Vehicle motion constraint dual-adjustment integrated navigation method based on machine learning

Abstract

The invention provides a vehicle motion constraint dual-adjustment integrated navigation method based on machine learning, which comprises the step of realizing accurate prediction of lateral and vertical speeds of a vehicle in different motion states by fusing an inertial navigation system, a speedometer and non-integrity constraint and combining a random forest regression model. The predicted value is used as an observed value in NHC Kalman filtering, and meanwhile, a covariance matrix of a Kalman filter is dynamically adjusted by utilizing speed variance obtained by calculating different motion state data when GNSS signals are available. According to the method, the precision of vehicle navigation can be improved under the condition of not increasing additional sensors, and particularly, the high-precision positioning capability can be kept in complex environments such as cities and the like. Through experimental verification, the algorithm significantly improves the horizontal positioning precision when the GNSS fails, and provides an efficient and low-cost solution for a vehicle navigation system.

Description

Technical Field

[0001] The present invention relates to the technical field of navigation, and in particular, to a combined navigation method for double adjustment of vehicle motion constraints based on machine learning. Background Art

[0002] The Global Navigation Satellite System (GNSS) is widely used in vehicle navigation because it can provide high-precision positioning information. However, GNSS signals are easily affected in environments with high-rise buildings in the city, tunnels or other occluded environments, resulting in the inability to provide continuous and reliable position information. Therefore, the data fusion technology that combines multiple sensors to achieve continuous positioning has become the key to the development of navigation systems.

[0003] Current combined navigation systems usually integrate GNSS, Micro-Electro-Mechanical System (MEMS) Inertial Measurement Unit (IMU) and vehicle sensor data, and combine the Nonholonomic Constraint (NHC) method to suppress the error accumulation of inertial navigation. NHC assumes that the lateral and vertical speeds of the vehicle are zero during normal driving, and uses this as a constraint condition. However, in practical applications, due to influencing factors such as the vehicle's motion state, vibration, and sensor installation position, this assumption is difficult to fully meet, resulting in errors in NHC observations and affecting the accuracy of inertial navigation.

[0004] In recent years, machine learning technology has been applied to vehicle navigation systems to identify the complex relationship between inertial measurement data and the actual motion state. By training a machine learning model, when the GNSS signal is available, the relatively high-precision speed data obtained by GNSS / SINS combined solution can be used as labels to learn the mapping relationship between these speeds and inertial measurement data. Summary of the Invention

[0005] The present invention provides a combined navigation method for double adjustment of vehicle motion constraints based on machine learning, which is used to solve the defects of the existing technology in terms of accuracy and stability of vehicle positioning when the GNSS signal fails, and to improve the accuracy and stability of the navigation system when the GNSS signal fails.

[0006] In a first aspect, the present invention provides a combined navigation method for double adjustment of vehicle motion constraints based on machine learning, including: Obtain SINS Kalman prediction data, determine the available state of GNSS, combine INS, ODO, and NHC, classify the vehicle motion state, determine the data selection features and label values, and obtain the vehicle motion data set; Construct a multi-level random forest regression model, train the multi-level random forest regression model based on the vehicle motion data set, and obtain a vehicle position, speed, and attitude prediction model; Determine the unavailable state of GNSS, output the vehicle motion prediction value by the vehicle position, speed, and attitude prediction model, and adjust the covariance matrix of the Kalman filter according to the vehicle motion state classification result; Update the Kalman filter using the vehicle motion prediction value and covariance matrix, obtain the vehicle position, speed, and attitude error compensation information, and output the optimized vehicle position, speed, and attitude output value according to the vehicle position, speed, and attitude error compensation information.

[0007] According to a vehicle motion constraint dual-adjustment integrated navigation method based on machine learning provided by the present invention, obtaining SINS Kalman prediction data includes: Combining IMU data, error compensation, mechanical arrangement, and Kalman prediction, collect the accelerometer, gyroscope data, and INS forward speed output by the IMU under the available state of GNSS to obtain SINS Kalman prediction data.

[0008] According to a vehicle motion constraint dual-adjustment integrated navigation method based on machine learning provided by the present invention, determining the available state of GNSS, combining INS, ODO, and NHC, classifying the vehicle motion state, determining the data selection features and label values, and obtaining the vehicle motion data set, including: Determine the lateral speed and vertical speed corresponding to the time stamp of the available state of GNSS as label values, and construct an initial training data set in combination with SINS Kalman prediction data; Calculate the average value and standard deviation of the heading angular velocity of the gyroscope Z-axis data of all SINS Kalman prediction data, and determine the first-level vehicle motion state, the second-level vehicle motion state, and the third-level vehicle motion state; According to the classification result, divide the initial training data set into training data sets of different levels; Calculate the statistical average value and variance of the lateral speed and vertical speed in the training data sets of different levels respectively, and randomly divide the data set into a training set, a test set, and a validation set to form the vehicle motion data set.

[0009] According to a vehicle motion constraint dual-adjustment integrated navigation method based on machine learning provided by the present invention, constructing a multi-level random forest regression model, training the multi-level random forest regression model based on the vehicle motion data set, and obtaining a vehicle position, speed, and attitude prediction model, including: Input the vehicle motion data set; Perform random feature selection to obtain a data subset, input the data subset into multiple decision trees, and use grid search to determine the hyperparameter set for each tree. The hyperparameter set includes depth parameters, sample splitting parameters, and leaf node parameters, obtain multiple predicted values, average the multiple predicted values to obtain an average prediction result, and obtain lateral velocity predicted values and vertical velocity predicted values; Evaluate the model using mean squared error, root mean squared error, and mean absolute error, obtain the optimal hyperparameter set, and output the vehicle position velocity attitude prediction model.

[0010] According to a vehicle motion constraint dual-adjustment integrated navigation method based on machine learning provided by the present invention, determine the GNSS unavailable state, output the vehicle motion predicted value by the vehicle position velocity attitude prediction model, and adjust the covariance matrix of the Kalman filter according to the vehicle motion state classification result, including: When GNSS is unavailable, input the vehicle motion data set into the vehicle position velocity attitude prediction model to obtain the predicted lateral velocity and vertical velocity; According to the gyroscope Z-axis data collected by the IMU at the current moment, determine the real-time state and classification of vehicle motion; Subtract the lateral velocity and vertical velocity obtained by mechanical arrangement from the predicted lateral velocity and vertical velocity to obtain the observation value updated by the Kalman filter; Construct the measurement matrix updated by the Kalman filter by taking the total differential of the conversion error equation from the inertial navigation coordinate system to the wheel coordinate system; Set covariance matrices for different levels through the variances of lateral velocity and vertical velocity at different levels obtained from statistical data.

[0011] According to a vehicle motion constraint dual-adjustment integrated navigation method based on machine learning provided by the present invention, update the Kalman filter using the vehicle motion predicted value and covariance matrix to obtain vehicle position velocity attitude error compensation information, including: Combine the minimum variance estimate of the current moment state vector obtained through ODO Kalman update and the variance corresponding to the minimum variance estimate, perform NHC Kalman update, and output the minimum variance estimate of the current moment state vector obtained after NHC Kalman update and the variance corresponding to the minimum variance estimate.

[0012] In a second aspect, the present invention also provides a vehicle motion constraint dual-adjustment integrated navigation system based on machine learning, including: An acquisition module, configured to acquire SINS Kalman prediction data, determine the available state of GNSS, combine INS, ODO, and NHC, classify the vehicle motion state, determine data selection features and label values, and obtain a vehicle motion data set; A training module, configured to build a multi-level random forest regression model, train the multi-level random forest regression model based on a vehicle motion data set, and obtain a vehicle position, speed and attitude prediction model; An adjustment module, configured to determine a GNSS unavailable state, output a vehicle motion prediction value by the vehicle position, speed and attitude prediction model, and adjust a covariance matrix of a Kalman filter according to a vehicle motion state classification result; An optimization module, configured to update the Kalman filter by using the vehicle motion prediction value and the covariance matrix, obtain vehicle position, speed and attitude error compensation information, and output an optimized vehicle position, speed and attitude output value according to the vehicle position, speed and attitude error compensation information.

[0013] 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 method for dual-adjustment integrated navigation of vehicle motion constraints based on machine learning as described in any one of the above is implemented.

[0014] 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 method for dual-adjustment integrated navigation of vehicle motion constraints based on machine learning as described in any one of the above is implemented.

[0015] The method for dual-adjustment integrated navigation of vehicle motion constraints based on machine learning provided by the present invention realizes accurate prediction of lateral and vertical speeds of vehicles in different motion states by fusing an inertial navigation system, an odometer, and nonholonomic constraints, and combining a random forest regression model. The prediction value is used as an observation value in the NHC Kalman filter, and at the same time, the covariance matrix of the Kalman filter is dynamically adjusted by using the speed variance calculated from different motion state data when the GNSS signal is available. The present invention can improve the accuracy of vehicle navigation without adding additional sensors, especially maintaining high-precision positioning ability in complex environments such as cities. Through experimental verification, the algorithm significantly improves the horizontal positioning accuracy when the GNSS fails, providing an efficient and low-cost solution for the vehicle navigation system. Description of the Drawings

[0016] 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 use in the embodiments or the description of the prior art. Obviously, the following drawings 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.

[0017] Figure 1It is one of the schematic flowcharts of the dual-adjustment integrated navigation method for vehicle motion constraints based on machine learning provided by the present invention; Figure 2 It is the system framework diagram provided by the present invention; Figure 3 It is the second schematic flowchart of the dual-adjustment integrated navigation method for vehicle motion constraints based on machine learning provided by the present invention; Figure 4 It is the data processing flowchart provided by the present invention; Figure 5 It is the hierarchical trajectory diagram of the vehicle motion state provided by the present invention; Figure 6 It is the structure diagram of the random forest regression model provided by the present invention; Figure 7 It is the update flowchart of the Kalman filter provided by the present invention; Figure 8 It is the comparison diagram of experimental results provided by the present invention; Figure 9 It is the schematic structure diagram of the dual-adjustment integrated navigation system for vehicle motion constraints based on machine learning provided by the present invention; Figure 10 It is the schematic structure diagram of the electronic device provided by the present invention. Detailed implementation manners

[0018] 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.

[0019] Figure 1 It is one of the schematic flowcharts of the dual-adjustment integrated navigation method for vehicle motion constraints based on machine learning provided by the present invention. As Figure 1 shown, it includes: Step 100: Obtain the SINS Kalman prediction data, determine the available state of the GNSS, combine the INS, ODO, and NHC, classify the vehicle motion state, determine the data selection features and label values, and obtain the vehicle motion data set; Step 200: Construct a multi-level random forest regression model, train the multi-level random forest regression model based on the vehicle motion data set, and obtain a vehicle position, speed, and attitude prediction model; Step 300: Determine the GNSS unavailable state, output the vehicle motion prediction value by the vehicle position, velocity, and attitude prediction model, and adjust the covariance matrix of the Kalman filter according to the vehicle motion state classification result; Step 400: Update the Kalman filter using the vehicle motion prediction value and the covariance matrix, obtain the vehicle position, velocity, and attitude error compensation information, and output the optimized vehicle position, velocity, and attitude output value according to the vehicle position, velocity, and attitude error compensation information.

[0020] Specifically, as Figure 2 shown, in the embodiment of the present invention, by fusing the inertial navigation system, the odometer, and the nonholonomic constraint, combined with the random forest regression model, the lateral and vertical velocities of the vehicle are predicted to optimize the performance of the Kalman filter. When the GNSS signal is available, by calculating the average value and standard deviation of the heading angular velocity of the IMU gyroscope data, the vehicle motion state is divided into different levels, and the predicted value of the random forest model is used to adjust the observed value of the Kalman filter in the observation domain, and the covariance matrix is adjusted using the variances of different level velocity data in the variance domain. The feature selection method is used to optimize the input of the IMU data and the INS forward velocity, so that the model can more accurately capture the complex relationship between the vehicle motion state and the inertial measurement data, and finally improve the accuracy of the velocity prediction.

[0021] It should be noted that Figure 3 the "high-precision position, velocity, and attitude information output" is the final output result in the GNSS available state, and this result is used to train the predictive AI model. Correspondingly, the "optimized position, velocity, and attitude information output" is the final output result optimized after settling the data predicted by the AI model trained with the high-precision output information when the GNSS is available in the GNSS unavailable state. By simulating at the GNSS available time and the unavailable time, it is assumed that the GNSS is unavailable during the simulation, but it is actually in the available state. In this way, the same data can be solved twice, that is, high-precision reference data is obtained when it is available, and the solved data optimized using the method of the present invention is used when it is assumed to be unavailable. The final effect verification is carried out by comparing the two types of data obtained simultaneously.

[0022] Based on the traditional NHC model, the present invention proposes an improved method: not only using the lateral and vertical velocities predicted by the random forest regression model in the observation domain to adjust the observed values, but also using the variances of different levels of velocity data collected when the GNSS signal is available in the variance domain to adjust the covariance matrix of the Kalman filter. This dual adjustment strategy combining the observation domain and the variance domain can effectively improve the accuracy and stability of the vehicle navigation system when the GNSS signal fails. This method can significantly improve the accuracy of vehicle navigation without adding additional sensors, especially maintaining high-precision positioning ability in complex environments, and effectively solving the positioning challenges when the GNSS fails.

[0023] In one embodiment, as Figure 3 shown, the positions and interaction relationships of GNSS, INS, ODO, IMU, and machine learning models in the entire system are described. The embodiment of the present invention combines IMU data, error compensation, mechanical arrangement, and Kalman prediction to collect the accelerometer, gyroscope data output by the IMU and the forward velocity of INS when the GNSS is available, and obtain the SINS Kalman prediction data.

[0024] In one embodiment, the available state of the GNSS is determined, and in combination with INS, ODO, and NHC, the vehicle motion state is classified, data selection features and label values are determined, and a vehicle motion data set is obtained, including: Determine the lateral velocity and vertical velocity corresponding to the time stamp of the available state of the GNSS as the label values, and construct an initial training data set in combination with the SINS Kalman prediction data; Calculate the average value and standard deviation of the heading angular velocity of the gyroscope Z-axis data of all SINS Kalman prediction data, and determine the first-level vehicle motion state, the second-level vehicle motion state, and the third-level vehicle motion state; According to the classified results, divide the initial training data set into training data sets of different levels; Calculate the statistical average values and variances of the lateral velocity and vertical velocity in the training data sets of different levels respectively, and randomly divide the data set into a training set, a test set, and a validation set to form the vehicle motion data set.

[0025] Specifically, as Figure 4 shown, for the data processing, model training, and calling process, when the GNSS is available, a training data set is constructed, the feature values are the accelerometer, gyroscope data output by the IMU during the available time period, and the forward velocity after mechanical arrangement, and the label values are the lateral velocity and vertical velocity corresponding to the time stamp in the output results during the available time period.

[0026] The constructed training dataset is graded. The statistical average value A (Average value) and standard deviation S (Standard deviation) of the gyroscope Z-axis data of all data are calculated. The absolute value P is obtained by subtracting the gyroscope Z-axis data H of each row of data from the statistical average value A, and each row of data is graded according to the following rules: (1) If P <= 1.5S, it is set to level 0, that is, the straight-line driving state; (2) If 1.5S < P <= 3S, it is set to level 1, that is, the small-angle turning state; (3) If P > 3S, it is set to level 2, that is, the large-angle turning state.

[0027] Each row of data here means that if the sampling rate of the IMU is 50Hz, that is, 50 groups of data will be collected in 1 second, and a group of data will be collected every 0.02 seconds. This group of data is combined and packed into a row of data, which is a group of data collected at the current moment. Then, it will be processed using the statistical values obtained from the statistical processing of a large amount of data saved previously, that is, the current moment and historical data.

[0028] As Figure 5 shown, it is the grading trajectory diagram of the vehicle motion state, which intuitively shows the specific trajectory diagrams of the vehicle in different motion states (straight-line driving, small-angle turning, large-angle turning). It can be clearly recognized that different states during the vehicle driving process can be basically well distinguished. Among them, only the states with grades 1 (small-angle turning), 2 (large-angle turning), and -1 (vehicle stationary) are marked, and the unmarked trajectories are defaulted to the state of grade 0 (straight-line driving).

[0029] According to the graded results, the initial training dataset is divided into three training datasets with different grades, namely grade 0, grade 1, and grade 2. Finally, the statistical average values and variances of the lateral speed and vertical speed in the grade 0, grade 1, and grade 2 datasets are calculated respectively.

[0030] Based on the grade 0, grade 1, and grade 2 datasets, feature selection and target variable definition are performed. The dataset is randomly split into a training set, a test set, and a validation set. Grid search is used to find the best hyperparameters of the model, and then the mean squared error (MSE), root mean squared error (RMSE), and mean absolute error (MAE) are calculated to evaluate the model performance to obtain the best model.

[0031] When GNSS is unavailable, predictive data is constructed, including the forward velocity after mechanical scheduling at the current moment, as well as the accelerometer and gyroscope data output by the IMU. According to the gyroscope Z-axis data, real-time status judgment and grading are performed based on statistical data. Then, regression models of level 0, level 1, and level 2 are called according to the level. Finally, the predicted lateral velocity and vertical velocity of the corresponding model are used as the observed values, and the covariance matrix is set correspondingly using the hierarchical variance obtained during data processing, and then NHC Kalman update is performed.

[0032] In one embodiment, a multi-level random forest regression model is constructed and trained based on a vehicle motion data set to obtain a vehicle position velocity attitude prediction model, including: Input the vehicle motion data set; Perform random feature selection to obtain a data subset, input the data subset into multiple decision trees, and use grid search to determine the hyperparameter set of each tree. The hyperparameter set includes depth parameters, sample splitting parameters, and leaf node parameters, obtain multiple predicted values, average the multiple predicted values to obtain an average prediction result, and obtain predicted lateral velocity and vertical velocity values; Evaluate the model using mean squared error, root mean squared error, and mean absolute error to obtain the best hyperparameter set, and output the vehicle position velocity attitude prediction model.

[0033] Specifically, as Figure 6 shown, the structure of the random forest regression model is shown, including feature input, model training, and prediction output.

[0034] In the input data, velocity_x represents the forward velocity of the vehicle calculated through mechanical scheduling and Kalman prediction, gyro_x, gyro_y, and gyro_z represent the gyroscope data collected by the IMU after error compensation, and accel_x, accel_y, and accel_z represent the accelerometer data collected by the IMU after error compensation.

[0035] Then perform random feature selection to obtain a data subset, use grid search to determine the hyperparameters of each tree in multiple different decision trees (decision tree 1, decision tree 2, decision tree 3,..., decision tree N), including depth parameter D, sample splitting parameter S, and leaf node parameter L, obtain multiple predictions (prediction 1, prediction 2, prediction 3,..., prediction N), average the multiple predictions, output the prediction result, that is, the predicted lateral velocity and vertical velocity, and finally calculate the mean squared error (MSE), root mean squared error (RMSE), and mean absolute error (MAE) to evaluate the model performance to obtain the hyperparameters N, D, S, and L of the best model.

[0036] In one embodiment, a GNSS unavailable state is determined, a vehicle motion prediction value is output by the vehicle position, velocity, and attitude prediction model, and according to the vehicle motion state classification result, the covariance matrix of the Kalman filter is adjusted, including: When GNSS is unavailable, input the vehicle motion data set into the vehicle position, velocity, and attitude prediction model to obtain the predicted lateral velocity and vertical velocity; According to the gyro Z-axis data collected by the IMU at the current moment, determine the real-time state and classification of vehicle motion; Subtract the predicted lateral velocity and vertical velocity from the lateral velocity and vertical velocity obtained through mechanical arrangement to obtain the updated observation value of the Kalman filter; Construct the measurement matrix for Kalman filter update by taking the total differential of the transformation error equation from the inertial navigation coordinate system to the wheel coordinate system; Set covariance matrices of different levels through the variances of lateral velocities and vertical velocities of different levels obtained from statistical data.

[0037] Among them, the Kalman filter is updated using the vehicle motion prediction value and the covariance matrix to obtain vehicle position, velocity, and attitude error compensation information, including: Combine the minimum variance estimate of the current moment state vector obtained through ODO Kalman update and the variance corresponding to the minimum variance estimate, perform NHC Kalman update, and output the minimum variance estimate of the current moment state vector obtained after NHC Kalman update and the variance corresponding to the minimum variance estimate.

[0038] Specifically, as Figure 7 shown, it is the Kalman filter update flowchart. First, obtain the input data set (velocity_x, gyro_x, gyro_y, gyro_z, accel_x, accel_y, accel_z), judge the real-time state of the vehicle body motion according to the gyro_z value at the current moment and set the classification L, input it into the random forest regression model, that is, the trained vehicle position, velocity, and attitude prediction model, and through the 0-level, 1-level, and 2-level regression models obtained by the foregoing classification, output the prediction, that is, the predicted lateral velocity V_y and vertical velocity V_z. Subtract V_y and V_z from the lateral and vertical velocities obtained through mechanical arrangement (which need to be converted to the wheel coordinate system and the lever arm error needs to be considered) to be used as the observation value Z for Kalman update. Construct the measurement matrix H in Kalman update by taking the total differential of the transformation error equation from the inertial navigation coordinate system to the wheel coordinate system. Set the covariance matrix R of different levels through the variances of lateral velocities and vertical velocities of different levels obtained from statistical data. Combine X k and P k, perform NHC Kalman update and output X obtained after intelligent NHC Kalman update k and P k .

[0039] Generally, the complete Kalman filtering process includes Kalman prediction and Kalman update. The above Xk refers to the minimum variance estimate of the current state vector X obtained after the second Kalman update in the Kalman filtering process, and Pk refers to the variance corresponding to Xk.

[0040] The embodiment of the present invention uses the predicted speed value and the adjusted covariance matrix to perform NHC Kalman filtering update to improve the inertial navigation accuracy.

[0041] Experimental verification shows that in a complex urban environment, this method can significantly improve the horizontal positioning accuracy. As Figure 8 shown, it shows the comparison of the navigation accuracy of the method of the present invention and the traditional method in different test environments. Among them Figure 8 in (1) and Figure 8 in (2) and Figure 8 in (3) and Figure 8 in (4) are the positioning results after using normal NHC integrated navigation and intelligent NHC integrated navigation respectively after manually disconnecting GNSS for 250s and 180s on different selected sections.

[0042] Among them Figure 8 in (1), the RMSE values of the northward position error and the eastward position error are 20.85m and 15.73m respectively, and the total horizontal position error RMSE value is 26.12m; the maximum northward position error and the eastward position error values are 47.23m and 64.07m respectively, and the maximum horizontal position error is 75.44m. And Figure 8 in (2), the RMSE values of the northward position error and the eastward position error are 2.27m and 3.20m respectively, and the total horizontal position error RMSE value is 3.93m; the maximum northward position error and the eastward position error values are 6.55m and 7.78m respectively, and the maximum horizontal position error is 8.07m. Therefore, after using the algorithm designed by the present invention, the RMSE value of the horizontal position positioning error has increased by 84.95%, and the maximum horizontal position error has increased by 89.3%.

[0043] Figure 8 in (3), the RMSE values of the northward position error and the eastward position error are 7.57m and 23.57m respectively, and the total horizontal position error RMSE value is 24.76m; the maximum northward position error and the eastward position error values are 18.68m and 67.84m respectively, and the maximum horizontal position error is 70.32m. And Figure 8The RMSE values of the (4) northward position error and eastward position error are 2.74 m and 5.68 m respectively, and the total horizontal position error RMSE value is 6.31 m; the maximum northward position error and eastward position error values are 6.03 m and 14.21 m respectively, and the maximum horizontal position error is 14.97 m. Therefore, after using the algorithm designed by the present invention, the RMSE value of the horizontal position positioning error has increased by 74.51%, and the maximum horizontal position error has increased by 78.71%.

[0044] It can be seen from the experimental results that the algorithm designed by the present invention significantly improves the horizontal positioning accuracy when GNSS fails. This method can improve the accuracy of vehicle navigation without adding additional sensors, especially in complex environments such as cities, and maintain high-precision positioning ability to provide an efficient and low-cost solution for vehicle navigation systems.

[0045] The vehicle motion constraint dual-adjustment integrated navigation system based on machine learning provided by the present invention will be described below. The vehicle motion constraint dual-adjustment integrated navigation system based on machine learning described below can be mutually referred to the vehicle motion constraint dual-adjustment integrated navigation method described above.

[0046] Figure 9 is a schematic structural diagram of the vehicle motion constraint dual-adjustment integrated navigation system based on machine learning provided by an embodiment of the present invention, as Figure 9 shown, including: an acquisition module 91, a training module 92, an adjustment module 93, and an optimization module 94, wherein: The acquisition module 91 is used to acquire SINS Kalman prediction data, determine the available state of GNSS, combine INS, ODO, and NHC, classify the vehicle motion state, determine the data selection features and label values, and obtain the vehicle motion data set; The training module 92 is used to construct a multi-level random forest regression model, train the multi-level random forest regression model based on the vehicle motion data set, and obtain a vehicle position velocity attitude prediction model; The adjustment module 93 is used to determine the unavailable state of GNSS, output the vehicle motion prediction value by the vehicle position velocity attitude prediction model, and adjust the covariance matrix of the Kalman filter according to the vehicle motion state classification result; The optimization module 94 is used to update the Kalman filter using the vehicle motion prediction value and the covariance matrix, obtain the vehicle position velocity attitude error compensation information, and output the optimized vehicle position velocity attitude output value according to the vehicle position velocity attitude error compensation information.

[0047] Figure 10 illustrates a schematic structural diagram of an electronic device, as Figure 10As shown in the figure, the electronic device may include: a processor 1010, a communications interface 1020, a memory 1030, and a communication bus 1040. Among them, the processor 1010, the communications interface 1020, and the memory 1030 complete communication with each other through the communication bus 1040. The processor 1010 may call the logical instructions in the memory 1030 to execute a combined navigation method for dual adjustment of vehicle motion constraints based on machine learning. The method includes: obtaining SINS Kalman prediction data, determining the available state of GNSS, combining INS, ODO, and NHC, grading the vehicle motion state, determining data selection features and label values, and obtaining a vehicle motion data set; constructing a multi-level random forest regression model, training the multi-level random forest regression model based on the vehicle motion data set to obtain a vehicle position, speed, and attitude prediction model; determining the unavailable state of GNSS, outputting vehicle motion prediction values by the vehicle position, speed, and attitude prediction model, and adjusting the covariance matrix of the Kalman filter according to the vehicle motion state grading result; updating the Kalman filter using the vehicle motion prediction values and the covariance matrix, obtaining vehicle position, speed, and attitude error compensation information, and outputting optimized vehicle position, speed, and attitude output values according to the vehicle position, speed, and attitude error compensation information.

[0048] In addition, when the logical instructions in the above-mentioned memory 1030 are implemented in the form of software functional units and sold or used as independent products, 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 foregoing storage medium includes: various media such as USB flash drives, mobile hard disks, read-only memories (ROM, Read-Only Memory), random access memories (RAM, Random Access Memory), magnetic disks, or optical discs that can store program codes.

[0049] On the other hand, the present invention also provides a computer program product, which includes a computer program. The computer program 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 machine learning-based dual-adjustment integrated navigation method for vehicle motion constraints provided by the above-mentioned various methods. The method includes: obtaining SINS Kalman prediction data, determining the available state of GNSS, combining INS, ODO, and NHC, grading the vehicle motion state, determining data selection features and label values, and obtaining a vehicle motion data set; constructing a multi-level random forest regression model, training the multi-level random forest regression model based on the vehicle motion data set to obtain a vehicle position, speed, and attitude prediction model; determining the unavailable state of GNSS, outputting vehicle motion prediction values by the vehicle position, speed, and attitude prediction model, and adjusting the covariance matrix of the Kalman filter according to the vehicle motion state grading result; updating the Kalman filter using the vehicle motion prediction values and the covariance matrix, obtaining vehicle position, speed, and attitude error compensation information, and outputting optimized vehicle position, speed, and attitude output values according to the vehicle position, speed, and attitude error compensation information.

[0050] In another aspect, 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 is implemented to execute the machine learning-based dual-adjustment integrated navigation method for vehicle motion constraints provided by the above-mentioned various methods. The method includes: obtaining SINS Kalman prediction data, determining the available state of GNSS, combining INS, ODO, and NHC, grading the vehicle motion state, determining data selection features and label values, and obtaining a vehicle motion data set; constructing a multi-level random forest regression model, training the multi-level random forest regression model based on the vehicle motion data set to obtain a vehicle position, speed, and attitude prediction model; determining the unavailable state of GNSS, outputting vehicle motion prediction values by the vehicle position, speed, and attitude prediction model, and adjusting the covariance matrix of the Kalman filter according to the vehicle motion state grading result; updating the Kalman filter using the vehicle motion prediction values and the covariance matrix, obtaining vehicle position, speed, and attitude error compensation information, and outputting optimized vehicle position, speed, and attitude output values according to the vehicle position, speed, and attitude error compensation information.

[0051] 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 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. A person of ordinary skill in the art can understand and implement it without creative efforts.

[0052] 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 essence of the above technical solution, or the part 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, magnetic disk, optical disk, etc., and includes several instructions to enable a computer device (which can be a personal computer, server, or network device, etc.) to execute the methods described in each embodiment or some parts of the embodiments.

[0053] 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 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 cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A dual-adjustment integrated navigation method for vehicle motion constraints based on machine learning, characterized in that Including: Obtain the Kalman prediction data of the strapdown inertial navigation system (SINS), determine the available state of the global navigation satellite system (GNSS), and combine the integrated inertial navigation system (INS), odometer (ODO), and nonholonomic constraint (NHC) to classify the vehicle motion state, determine the data selection features and label values, and obtain the vehicle motion data set. Construct a multi-level random forest regression model, and train the multi-level random forest regression model based on the vehicle motion data set to obtain a vehicle position, velocity, and attitude prediction model. Determine the unavailable state of GNSS, output the vehicle motion prediction value by the vehicle position, velocity, and attitude prediction model, and adjust the covariance matrix of the Kalman filter according to the vehicle motion state classification result. Update the Kalman filter using the vehicle motion prediction value and covariance matrix, obtain the vehicle position, velocity, and attitude error compensation information, and output the optimized vehicle position, velocity, and attitude output value according to the vehicle position, velocity, and attitude error compensation information.

2. The method for dual-adjustment integrated navigation of vehicle motion constraints based on machine learning according to claim 1, wherein Obtain the SINS Kalman prediction data, including: Combining IMU data, error compensation, mechanical arrangement, and Kalman prediction, collect the accelerometer, gyroscope data, and INS forward velocity output by the IMU under the available state of GNSS to obtain the SINS Kalman prediction data.

3. The machine learning-based dual adjustment integrated navigation method for vehicle motion constraints according to claim 2, wherein Determine the available state of GNSS, combine INS, ODO, and NHC to classify the vehicle motion state, determine the data selection features and label values, and obtain the vehicle motion data set, including: Determine the lateral velocity and vertical velocity corresponding to the time stamp of the available state of GNSS as the label values, and construct an initial training data set in combination with the SINS Kalman prediction data. Calculate the average value and standard deviation of the heading angular velocity of the gyroscope Z-axis data of all SINS Kalman prediction data, and determine the first-level vehicle motion state, the second-level vehicle motion state, and the third-level vehicle motion state. According to the classification result, divide the initial training data set into training data sets of different levels. Calculate the statistical average value and variance of the lateral velocity and vertical velocity in the training data sets of different levels respectively, and randomly divide the data set into a training set, a test set, and a validation set to form the vehicle motion data set.

4. The method for dual-adjustment integrated navigation of vehicle motion constraints based on machine learning according to claim 1, wherein Construct a multi-level random forest regression model, and train the multi-level random forest regression model based on the vehicle motion data set to obtain a vehicle position, velocity, and attitude prediction model, including: Input the vehicle motion data set. Perform random feature selection to obtain a data subset, input the data subset into multiple decision trees, and use grid search to determine the hyperparameter set of each tree. The hyperparameter set includes depth parameters, sample splitting parameters, and leaf node parameters, obtain multiple prediction values, and average the multiple prediction values to obtain an average prediction result, and obtain the lateral velocity prediction value and the vertical velocity prediction value. Evaluate the model using the mean square error, root mean square error, and mean absolute error to obtain the best hyperparameter set, and output the vehicle position, velocity, and attitude prediction model.

5. The combined navigation method for dual adjustment of vehicle motion constraints based on machine learning according to claim 1, characterized in that, Determine the unavailable state of GNSS, output the vehicle motion prediction value by the vehicle position, velocity, and attitude prediction model, and adjust the covariance matrix of the Kalman filter according to the vehicle motion state classification result, including: When GNSS is unavailable, input the vehicle motion data set into the vehicle position, velocity, and attitude prediction model to obtain the predicted lateral velocity and vertical velocity; Determine the real-time state and classification of vehicle motion according to the gyroscope Z-axis data collected by the IMU at the current moment; Subtract the predicted lateral velocity and vertical velocity from the lateral velocity and vertical velocity obtained through mechanical arrangement to obtain the observation value updated by the Kalman filter; Construct the measurement matrix updated by the Kalman filter by taking the total differential of the conversion error equation from the inertial navigation coordinate system to the wheel coordinate system; Set covariance matrices of different levels through the variances of lateral velocities and vertical velocities of different levels obtained from statistical data; 6. The method for dual-adjustment integrated navigation of vehicle motion constraints based on machine learning according to claim 5, wherein Update the Kalman filter using the vehicle motion prediction value and the covariance matrix to obtain the vehicle position, velocity, and attitude error compensation information, including: Combine the minimum variance estimate of the current moment state vector obtained through ODO Kalman update and the variance corresponding to the minimum variance estimate, perform NHC Kalman update, and output the minimum variance estimate of the current moment state vector obtained after NHC Kalman update and the variance corresponding to the minimum variance estimate.

7. A dual-adjustment integrated navigation system for vehicle motion constraints based on machine learning, characterized in that, Including: An acquisition module for acquiring SINS Kalman prediction data, determining the available state of GNSS, combining INS, ODO, and NHC, classifying the vehicle motion state, determining data selection features and label values, and obtaining the vehicle motion data set; A training module for constructing a multi-level random forest regression model and training the multi-level random forest regression model based on the vehicle motion data set to obtain the vehicle position, velocity, and attitude prediction model; An adjustment module for determining the unavailable state of GNSS, outputting the vehicle motion prediction value by the vehicle position, velocity, and attitude prediction model, and adjusting the covariance matrix of the Kalman filter according to the vehicle motion state classification result; An optimization module for updating the Kalman filter using the vehicle motion prediction value and the covariance matrix to obtain the vehicle position, velocity, and attitude error compensation information, and outputting the optimized vehicle position, velocity, and attitude output value according to the vehicle position, velocity, and attitude error compensation information.

8. 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 machine learning-based dual-adjustment integrated navigation method for vehicle motion constraints according to any one of claims 1 to 6.

9. 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 machine learning-based dual-adjustment integrated navigation method for vehicle motion constraints according to any one of claims 1 to 6.

10. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by the processor, it implements the machine learning-based dual-adjustment integrated navigation method for vehicle motion constraints according to any one of claims 1 to 6.

Images