zero bias vector of the three-axis gyroscopes, represents the zero bias vector of the three-axis accelerometers; The state equation of system discretization is:

[0015]

[0016] Among them, is the system discretization state transition matrix from time k to k - 1, is the system excitation noise, is the process noise covariance matrix; Define the difference between the position vector calculated by INS mechanical alignment and the GNSS position vector as the observation vector of the Kalman filter , and the Kalman filter observation equation is:

[0017]

[0018] where, represents the Kalman filter design matrix, represents the Kalman filter observation vector noise, is the measurement noise covariance matrix; The Kalman filter time update equation is:

[0019]

[0020] where, is the one-step predicted state value, is the state prediction covariance matrix, is the state covariance matrix of the optimal estimate of the previous moment's state.

[0021] The Kalman filter measurement update equation is:

[0022]

[0023]

[0024] where, is the Kalman gain, is the optimal estimate of the current moment's state, is the optimal estimate covariance matrix of the state.

[0025] According to a navigation enhancement method for integrated navigation during interruption provided by the present invention, when the GNSS signal is interrupted, a trained TA-LSTM prediction model is used to predict the pseudo-GNSS position increment, the pseudo-GNSS position information of the current epoch is obtained through increment accumulation, and the optimal filtering state is obtained based on the error state Kalman filter, and the navigation position information is output, including: Starting from the first epoch after the GNSS signal interruption, accumulate the pseudo-GNSS position increments predicted by the model to obtain the pseudo-GNSS position of any epoch after the interruption. The formula for the pseudo-GNSS position predicted by the model can be expressed as:

[0026] Wherein, is the position at the previous epoch before the GNSS signal interruption; After the GNSS signal is interrupted, the pseudo-GNSS position information is predicted by the AI model, and the prediction noise error between the pseudo-GNSS position information and the GNSS true position information is calculated:

[0027] Wherein, is the predicted position noise error of the model; Perform Sage-Husa moving window estimation on the pseudo-GNSS position information, calculate the scale factor through covariance matching to achieve adaptive estimation of the noise covariance matrix, and obtain the optimal filtering state after time update and measurement update.

[0028] According to a navigation enhancement method for integrated navigation during interruption provided by the present invention, performing Sage-Husa moving window estimation on the pseudo-GNSS position information, calculating the scale factor through covariance matching to achieve adaptive estimation of the noise covariance matrix, and obtaining the optimal filtering state after time update and measurement update, including: In the AI / INS integrated navigation system, the pseudo-GNSS position information vector is expressed as:

[0029] Wherein, , according to the covariance propagation law, the theoretical value of the covariance matrix of the pseudo-GNSS position innovation vector at the current epoch is:

[0030] Perform moving window estimation on the pseudo-GNSS position innovation vector, and take equal-weighted averaging of the innovation vectors within the window. Assuming that the information vector is normally distributed, for the information vector with a window of d, the estimated value of the covariance matrix of the pseudo-GNSS position information vector at the current epoch is:

[0031] Determine that the theoretical value and the estimated value of the covariance matrix of the pseudo-GNSS position innovation vector are approximately equal, that is , and the estimated measurement noise covariance matrix is expressed as:

[0032] Inflate the preset measurement noise covariance matrix through the robust factor. If the estimated covariance matrix matches the theoretical value, the traces of the two are equal; otherwise, there is a deviation in the traces of the two. Among them, the robust factor is:

[0033] wherein, is the trace of the matrix, is a preset empirical threshold, is the prior measurement noise covariance matrix, obtained by predicting experimental statistics through an AI model; The actually estimated measurement noise covariance matrix at the current epoch is expressed as:

[0034] The inflated estimated measurement noise covariance matrix is sent into the measurement update process to obtain the optimal state estimate at the current epoch, expressed as:

[0035] According to the covariance propagation law, the process noise covariance matrix is estimated, specifically expressed as:

[0036] A scale factor is constructed through the operation of the trace, and the preset process noise covariance matrix is inflated, wherein the scale factor is:

[0037] wherein, is a preset empirical threshold, is the prior process noise covariance matrix, obtained by estimating through the Allan variance method; The actually estimated process noise covariance matrix at the current epoch can be expressed as:

[0038] The inflated process noise covariance matrix is stored and used for the time update process of the next epoch.

[0039] In a second aspect, the present invention also provides a navigation enhancement system for integrated navigation during interruption, including: A construction module, configured to construct a TA-LSTM prediction model for predicting pseudo GNSS position information; A first enhancement module, configured to perform GNSS / INS integrated navigation based on a Kalman filter and train the TA-LSTM prediction model when the GNSS signal is present and reliable; A second enhancement module, configured to predict the pseudo GNSS position increment by using the trained TA-LSTM prediction model when the GNSS signal is interrupted, obtain the pseudo GNSS position information at the current epoch through increment accumulation, and obtain the optimal filtering state based on the error state Kalman filter, and output the navigation position information.

[0040] In a third aspect, the present invention further provides an electronic device, including a memory, a processor, and a computer program stored on the memory and executable on the processor. When the processor executes the program, the navigation enhancement method for integrated navigation during interruption as described in any one of the above is implemented.

[0041] In a fourth aspect, the present invention further provides a non-transitory computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, the navigation enhancement method for integrated navigation during interruption as described in any one of the above is implemented.

[0042] The navigation enhancement method and system for integrated navigation during interruption provided by the present invention predict the pseudo GNSS position increment by constructing a TA-LSTM model, and make the LSTM network pay more attention to important temporal features during the learning process by introducing a time attention mechanism, dynamically adjusting the attention degree at different time steps of the LSTM network, so as to improve the utilization efficiency and prediction accuracy of long time series data; also propose an RKF data fusion algorithm based on the pseudo GNSS position innovation, perform Sage-Husa moving window estimation on the pseudo GNSS position innovation, and adaptively inflate the preset noise covariance matrix by means of covariance matching to reduce the influence of the prediction noise error of the model on the navigation performance and improve the robustness of the AI / INS integrated navigation system. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] In order to more clearly illustrate the technical solutions in the present invention or the prior art, the following will briefly introduce the drawings required for the description of the embodiments or the prior art. Obviously, the drawings in the following description are some embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0044] Figure 1 is a schematic flowchart of the navigation enhancement method for integrated navigation during interruption provided by the present invention; Figure 2 is a block diagram of a GNSS / INS loosely coupled navigation system provided by the present invention; Figure 3 is a network structure diagram of a TA-LSTM prediction model provided by the present invention; Figure 4 is a framework diagram of a GNSS / INS integrated navigation system during the training stage of the TA-LSTM model provided by the present invention; Figure 5 is a framework diagram of an AI / INS integrated navigation system during the application stage of the TA-LSTM model provided by the present invention; Figure 6It is the block diagram of the RKF algorithm based on the pseudo-GNSS position innovation provided by the present invention; Figure 7 It is the structural schematic diagram of the navigation enhancement system for integrated navigation during interruption provided by the present invention; Figure 8 It is the structural schematic diagram of the electronic device provided by the present invention. Specific implementation manners

[0045] To make the objectives, technical solutions and advantages of the present invention clearer, the technical solutions in the present invention will be clearly and completely described below with reference to the accompanying drawings in the present invention. Obviously, the described embodiments are some but not all of the embodiments of the present invention. All other embodiments obtained by those of ordinary skill in the art based on the embodiments in the present invention without creative efforts shall fall within the protection scope of the present invention.

[0046] Aiming at the problem that the navigation error of the GNSS / INS integrated navigation system continues to diverge during the GNSS interruption in the prior art, a navigation enhancement method based on Time Attention-Long Short Term Memory (TA-LSTM) and Robust Kalman filter (RKF) is proposed for the GNSS / INS loose coupling system. The main objective is to improve the prediction accuracy of the pseudo-GNSS position increment and reduce the cumulative error of the model prediction noise on the navigation performance, so as to improve the navigation performance of the GNSS / INS integrated navigation system during the GNSS interruption.

[0047] Figure 1 It is the flow schematic diagram of the navigation enhancement method for integrated navigation during interruption provided by the embodiment of the present invention, as Figure 1 shown, including: Step 100: Construct a TA-LSTM prediction model for predicting pseudo-GNSS position information; Step 200: When the GNSS signal exists and is reliable, perform GNSS / INS integrated navigation based on the Kalman filter and train the TA-LSTM prediction model; Step 300: When the GNSS signal is interrupted, use the trained TA-LSTM prediction model to predict the pseudo-GNSS position increment, obtain the pseudo-GNSS position information of the current epoch through increment accumulation, and obtain the optimal filtering state based on the error state Kalman filter, and output the navigation position information.

[0048] Specifically, as Figure 2As shown in the figure, the system consists of a GNSS receiver, an IMU, and a Kalman filter. The IMU obtains real-time position, velocity, and attitude through a mechanical alignment algorithm. The GNSS receiver provides position observations, and data fusion is performed through the Kalman filter to correct the navigation results of INS mechanization, providing a strong guarantee for the vehicle integrated navigation system and greatly improving the positioning accuracy of the vehicle integrated navigation system.

[0049] In an embodiment of the present invention, a TA-LSTM prediction model is constructed by introducing a time attention mechanism into the LSTM network to adaptively adjust the attention of the LSTM network at different time steps. By inputting angular rate, specific force, INS velocity, and attitude information into the TA-LSTM prediction model, latitude, longitude, and elevation increments are predicted to assist INS navigation, and the scale factor is calculated through covariance matching to adaptively inflate the noise covariance matrix in the Kalman filter to reduce the impact of the prediction noise error of the pseudo-GNSS position on the navigation performance.

[0050] In one embodiment, as Figure 3 shown, it is the network structure diagram of the TA-LSTM prediction model. First, the feature data composed of the feature sequence is sent into the input layer, and after normalization, it is sent into the LSTM layer to capture time correlation. Then, the time attention weights of the state vector output by the LSTM layer are calculated and allocated, and then linear summation processing is performed in the dense layer. Finally, the prediction output is obtained through denormalization.

[0051] Specifically, first, the feature data composed of the feature sequence is sent into the input layer, represents angular rate, represents specific force, represents INS velocity, and represents attitude information. After normalization, it is sent into the LSTM layer to capture time correlation. Then, the time attention weights of the state vector output by the LSTM layer are calculated and allocated, and then linear summation processing is performed in the dense layer. Finally, the prediction output is obtained through denormalization.

[0052] Generally, the output frequency of GNSS is 1Hz, while the sampling frequency of IMU is much higher than that of GNSS. Assuming the sampling frequency of IMU is nHz, then within one GNSS cycle (1s), there will be n groups of feature sequences. Since the current epoch GNSS position observation increment is also related to the navigation data of historical epochs, it is necessary to input more historical feature sequences, that is, it is necessary to design a certain time step m so that the model can better learn the complex input-output non-linear correspondence relationship. Therefore, the dimension of the feature data input to the model is Dimension, that is, predicting the pseudo-GNSS position increment of the current epoch through the feature sequence within the first m GNSS cycles.

[0053] The design of the LSTM layer is the most important part of the entire prediction model, determining the prediction accuracy of the model. The hyperparameters that have the greatest impact on the prediction accuracy are the number of LSTM layers and the number of hidden layer units. Too many hidden layer units will reduce the correlation of the training data in the time dimension, while too few units will not be able to fully learn the non-linear features of the input and output. Too many layers will result in overfitting, while insufficient layers will reduce the generalization performance of the model. At the same time, to prevent overfitting, adding a Dropout layer can achieve a regularization effect.

[0054] The core idea of the attention mechanism is to allow the model to focus on the key parts of the input sequence during the decision-making process of the model. By adaptively assigning weights to each input parameter, it is used for weighting the input sequence. Since the LSTM network performs poorly in highlighting important information, a temporal attention mechanism is introduced to model the correlation between different time steps of the LSTM network, enabling the model to dynamically adjust the attention at different time steps of the LSTM network, thereby improving the utilization efficiency of temporal information. At each time step, the LSTM generates a hidden state that contains the sequence information of the current time step. For the LSTM hidden state, calculate the correlation score between it and each time step in the input sequence, and normalize the attention scores of different time steps to obtain the corresponding attention weights, which represent the degree of association between the current hidden state and different time steps in the input sequence. Finally, at each time step, weight the LSTM hidden state and its corresponding attention weights to obtain the hidden state after weight assignment. The calculation formula of the temporal attention mechanism can be expressed as:

[0055] Among them, and are the temporal attention score and weight respectively, and are the parameters of the attention layer, is the LSTM hidden layer state at the t-th time step, is the hidden state after attention weight assignment.

[0056] The role of the dense layer is to perform a linear summation on the hidden state after attention weight assignment. The calculation formula of the dense layer can be expressed as:

[0057] At the output layer After the activation function processing, the predicted values of the latitude increment, longitude increment, and elevation increment can be obtained through inverse normalization. To make the prediction efficiency and accuracy of the model higher, this paper selects the Adam algorithm as the model optimizer and uses the Mean Square Error (MSE) as the loss function of the prediction model. By continuously adjusting the network weights inside the TA-LSTM model until the value of the loss function is minimized. The loss function for pseudo-GNSS position increment prediction can be expressed as:

[0058] where M is the number of samples, is the pseudo-GNSS position increment obtained by model prediction, is the true GNSS position increment in the collected dataset.

[0059] In one embodiment, as Figure 4 shown, it is the GNSS / INS integrated navigation system framework during the training stage of the TA-LSTM model. When the GNSS signal is present and reliable, the GNSS / INS integrated navigation system operates in the model training stage. As Figure 4 shown, the position information obtained by INS mechanical alignment and the position information solved by GNSS are input into the KF for data fusion to obtain the pose result of integrated navigation and perform error correction. During the operation of the GNSS / INS integrated navigation system, at the same time, the specific force and angular rate data provided by the IMU, the speed and attitude data obtained by INS mechanical alignment are stored, and the stored data of multiple historical epochs are taken out and input into the TA-LSTM model together with the current epoch GNSS increment data (latitude increment

[0060] longitude increment elevation increment

[0061] for training. Specifically, in order to fuse the INS mechanization result with the GNSS measurement value, give full play to the advantages of each system and make up for their respective disadvantages, the present invention adopts indirect Kalman filtering (error-state Kalman filtering), which can solve the nonlinear problem of the system. The state vector denotes the error state vector, denotes the position error vector, denotes the velocity error vector, denotes the attitude error vector, denotes the zero bias vector of the three-axis gyroscope, denotes the zero bias vector of the three-axis accelerometer.

[0062] The state equation of the system discretization is:

[0063]

[0064] where, is the state transition matrix of the system discretization from time k to k-1, is the system excitation noise, is the process noise covariance matrix.

[0065] In the present invention, the difference between the position vector calculated by the INS mechanical arrangement and the GNSS position vector is defined as the observation vector of the Kalman filter , and the observation equation of the Kalman filter is:

[0066]

[0067] where, represents the Kalman filter design matrix, represents the Kalman filter observation vector noise, is the measurement noise covariance matrix.

[0068] The Kalman filter is mainly divided into two parts. The first part is the prediction based on the state error, called the time update, and the Kalman filter time update equation is:

[0069]

[0070] where, is the one-step predicted state value, is the state prediction covariance matrix, is the state covariance matrix of the optimal estimated value of the previous state.

[0071] The second part is the measurement update process based on the measurement error, and the Kalman filter measurement update equation is:

[0072]

[0073]

[0074] Among them, is the Kalman gain, is the optimal state estimate at the current moment, is the covariance matrix of the optimal state estimate.

[0075] In one embodiment, as Figure 5 shown, it is the AI / INS integrated navigation system framework in the application stage of the TA-LSTM model. When the GNSS signal is interrupted, the AI / INS integrated navigation system works in the model application stage, as Figure 5 shown. The specific force, angular rate, velocity, and attitude data of multiple stored historical epochs are input into the trained TA-LSTM model to obtain the pseudo-latitude increment, pseudo-longitude increment, and pseudo-elevation increment of the current epoch. And starting from the first epoch after the GNSS signal interruption, the pseudo-GNSS position increment predicted by the TA-LSTM model is accumulated to obtain the pseudo-GNSS position information, and the RKF is used to perform data fusion on the position information obtained by the INS mechanical alignment to obtain the optimal pose result.

[0076] Specifically, starting from the first epoch after the GNSS signal interruption, the pseudo-GNSS position increment predicted by the model is accumulated to obtain the pseudo-GNSS position at any epoch after the interruption. The formula for the pseudo-GNSS position predicted by the model can be expressed as:

[0077] Among them, is the position at the previous epoch before the GNSS signal interruption.

[0078] When the GNSS signal is interrupted, the AI model is used to predict the pseudo-GNSS position information. Due to the inevitable introduction of model prediction noise, there is a certain error between the pseudo-GNSS and the true GNSS value, and the prediction noise error accumulates with the increase of the GNSS interruption time. The prediction noise error of the pseudo-GNSS position can be expressed as:

[0079] Among them, is the prediction position noise error of the model.

[0080] If the standard KF is persistently used to perform data fusion on the pseudo-GNSS position information and the position information obtained by the INS mechanical alignment, since the actual prediction noise does not match the preset noise covariance matrix, the optimal pose result cannot be obtained. To further improve the positioning accuracy and considering the influence of the prediction noise, the present invention proposes an RKF algorithm based on the innovation of the pseudo-GNSS position, as Figure 6As shown, when the GNSS signal is interrupted, the AI model is used to predict the pseudo-GNSS position information. Due to the inevitable introduction of the prediction noise of the model, there is a certain error between the pseudo-GNSS and the true GNSS value, and the prediction noise error accumulates with the increase of the GNSS interruption time. To further improve the positioning accuracy and considering the influence of the prediction noise, the present invention performs Sage-Husa moving window estimation on the pseudo-GNSS position innovation, and calculates the scale factor by covariance matching to realize the adaptive estimation of the noise covariance matrix. Then, after the time update and measurement update processes, the optimal filtering state will be obtained, and finally the optimal positioning result will be obtained.

[0081] In the AI / INS integrated navigation system, the pseudo-GNSS position information vector can be expressed as:

[0082] where , and according to the covariance propagation law, the theoretical value of the covariance matrix of the pseudo-GNSS position innovation vector at the current epoch is:

[0083] To avoid the influence of excessive noise on the system state, moving window estimation is performed on the pseudo-GNSS position innovation vector, and equal weighted averaging is taken for the innovation vectors within the window. Based on the assumption that the information vector is normally distributed, for the innovation vector with a window of d, the estimated value of the covariance matrix of the pseudo-GNSS position innovation vector at the current epoch is:

[0084] To further estimate the noise covariance matrix, since the divergence error of the dynamic model is small within the epoch interval time, the present invention believes that the theoretical value and the estimated value of the covariance matrix of the pseudo-GNSS position innovation vector are approximately equal, that is . Therefore, the estimated measurement noise covariance matrix can be expressed as:

[0085] To prevent the situation that the filtering algorithm cannot be executed due to the non-positive definiteness of the matrix, the present invention designs a robust factor to inflate the preset measurement noise covariance matrix. If the estimated covariance matrix matches the theoretical value, then the traces of the two are equal. Conversely, there is a deviation in their traces. Therefore, the present invention can construct a robust factor through the operation of the trace. The robust factor constructed by the present invention is as follows:

[0086] where is the trace of the matrix, is a preset empirical threshold, is the prior measurement noise covariance matrix, which is obtained by predicting experimental statistics through the AI model in the present invention. Therefore, the actually estimated measurement noise covariance matrix at the current epoch can be expressed as:

[0087] Sending the inflated estimated measurement noise covariance matrix into the measurement update process, the optimal state estimate at the current epoch can be obtained, which can be expressed as:

[0088] According to the covariance propagation law, the present invention can estimate the process noise covariance matrix, and the specific formula can be expressed as:

[0089] Meanwhile, by performing operations on the trace to construct a scaling factor and inflating the preset process noise covariance matrix, the scaling factor constructed by the present invention is as follows:

[0090] wherein, is the preset empirical threshold, is the prior process noise covariance matrix, which is estimated by the Allan variance method in the present invention. Therefore, the actually estimated process noise covariance matrix at the current epoch can be expressed as:

[0091] Storing the inflated process noise covariance matrix and using it for the time update process of the next epoch.

[0092] The navigation enhancement system for integrated navigation during interruption provided by the present invention will be described below. The navigation enhancement system for integrated navigation during interruption described below can be correspondingly referred to the navigation enhancement method for integrated navigation during interruption described above.

[0093] Figure 7 is a schematic structural diagram of the navigation enhancement system for integrated navigation during interruption provided by an embodiment of the present invention. As Figure 7 shown, it includes: a construction module 71, a first enhancement module 72, and a second enhancement module 73, wherein: The building module 71 is used to build a TA-LSTM prediction model for predicting pseudo GNSS position information; the first enhancement module 72 is used to perform GNSS / INS integrated navigation based on a Kalman filter and train the TA-LSTM prediction model when the GNSS signal is present and reliable; the second enhancement module 73 is used to predict the pseudo GNSS position increment using the trained TA-LSTM prediction model when the GNSS signal is interrupted, obtain the pseudo GNSS position information of the current epoch through increment accumulation, and obtain the optimal filtering state based on the error state Kalman filter to output the navigation position information.

[0094] Figure 8 An entity structure diagram of an electronic device is exemplified, as [[ID= shown. The electronic device may include: a processor 810, a communications interface 820, a memory 830, and a communication bus 840. Among them, the processor 8, the communications interface 820, and the memory 830 complete mutual communication through the communication bus 840. The processor 810 can call the logical instructions in the memory 830 to execute a navigation enhancement method for integrated navigation during interruption, and the method includes: building a TA-LSTM prediction model for predicting pseudo GNSS position information; when the GNSS signal is present and reliable, performing GNSS / INS integrated navigation based on a Kalman filter and training the TA-LSTM prediction model; when the GNSS signal is interrupted, predicting the pseudo GNSS position increment using the trained TA-LSTM prediction model, obtaining the pseudo GNSS position information of the current epoch through increment accumulation, and obtaining the optimal filtering state based on the error state Kalman filter to output the navigation position information.

[0095] In addition, when the logical instructions in the above-mentioned memory 830 can be implemented in the form of software functional units and sold or used as an independent product, they can be stored in a computer-readable storage medium. Based on such an understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or a part of this technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes several instructions for causing a computer device (which may be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in various embodiments of the present invention. The aforementioned storage medium includes: various media such as a USB flash drive, a mobile hard disk, a read-only memory (ROM, Read-Only Memory), a random access memory (RAM, Random Access Memory), a magnetic disk, or an optical disc that can store program codes.

[0096] On the other hand, the present invention also provides a non-transitory computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, it implements a navigation enhancement method for integrated navigation during interruption, and the method includes: constructing a TA-LSTM prediction model for predicting pseudo-GNSS position information; when the GNSS signal exists and is reliable, performing GNSS / INS integrated navigation based on a Kalman filter, and training the TA-LSTM prediction model; when the GNSS signal is interrupted, using the trained TA-LSTM prediction model to predict the pseudo-GNSS position increment, obtaining the pseudo-GNSS position information of the current epoch through increment accumulation, and obtaining the optimal filtering state based on an error-state Kalman filter, and outputting navigation position information.

[0097] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place, or may be distributed to multiple network units. Some or all of the modules can be selected according to actual needs to achieve the purpose of the solution of this embodiment. Those of ordinary skill in the art can understand and implement it without creative efforts.

[0098] Through the description of the above embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus a necessary general hardware platform, and of course, it can also be implemented by hardware. Based on such an understanding, the above technical solution, in essence, or the part that contributes to the prior art, can be embodied in the form of a software product. The computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or some parts of the embodiments.

[0099] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions described in the foregoing embodiments, or perform equivalent replacements for some of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of each embodiment of the present invention.

Claims

1. A navigation enhancement method for integrated navigation during interruption, characterized in that, Including: Construct a TA-LSTM prediction model for predicting pseudo GNSS position information; When the GNSS signal is present and reliable, perform GNSS / INS integrated navigation based on the Kalman filter and train the TA-LSTM prediction model; When the GNSS signal is interrupted, use the trained TA-LSTM prediction model to predict the pseudo GNSS position increment, obtain the pseudo GNSS position information of the current epoch through increment accumulation, and obtain the optimal filtering state based on the error-state Kalman filter to output the navigation position information.

2. The method for enhancing navigation during interruption for integrated navigation according to claim 1, wherein Construct a TA-LSTM prediction model for predicting pseudo GNSS position information, including: Determine that the TA-LSTM prediction model includes an input layer, an LSTM layer, a dense layer, and an output layer; Input the feature sequence into the input layer, which represents the angular rate, represents the specific force, represents the INS velocity, and represents the attitude information. After normalization, it is sent to the LSTM layer to capture temporal correlations. Calculate and allocate the temporal attention weights for the state vectors output by the LSTM layer, perform linear summation processing in the dense layer, and after denormalization, the prediction output is made by the output layer.

3. The method for enhancing navigation during interruption for integrated navigation according to claim 2, wherein After normalization, it is sent to the LSTM layer to capture time correlation, calculate and allocate the time attention weights for the state vectors output by the LSTM layer, perform linear summation processing in the dense layer, and after inverse normalization, the prediction output is performed by the output layer, including: Determine the number of layers and hidden layer units of the LSTM, and add a Dropout layer for regularization; Calculate the correlation score between the LSTM hidden state and each time step in the input feature sequence, normalize the attention scores at different time steps to obtain the attention weights, and the attention weights represent the degree of association between the current hidden state and different time steps of the input feature sequence. At each time step, the LSTM hidden state and the corresponding attention weights are weighted to obtain the hidden state after weight allocation. The time attention mechanism is expressed as: wherein, and are the temporal attention score and weight respectively, and are the parameters of the attention layer, is the LSTM hidden layer state at the t-th time step, is the hidden state after attention weight assignment; Perform linear summation on the hidden state after attention weight allocation in the dense layer: After the activation function processing in the output layer, the latitude increment prediction value, longitude increment prediction value, and elevation increment prediction value are obtained through inverse normalization; Use the Adam algorithm as the model optimizer, use the mean square error as the loss function of the prediction model, and adjust the network weights inside the TA-LSTM model until the value of the loss function reaches the lowest. The loss function is expressed as: where M is the number of samples, is the pseudo GNSS position increment obtained by model prediction, is the true GNSS position increment in the collected dataset.

4. The method for enhancing navigation during interruption for integrated navigation according to claim 1, wherein When the GNSS signal is present and reliable, perform GNSS / INS integrated navigation based on the Kalman filter and train the TA-LSTM prediction model, including: The position information obtained by INS mechanical scheduling and the position information obtained by GNSS solution are input into the Kalman filter for data fusion to obtain the pose result of integrated navigation and perform error correction; The specific force provided by the IMU and the angular rate , as well as the INS velocity and attitude obtained by the INS mechanical alignment are stored, and the stored data of multiple historical epochs and the current epoch GNSS incremental data are extracted and input into the TA-LSTM model for training.

5. The method for enhancing navigation during interruption for integrated navigation according to claim 4, wherein The position information obtained by INS mechanical scheduling and the position information obtained by GNSS solution are input into a Kalman filter for data fusion to obtain the pose result of integrated navigation and perform error correction, including: Determine the system state vector It is the position error, velocity error, attitude error related to the navigation state, and the zero biases of the three-axis gyroscopes and three-axis accelerometers related to the sensor errors, with a total of 15 dimensions: Among them, represents the error state vector, represents the position error vector, represents the velocity error vector, represents the attitude error vector, represents the zero bias vector of the three-axis gyroscope, represents the zero bias vector of the three-axis accelerometer; The state equation of system discretization is: Among them, is the state transition matrix for the discretization of the system from time k to time k - 1, is the system excitation noise, is the process noise covariance matrix; Define the difference between the position vector calculated by INS mechanical scheduling and the GNSS position vector as the observation vector of the Kalman filter , and the observation equation of the Kalman filter is as follows: Among them, represents the Kalman filter design matrix, represents the Kalman filter observation vector noise, is the measurement noise covariance matrix; The Kalman filter time update equation is: Among them, is the one-step predicted state value, is the state prediction covariance matrix, is the state covariance matrix of the optimal estimate of the previous state; The Kalman filter measurement update equation is: Among them, is the Kalman gain, is the optimal state estimate at the current moment, is the covariance matrix of the optimal state estimate.

6. The method for enhancing navigation during interruption for integrated navigation according to claim 1, wherein When the GNSS signal is interrupted, use the trained TA-LSTM prediction model to predict the pseudo GNSS position increment, obtain the pseudo GNSS position information of the current epoch through increment accumulation, and obtain the optimal filtering state based on the error-state Kalman filter to output the navigation position information, including: Starting from the first epoch after the GNSS signal interruption, accumulate the pseudo GNSS position increments predicted by the model to obtain the pseudo GNSS position at any epoch after the interruption. The formula for the pseudo GNSS position predicted by the model can be expressed as: Among them, is the position at the previous epoch before the GNSS signal interruption; After the GNSS signal is interrupted, predict the pseudo GNSS position information through the AI model, and calculate the prediction noise error between the pseudo GNSS position information and the GNSS true position information: Among them, is the predicted position noise error of the model; Perform Sage-Husa moving window estimation on the pseudo-GNSS position information, calculate the scale factor through covariance matching to achieve adaptive estimation of the noise covariance matrix, and obtain the optimal filtering state after time update and measurement update.

7. The method for enhancing navigation during interruption for integrated navigation according to claim 6, wherein Perform Sage-Husa moving window estimation on the pseudo-GNSS position information, calculate the scale factor through covariance matching to achieve adaptive estimation of the noise covariance matrix, and obtain the optimal filtering state after time update and measurement update, including: In the AI / INS integrated navigation system, the pseudo GNSS position information vector is expressed as: Among them, , according to the covariance propagation law, the theoretical value of the covariance matrix of the pseudo-GNSS position innovation vector at the current epoch is: Perform a moving window estimation on the pseudolite GNSS position innovation vector, and take an equal weighted average of the innovation vectors within the window. Assuming that the information vector is normally distributed, for an information vector with a window size of d, the estimated covariance matrix of the pseudolite GNSS position information vector at the current epoch is as follows: The theoretical value and the estimated value of the covariance matrix of the pseudo-GNSS position innovation vector are approximately equal, i.e., , and the estimated measurement noise covariance matrix is expressed as: Inflate the preset measurement noise covariance matrix through the robust factor. If the estimated covariance matrix matches the theoretical value, the traces of the two are equal; otherwise, there is a deviation between the traces of the two. Among them, the robust factor is: Among them, is the trace of the matrix, is a preset empirical threshold, is the prior measurement noise covariance matrix, which is obtained by predicting experimental statistics through an AI model; The actually estimated measurement noise covariance matrix at the current epoch is expressed as: Send the inflated estimated measurement noise covariance matrix into the measurement update process to obtain the optimal state estimation at the current epoch, expressed as: Estimate the process noise covariance matrix according to the covariance propagation law, specifically expressed as: Construct the scale factor through the operation of the trace and inflate the preset process noise covariance matrix. Among them, the scale factor is: Among them, is a preset empirical threshold value, is a prior process noise covariance matrix, which is estimated by the Allan variance method; The actually estimated process noise covariance matrix at the current epoch can be expressed as: Store the inflated process noise covariance matrix and use it for the time update process of the next epoch.

8. A navigation enhancement system for integrated navigation during interruptions, characterized in that, Including: A construction module for constructing a TA-LSTM prediction model for predicting pseudo-GNSS position information; A first enhancement module for performing GNSS / INS integrated navigation based on the Kalman filter and training the TA-LSTM prediction model when the GNSS signal is present and reliable; A second enhancement module for predicting the pseudo-GNSS position increment using the trained TA-LSTM prediction model when the GNSS signal is interrupted, obtaining the pseudo-GNSS position information at the current epoch through increment accumulation, and obtaining the optimal filtering state based on the error state Kalman filter and outputting the navigation position information.

9. An electronic device, comprising a memory, a processor, and a computer program stored on the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the navigation enhancement method for integrated navigation during interruption as described in any one of claims 1 to 7.

10. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the navigation enhancement method for integrated navigation during interruption as described in any one of claims 1 to 7.

Images