Self-adaptive robust vehicle-mounted navigation method and equipment based on LSTM (Long Short Term Memory) assistance

Abstract

The invention provides a self-adaptive robust vehicle navigation method and device based on LSTM assistance, and the method comprises the steps: calculating a process noise scale factor through process noise covariance estimation, and dynamically adjusting a process noise covariance matrix Q in a Kalman filter; executing a prediction process of Kalman filtering to obtain a one-step prediction state value and a state prediction covariance matrix; calculating a robust factor and a mahalanobis distance adaptive factor based on the prediction residual vector, and performing expansion adjustment on the prior measurement noise covariance matrix R and the predicted state prediction covariance matrix P; executing an updating process of Kalman filtering to obtain a state optimal estimation and a covariance matrix at the current moment; and when the GNSS loses lock, pseudo GNSS extended Kalman filtering is carried out by using a pseudo GNSS measurement value generated by the LSTM neural network and combining a process noise covariance estimation method. The method can better adapt to the dynamic characteristics of the navigation system in different environments, and provides powerful support for the stability and reliability of the combined navigation system.

Description

Technical Field

[0001] The present invention belongs to the technical field of vehicle-mounted navigation and relates to an adaptive robust vehicle-mounted navigation technical solution based on LSTM assistance. Background Art

[0002] The GNSS / INS integrated navigation system, by integrating the high-precision positioning of the Global Navigation Satellite System (GNSS) with the autonomous dead reckoning capabilities of the Inertial Navigation System (INS), achieves complementary advantages and has become a classic solution in the field of in-vehicle navigation. However, existing technologies still have significant shortcomings in practical applications, especially in complex urban environments and dynamic vehicle motion scenarios: 1) Noise interference and error accumulation issues of low-cost MEMS IMU In tunnels, boulevards, urban canyons, and other scenarios where satellite signals are obscured, GNSS signals are interrupted or severely attenuated, requiring the INS to operate independently. However, low-cost microelectromechanical systems (MEMS) inertial measurement units (IMUs) suffer from high sensor noise and unstable zero bias, causing INS navigation errors to accumulate and diverge rapidly over time. Although existing technologies attempt to mitigate errors through odometers (ODOs), odometer data is susceptible to interference from wheel slip and road surface unevenness. Under dynamic conditions such as sudden acceleration and sharp turns, the assistance is significantly reduced, making it impossible to effectively control error growth.

[0003] 2) Mismatch between dynamic vehicle motion model and noise statistics Traditional Kalman filters (KFs) and extended Kalman filters (EKFs) rely on preset process noise covariance matrices (Q) and measurement noise covariance matrices (R). However, under maneuvering conditions such as frequent acceleration and deceleration, lane changes, and sharp turns, the actual motion model and the preset model become significantly mismatched, resulting in the inability of the filter parameters to dynamically adapt to changes in the vehicle's motion state. Furthermore, multipath effects and signal obstruction in complex urban environments lead to gross errors in GNSS observations. Existing methods lack the ability to dynamically adjust to observation noise, further exacerbating positioning drift.

[0004] 3) Loss of positioning continuity in GNSS lock-out scenarios Existing integrated navigation systems primarily rely on independent INS navigation when GNSS lock is lost for extended periods (e.g., in underground tunnels and parking lots). However, the accumulated error of the MEMS IMU rapidly increases, leading to a sharp decline in positioning accuracy. Although some studies have used linear extrapolation based on historical data or simple neural networks (such as GRU) to predict pseudo-GNSS measurements, these methods lack generalizability for complex vehicle motion patterns. In particular, prediction errors increase significantly during non-stationary motion (e.g., frequent starts and stops, and curves), making it difficult to meet high-precision positioning requirements.

[0005] 4) Limitations of existing adaptive filtering algorithms While existing adaptive Kalman filtering algorithms (such as the Sage-Husa filter) can adjust noise covariance to a certain extent, they are computationally complex and lack the ability to track dynamic noise in real time. In particular, when the vehicle's motion state changes suddenly, traditional methods cannot quickly respond to the instantaneous changes in process noise, resulting in delayed state estimation. Furthermore, existing algorithms lack robustness against gross observation noise errors, making them susceptible to introducing anomalous observations when GNSS signals are interfered with, further reducing system stability.

[0006] In summary, existing GNSS / INS integrated navigation technology faces core problems such as noise sensitivity, poor dynamic adaptability, and error loss when GNSS loses lock in complex urban environments. There is an urgent need for a navigation method that can dynamically adjust filtering parameters, integrate multi-source anti-error strategies, and provide high-precision pseudo-measurements when GNSS loses lock, in order to improve the robustness and continuity of the system. Summary of the Invention

[0007] In order to achieve continuous high-precision positioning of a vehicle-mounted navigation system in a complex urban environment, the present invention provides an adaptive robust vehicle-mounted navigation method based on LSTM assistance.

[0008] The above technical problems of the present invention are mainly solved by the following technical solutions: An adaptive robust vehicle navigation method based on LSTM assistance includes the following steps: The process noise covariance matrix Q in the Kalman filter is dynamically adjusted by estimating the process noise covariance and calculating the process noise scaling factor. Execute the prediction process of Kalman filter to obtain the one-step predicted state value and state prediction covariance matrix; Based on the prediction residual vector, the robustness factor and Mahalanobis distance adaptive factor are calculated, and the prior measurement noise covariance matrix R and the predicted state prediction covariance matrix P are expanded and adjusted respectively; Execute the Kalman filter update process to obtain the optimal state estimate and covariance matrix at the current moment; When GNSS loses lock, the pseudo-GNSS measurement values generated by the LSTM neural network are combined with the process noise covariance estimation method to perform pseudo-GNSS extended Kalman filtering.

[0009] Moreover, according to the covariance propagation law of the position prediction residual vector, the process noise scaling factor of the current epoch is estimated, and the process noise covariance matrix Q is dynamically adjusted.

[0010] Moreover, the calculation method of the said robustness factor is to construct a robustness statistic as the ratio of the measurement noise covariance matrix trace of the position prediction residual vector estimate to the prior measurement noise covariance matrix trace, and generate the robustness factor through a three-segment logarithmic function to expand and adjust the measurement noise covariance matrix R.

[0011] Moreover, the calculation method of the adaptive factor is to construct an adaptive statistic based on the Mahalanobis distance of the prediction residual vector, make a judgment based on the critical value of the chi-square distribution, and if it is judged that there is an abnormality, solve the adaptive factor through the Newton iteration method and expand the state prediction covariance matrix P.

[0012] Moreover, when GNSS loses lock, the IMU and odometer feature sequences of several previous GNSS moments are input into the LSTM model to predict the current position observation increment and generate pseudo-GNSS measurement values. Combined with the process noise covariance estimation method, continuous positioning is achieved through the extended Kalman filter.

[0013] Moreover, the input feature sequence of the LSTM model consists of multiple groups of sampled data from the IMU and multiple groups of sampled data from the odometer, and the output is the GNSS position observation increment.

[0014] Moreover, when the GNSS is not locked, the Kalman filter is optimized by adjusting the Q, R, and P matrices. When the GNSS is locked, pseudo-measurement values are generated through LSTM and the measurement noise covariance matrix R is fixed.

[0015] On the other hand, the present invention also provides an electronic device, comprising a memory, a processor, and a computer program stored on the memory and runnable on the processor, wherein when the processor executes the program, the adaptive anti-error vehicle navigation method based on LSTM assistance as described above is implemented.

[0016] On the other hand, the present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the LSTM-assisted adaptive robust vehicle navigation method as described above.

[0017] On the other hand, the present invention also provides a computer program product, comprising a computer program, which, when executed by a processor, implements the LSTM-assisted adaptive robust vehicle navigation method as described above.

[0018] The technical solution of the present invention can better adapt to the dynamic characteristics of the navigation system under different environments by comprehensively adjusting the three covariance matrices P, Q, and R of the Kalman filter, and provide strong support for the stability and reliability of the integrated navigation system.

[0019] The solution of the present invention is simple and convenient to implement and has strong practicality. It solves the problems of low practicality and inconvenience in actual application existing in related technologies, can improve user experience, and has important market value. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] Figure 1 This is a framework diagram of the LSTM-assisted vehicle navigation solution in an embodiment of the present invention.

[0021] Figure 2 4 is a flow chart of an adaptive robust Kalman filter according to an embodiment of the present invention.

[0022] Figure 3 This is a diagram of the LSTM network structure of an embodiment of the present invention.

[0023] Figure 4 Schematic diagram of the experimental route of an embodiment of the present invention. DETAILED DESCRIPTION

[0024] The following will further illustrate the concept, specific structure and technical effects of the present invention in conjunction with the accompanying drawings and embodiments, so as to fully understand the purpose, characteristics and effects of the present invention.

[0025] This paper proposes an adaptive, robust vehicle navigation method based on LSTM assistance. The method includes a process noise covariance estimation algorithm adapted to vehicle mobility, a GNSS adaptive robustness algorithm adapted to complex urban environments, and an LSTM-based algorithm for providing GNSS pseudo-measurement data to assist the INS. The first two algorithms simultaneously form an adaptive robustness filter that can cope with changes in vehicle mobility. This filter calculates process noise statistics and robustness statistics from the position prediction residual vector to obtain process noise scaling factors and robustness factors. It then inflates the process noise covariance matrix and the observation noise covariance matrix, introduces the Mahalanobis distance based on the prediction residual vector to detect system anomalies, and adjusts the state prediction covariance matrix using a designed adaptive factor. Furthermore, the pseudo-GNSS measurement data generated by the LSTM model is combined with the process noise covariance estimation method to improve the positioning accuracy of the integrated navigation system when the vehicle loses the GNSS signal in urban tunnel scenarios. By comprehensively adjusting the three covariance matrices P, Q, and R of the Kalman filter, it can better adapt to the dynamic characteristics of the navigation system in different environments, improve filtering performance, and provide strong support for the stability and reliability of the integrated navigation system.

[0026] See also Figure 1 The solution provided by the embodiment of the present invention mainly consists of three parts: prediction of pseudo GNSS measurement values based on LSTM network model, process noise covariance estimation method and adaptive robust Kalman filter algorithm. The core idea of the algorithm is to adaptively adjust the Kalman filter in the 、 and ,On the other hand, when GNSS loses lock, the LSTM model provides pseudo GNSS ,measurements for the integrated navigation system to achieve high-precision ,positioning of the vehicle.

[0027] When the GNSS is not locked, the difference between the position vector calculated by the INS mechanical arrangement and the GNSS observed position vector is fed into the process noise covariance estimation algorithm to estimate the matrix at the current epoch After that, they are used in the GNSS adaptive robust Kalman filter and ODO / NHC Kalman filter respectively, and the measurement noise covariance matrix at the current epoch is converted to and the state covariance matrix Save, the matrix for estimating the next epoch When the vehicle is in a tunnel, that is, when the GNSS is locked, the algorithm flow is similar to the above process description and will not be elaborated here. However, there are two differences: first, the GNSS pseudo-measurement values predicted by the LSTM network model are used for auxiliary positioning; second, the pseudo-GNSS measurement values use the extended Kalman filter and do not perform the measurement noise covariance matrix. and the state prediction covariance matrix This is because there is only a single tunnel scenario and the pseudo-measurement values predicted by LSTM are used. The matrix Generally speaking, it is fixed.

[0028] See also Figure 2 The present invention integrates the process noise covariance estimation algorithm into the adaptive robust algorithm to form an adaptive robust Kalman filter. The process noise covariance matrix, measurement noise covariance matrix and state prediction covariance matrix are adjusted successively. It not only solves the influence of process noise and measurement noise on the Kalman filter system during the vehicle maneuverability movement, but also proposes a solution to the system state abnormality caused by the mismatch of INS dynamic model caused by vehicle maneuverability changes. It is an on-board combined navigation Kalman filter suitable for complex urban environments.

[0029] When a vehicle maneuvers through complex urban environments, such as boulevards, city blocks, and overpasses, the GNSS signal is obstructed by obstacles, preventing it from providing accurate position observations. Furthermore, the vehicle's maneuvering can cause system state anomalies. This filter can be used to achieve high-precision on-board navigation. First, the prediction residual vector is calculated from the observation vector representing the difference between the INS's mechanical position vector and the GNSS position vector. The process noise covariance estimation algorithm (OQEA) proposed in this invention is then used to calculate the process noise statistics and process noise scaling factors. This process noise covariance matrix at the current epoch is then estimated and fed into the one-step prediction process of the GNSS Kalman filter to calculate the state vector prediction value and the state prediction covariance matrix. Next, robustness statistics and robustness factors are calculated from the prediction residual vector and the state prediction covariance matrix. The measurement noise covariance matrix at the current epoch is estimated, and then the adaptive statistics are calculated to determine whether the system is abnormal. If the system is abnormal, the state prediction covariance matrix is inflated by the adaptive factors. Finally, the GNSS Kalman filter measurement update process is performed and the measurement noise covariance matrix and state covariance matrix of the current epoch are saved to prepare for the next epoch.

[0030] Specifically, an embodiment of the present invention provides an LSTM-assisted adaptive robust vehicle navigation method, which is implemented in the following steps: Step 1: Calculate the process noise scaling factor based on the process noise covariance estimation method to adjust the Kalman filter Formation.

[0031] Preferably, the specific contents of step one include the following: The position observation vector of the Kalman filter system for:

[0032] in, The position vector calculated for the INS mechanical arrangement, is the actual GNSS position observation vector or pseudo-GNSS measurement vector.

[0033] Position prediction residual vector Defined as:

[0034] in, is the design matrix, is the state prediction vector.

[0035] From the above two equations, we can see that when the vehicle's motion state changes significantly, The error will increase, becomes larger, resulting in the position prediction residual vector Become bigger, that is It is consistent with the change of vehicle motion state, so it is used It is reasonable to express the degree of change in the vehicle's motion state. According to the covariance propagation law, for the position prediction residual vector with a window of m, the prediction covariance matrix of the current epoch is:

[0036] Among them, the superscript T is the transposition symbol, j is the counting symbol, and They are the measurement noise covariance matrix and state covariance matrix of the position prediction residual vector estimation, both of which are unknown and cannot be solved accurately.

[0037] definition , is the observation noise covariance matrix of the previous epoch estimated by the proposed adaptive robust Kalman filter algorithm, then:

[0038] in, is the measurement noise covariance matrix estimated for the previous epoch.

[0039] In order to estimate the process noise covariance matrix at the current epoch , define the process noise scaling factor as follows for:

[0040] in,

[0041] in, is the process noise statistic, Indicates the absolute value symbol, represents the matrix trace operation, is the state one-step transfer matrix, is the state covariance matrix estimated by the proposed adaptive robust Kalman filter algorithm at the previous epoch, is the noise covariance matrix of the process at the previous epoch. if , indicating that the vehicle motion state changes significantly at the current epoch. Since the vehicle motion state changes are often instantaneous, we design the process noise scaling factor as , to highlight the instantaneous nature; otherwise, we scale the noise of the process at the current epoch by Designed as the scale factor of the previous epoch , rather than the general design of 1, because the change of vehicle motion state is also continuous.

[0042] Then the process noise covariance matrix at the current epoch is It can be expressed as:

[0043] in, The default value for initialization.

[0044] The estimated process noise covariance matrix is sent to the subsequent Kalman filter system for one-step prediction to compensate for the defect of mismatch between the established motion model and the actual motion state caused by the change of the vehicle's motion state.

[0045] Step 2: Perform the Kalman filter prediction process.

[0046] Preferably, the specific contents of step 2 include the following: The Kalman filter time update equation is:

[0047] Where, is the one-step predicted state value, is the state vector estimated at the previous epoch, is the covariance matrix of the one-step predicted state vector, is the state covariance matrix of the optimal estimate of the state at the previous moment, is the system state noise variance matrix, is the system noise, represents the first-order moment.

[0048] Step 3: Calculate the robustness factor based on the prediction residual vector and the adaptive factor based on the Mahalanobis distance, respectively for the prior measurement noise covariance matrix And the predicted state prediction covariance matrix To expand.

[0049] Preferably, the specific contents of step three include the following: 1) Robustness factor From the above two equations, we can see that when the GNSS observation quality deteriorates, The error will increase, becomes larger, resulting in the position prediction residual vector Become bigger, that is This is consistent with the change in GNSS observation quality, so It is reasonable to express the degree of change in GNSS observation quality. According to the covariance propagation law, the measurement noise covariance matrix estimated for the position prediction residual vector with a window of m is:

[0050] in, The state forecast covariance matrix after making a one-step forecast of the process noise covariance matrix estimated using the process noise covariance estimation algorithm.

[0051] Constructed robust statistics is the measurement noise covariance matrix of the position prediction residual vector estimate The trace of the prior measurement noise covariance matrix The ratio of the traces of :

[0052] in, is the prior measurement noise covariance matrix, which can be obtained from GNSS receivers or experimental statistics.

[0053] In order to estimate the measurement noise covariance matrix at the current epoch , define the following three-segment logarithmic robustness factor for:

[0054] in, is a natural constant, is the natural logarithm function.

[0055] The robustness factor of the structure For the prior measurement noise covariance matrix For expansion, we have:

[0056] The expanded measurement noise covariance matrix is used in the subsequent adaptive factor calculation and measurement update process to improve the matching degree of the state prediction covariance matrix and optimize the Kalman filter.

[0057] 2) Adaptive factor In addition to system state anomalies caused by inappropriate selection of process noise covariance and measurement noise covariance, an inappropriate INS dynamics model caused by changes in vehicle maneuverability can also cause system state anomalies. To address these issues, a Mahalanobis distance based on the prediction residual vector is introduced to detect system anomalies, and an adaptive factor is proposed to expand the state prediction covariance matrix.

[0058] If the GNSS observations are uncontaminated, the position observation vector The mean is , the covariance is The standard normal distribution of arrive The square of the Mahalanobis distance between , and use it to detect system anomalies and construct adaptive factors, Follows a chi-square distribution with 3 degrees of freedom , then:

[0059] Among them, the superscript represents the inverse of the matrix, is the Mahalanobis distance, Represents the covariance sign.

[0060] Will Bringing in the above formula, it can be written as the position-based prediction residual vector Adaptive statistics of for:

[0061] in The state prediction covariance matrix after one-step prediction of the process noise covariance matrix estimated by the process noise covariance estimation algorithm is is the measurement noise covariance matrix after expansion by the robustness factor.

[0062] Choose a low probability value , if the adaptive statistic Greater than the chi-square distribution critical value The probability is greater than , it indicates that the system is inconsistent with some of the above assumptions, that is, the system is abnormal, and the adaptive expansion state prediction covariance matrix is needed Otherwise, the system is normal. The following formula represents the critical situation:

[0063] in, represents the probability function, Taking the value as 1%, we can find out after looking up the critical value table of chi-square distribution The value is approximately 11.345.

[0064] In critical cases, the following equations are satisfied:

[0065] To calculate the adaptive factor , define the function for:

[0066] Newton's iteration method is an approximate method for solving nonlinear equations. The solution of the above nonlinear equation using Newton's iteration method is:

[0067] in,

[0068]

[0069] The subscript i of the adaptive factor represents the number of iterations, and the adaptive factor of the 0th iteration is Defined as 1 until the adaptive statistic When it becomes stable, the iteration ends and As an adaptive factor to expand the state prediction covariance matrix, the adaptive factor Expressed as:

[0070] The constructed adaptive factor is used to expand the state prediction covariance matrix, and the expanded state prediction covariance matrix is It can be expressed as: Then:

[0071] The expanded state prediction covariance matrix is used in the measurement update process to adaptively adjust the Kalman filter.

[0072] Step 4: Perform the Kalman filter update process.

[0073] Preferably, the specific contents of step 4 include the following: The Kalman filter measurement update equation is:

[0074] in, is the Kalman gain, is the measurement noise variance matrix, is the optimal estimate of the current state, is the identity matrix, To measure noise, is the state optimal estimate covariance matrix.

[0075] Step 5: If GNSS is lost, the LSTM-based vehicle navigation system is combined with the process noise covariance estimation method and a pseudo-GNSS extended Kalman filter is performed.

[0076] Preferably, the specific contents of step five include the following: See also Figure 3When a vehicle maneuvers through an urban tunnel, the feature sequences of the previous n-1 GNSS moments and the feature sequence mechanically obtained by the INS at the current moment are fed into the LSTM neural network to predict the position observation increment at the current moment. Combined with the GNSS pseudo-observation value at the previous moment, the current GNSS pseudo-observation value is fed into the process noise covariance estimation algorithm, and ODO / NHC Kalman filtering and pseudo-GNSS extended Kalman filtering are performed.

[0077] Generally speaking, there are differences between the IMU sampling frequency, the odometer sampling frequency, and the GNSS signal acquisition frequency. If the IMU sampling frequency is N times the GNSS signal frequency, and the odometer sampling frequency is M times the GNSS signal frequency, then the GNSS position observation increment at each adjacent moment is Corresponding to N groups IMU signature sequence and Group M Odometer feature sequence , together constitute an input feature sequence of GNSS time , are IMU specific force, IMU angular rate, odometer speed, pitch angle and heading angle respectively. Therefore, the input feature sequence dimension of the LSTM neural network model of the present invention is , the output feature sequence dimension is .

[0078] exist Figure 3 In the figure, the green dotted box is the input feature sequence of the LSTM network model at a GNSS moment; the yellow module represents the input layer of the LSTM model, which takes the first n epochs The input feature sequence is sent to the input layer; the blue module represents the hidden layer of the LSTM model; the red module represents the output layer of the LSTM model, which takes the state of the last hidden layer at the current moment as the model output and selects a The function is used as the fully connected layer to process the output of the activation function to obtain the predicted value of the LSTM model. .

[0079] When a vehicle maneuvers through an urban tunnel, the feature sequences of the previous n-1 GNSS moments and the feature sequence mechanically obtained by INS at the current moment are fed into the LSTM neural network to predict the position observation increment at the current moment. Combined with the GNSS pseudo-observation value at the previous moment, the current GNSS pseudo-observation value is fed into the process noise covariance estimation algorithm, and ODO / NHC Kalman filtering and pseudo-GNSS extended Kalman filtering are performed.

[0080] In order to verify the performance superiority of the technical solution of the present invention, experimental analysis was carried out when the GNSS was not locked and when it was locked. The urban and suburban roads around a university were selected as the experimental route, which is about 10km in total. The specific experimental route map is as follows: Figure 1 shown.

[0081] 1. GNSS is not locked When the GNSS is not locked, the navigation performance of the scheme of the present invention is compared with that of the EKF algorithm. The comparative values of the RMS and 90% errors of the horizontal position, horizontal velocity and heading are shown in Table 1.

[0082] Table 1 Comparison of RMS and 90% error of position, velocity and heading between EKF and the present invention when GNSS is not locked

[0083] The detailed analysis is as follows: when GNSS lock is maintained, the RMS and 90% errors of the proposed solution are reduced by 89.7% and 58.9% respectively compared to the EKF in terms of horizontal position error. The RMS and 90% errors of the proposed solution are reduced by 81.0% and 72.9% respectively compared to the EKF in terms of horizontal velocity error. The RMS and 90% errors of the proposed solution are reduced by 68.3% and 68.2% respectively compared to the EKF in terms of heading error.

[0084] 2. GNSS lock loss When GNSS is locked, the navigation performance of the proposed solution is compared with that of a single INS, GRU, and LSTM. The RMS and 90% error comparison values of horizontal position, horizontal velocity, and heading are shown in Table 2.

[0085] Table 2 Comparison of the RMS errors of the north, east, and ground directions of single INS, GRU, LSTM, and the present invention when GNSS is not locked

[0086] The detailed analysis is as follows: when GNSS lock is lost, the RMS error of the proposed solution in northing position accuracy is reduced by 79.3%, 76.1%, and 45.0% compared to single INS, GRU, and LSTM, respectively. In easting position accuracy, the RMS error of the proposed solution is reduced by 62.9%, 45.5%, and 49.6% compared to single INS, GRU, and LSTM, respectively. In ground position accuracy, the RMS error of the proposed solution is reduced by 82.2%, 36.4%, and 34.3% compared to single INS, GRU, and LSTM, respectively.

[0087] The above results show that the technical solution of the present invention can effectively reduce the gross errors of GNSS observations and the gross errors of INS state of vehicle maneuverability in complex urban environments, and has good system robustness and high navigation accuracy.

[0088] In specific implementation, the method proposed in the technical solution of the present invention can be automatically run by those skilled in the art using computer software technology. System devices that implement the method, such as computer-readable storage media that store the corresponding computer program of the technical solution of the present invention and computer equipment that runs the corresponding computer program, should also be within the scope of protection of the present invention.

[0089] The following describes the LSTM-assisted adaptive robust vehicle navigation electronic device provided by the present invention. The LSTM-assisted adaptive robust vehicle navigation electronic device described below and the LSTM-assisted adaptive robust vehicle navigation method described above can refer to each other.

[0090] The electronic device may include: a processor, a communications interface, a memory, and a communications bus, wherein the processor, the communications interface, and the memory communicate with each other via the communications bus. The processor may call logic instructions in the memory to execute the LSTM-assisted adaptive robust vehicle navigation method, which mainly includes the software processing portion of the above steps.

[0091] Furthermore, the logical instructions in the aforementioned memory can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the portion that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions for enabling a computer device (which can be a personal computer, server, or 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 media include various media capable of storing program code, such as USB flash drives, mobile hard drives, read-only memories (ROMs), random access memories (RAMs), magnetic disks, or optical disks.

[0092] On the other hand, the present invention also provides a computer program product, which includes a computer program, which can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer can execute the software processing part of the LSTM-assisted adaptive anti-error vehicle navigation method provided by the above methods.

[0093] On the other hand, the present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, is implemented to execute the software processing part of the LSTM-assisted adaptive anti-error vehicle navigation method provided by the above-mentioned methods.

[0094] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units, i.e., they may be located in one location or distributed across multiple network units. Some or all of the modules may be selected based on actual needs to achieve the objectives of the present embodiment. Persons of ordinary skill in the art will be able to understand and implement the present invention without inventive effort.

[0095] Through the above description of the embodiments, those skilled in the art will clearly understand that each embodiment can be implemented using software plus a necessary general-purpose hardware platform, or of course, hardware. Based on this understanding, the essence of the above technical solution, or the portion that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a magnetic disk, or an optical disk, and includes a number of instructions for causing a computer device (such as a personal computer, server, or network device) to execute the methods described in each embodiment or certain portions of the embodiments.

[0096] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the various embodiments of the present invention.

Claims

1. An adaptive robust vehicle navigation method based on LSTM assistance, characterized by: The following processes are included: The process noise covariance matrix Q in the Kalman filter is dynamically adjusted by estimating the process noise covariance and calculating the process noise scaling factor. Execute the prediction process of Kalman filter to obtain the one-step predicted state value and state prediction covariance matrix; Based on the prediction residual vector, the robustness factor and Mahalanobis distance adaptive factor are calculated, and the prior measurement noise covariance matrix R and the predicted state prediction covariance matrix P are expanded and adjusted respectively; Execute the Kalman filter update process to obtain the optimal state estimate and covariance matrix at the current moment; When GNSS loses lock, the pseudo-GNSS measurement values generated by the LSTM neural network are combined with the process noise covariance estimation method to perform pseudo-GNSS extended Kalman filtering.

2. The LSTM-assisted adaptive robust vehicle navigation method according to claim 1, characterized in that: According to the covariance propagation law of the position prediction residual vector, the process noise scaling factor of the current epoch is estimated, and the process noise covariance matrix Q is dynamically adjusted.

3. The LSTM-assisted adaptive robust vehicle navigation method according to claim 1, characterized in that: The calculation method of the said robustness factor is to construct a robustness statistic as the ratio of the measurement noise covariance matrix trace of the position prediction residual vector estimate to the prior measurement noise covariance matrix trace, and generate the robustness factor through a three-segment logarithmic function to expand and adjust the measurement noise covariance matrix R.

4. The LSTM-assisted adaptive robust vehicle navigation method according to claim 1, characterized in that: The adaptive factor is calculated by constructing an adaptive statistic based on the Mahalanobis distance of the prediction residual vector, making a judgment based on the critical value of the chi-square distribution, and solving the adaptive factor through the Newton iteration method if an abnormality is judged, and expanding the state prediction covariance matrix P.

5. The LSTM-assisted adaptive robust vehicle navigation method according to claim 1, characterized in that: When GNSS loses lock, the IMU and odometer feature sequences of several previous GNSS moments are input into the LSTM model to predict the current position observation increment and generate pseudo-GNSS measurement values. Combined with the process noise covariance estimation method, continuous positioning is achieved through the extended Kalman filter.

6. The LSTM-assisted adaptive robust vehicle navigation method according to claim 5, characterized in that: The input feature sequence of the LSTM model consists of multiple sets of sampled data from the IMU and multiple sets of sampled data from the odometer, and the output is the GNSS position observation increment.

7. The LSTM-assisted adaptive robust vehicle navigation method according to claim 5, characterized in that: When the GNSS is not locked, the Kalman filter is optimized by adjusting the Q, R, and P matrices. When the GNSS is locked, pseudo-measurement values are generated through LSTM and the measurement noise covariance matrix R is fixed.

8. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the program, the LSTM-assisted adaptive robust vehicle navigation method according to any one of claims 1 to 7 is implemented.

9. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the LSTM-assisted adaptive robust vehicle navigation method according to any one of claims 1 to 7 is implemented.

10. A computer program product comprising a computer program, characterized in that: When the computer program is executed by a processor, the LSTM-assisted adaptive robust vehicle navigation method according to any one of claims 1 to 7 is implemented.

Images