An ins / gnss rod arm error online estimation method based on dimension reduction filtering

Abstract

The application discloses an INS / GNSS rod arm error online estimation method based on dimension reduction filtering. The method splits the INS / GNSS rod arm error and other error variables, constructs multiple low-dimensional filters, reduces the coupling between the error variables, has clear mechanism, simplifies the model, reduces the requirements of the combined navigation algorithm on the chip main frequency and the storage space, improves the calculation speed and the filtering estimation precision, and can improve the positioning performance of the vehicle-mounted combined navigation system after the INS / GNSS rod arm error online estimation and compensation, and has high engineering application value.

Description

Technical Field

[0001] This application belongs to the technical field of vehicle-mounted integrated navigation systems, and particularly relates to an online estimation method for INS / GNSS lever error based on dimensionality reduction filtering. Background Technology

[0002] Integrated navigation systems are indispensable modules in vehicle-assisted and autonomous driving, responsible for providing the vehicle with real-time, high-precision position, velocity, and attitude information. In-vehicle integrated navigation systems mainly consist of an inertial navigation system (INS) and a global navigation satellite system (GNSS). The INS primarily uses micro-electro-mechanical systems (MEMS), which are small, low-cost, and have a high data update rate, and do not rely on external information. However, due to the relatively low accuracy of the internal gyroscopes and accelerometers, they cannot operate independently for extended periods. Therefore, data fusion between the INS and GNSS modules is necessary to ensure high-precision navigation performance.

[0003] INS typically uses the geometric center of its internal Inertial Measurement Unit (IMU) as the navigation and positioning reference. GNSS uses the phase center of the receiving antenna as the reference. When GNSS operates in Real-time Kinematic (RTK) mode, its positioning accuracy can theoretically be better than 3-5 centimeters. In practical applications, there is a certain deviation in the installation position between INS and GNSS antennas, known as lever error. Lever error is unavoidable in practical use. If the lever error compensation of the vehicle-mounted integrated navigation system is not accurate enough, it will severely reduce the integrated navigation performance of INS / GNSS, making it impossible to achieve centimeter-level positioning accuracy.

[0004] Existing literature generally employs two approaches to handling INS / GNSS lever errors: physical measurement and filtering estimation. The physical measurement approach, as the name suggests, involves pre-measuring the lever length and then compensating during integrated navigation. This method requires high measurement accuracy, but in vehicle-mounted environments, numerous curved surfaces make precise measurement difficult. Furthermore, the accurate reference points for INS and GNSS are not readily visible, hindering positioning. Therefore, the physical measurement approach has certain limitations. The filtering estimation approach treats the INS / GNSS lever as an extended state variable, estimating it online along with velocity and position errors, sensor bias, etc., using a filter. The filtering estimation approach suffers from a large dimensionality of state variables, leading to a significant increase in storage space and computational load. Moreover, the severe coupling between state variables affects the convergence speed and accuracy of the filtering estimation. Therefore, to improve the positioning performance of high-precision vehicle-mounted integrated navigation systems, it is necessary to research a simple and easy-to-implement online estimation method for INS / GNSS lever errors, reducing the requirements of the integrated navigation algorithm on chip clock speed and storage space, improving computational speed and filtering estimation accuracy, and ensuring the positioning performance of vehicle-mounted integrated navigation systems in high-precision applications. Summary of the Invention

[0005] This application proposes an online estimation method for INS / GNSS lever error based on dimensionality reduction filtering. This method decomposes the INS / GNSS lever error and other error variables, constructs multiple dimensionality reduction filters, reduces the coupling between errors and storage space, has a clear mechanism, a simplified model, fast calculation speed, and high estimation accuracy. After online estimation and compensation of lever error, the positioning accuracy of the vehicle-mounted INS / GNSS integrated navigation system can be improved, demonstrating high engineering application value.

[0006] The technical solution adopted in this application is as follows:

[0007] An online estimation method for INS / GNSS boom error based on dimensionality reduction filtering is proposed, which includes the following steps:

[0008] Step 1: Start the INS / GNSS integrated navigation system and complete the initialization of the position and attitude navigation parameters of the integrated navigation system;

[0009] Step 2: The INS inertial navigation system updates its own navigation parameters, which include velocity, position, and attitude.

[0010] Step 3: Decompose the error state variables and construct a dimension reduction filter;

[0011] Step 4: Estimate the various error state variables online using Kalman filtering;

[0012] Step 5: Compensate for lever arm error and other error state variables.

[0013] Furthermore, in step 3, the construction of the dimensionality reduction filter involves constructing four state estimation filters, the model of which is represented as follows:

[0014]

[0015] Where i = 1, 2, 3, 4 corresponds to 4 filters, X i Z represents the state variable. i F represents the measurement variable. i H represents the system matrix. i Represents the measurement matrix; W i With V i These are system noise and measurement noise, respectively, both of which are white noise.

[0016] Furthermore, in step 3, the error state variable is broken down, including:

[0017] The INS / GNSS lever arm error to be estimated, along with other error variables, is split into four state estimation filters. The expressions for the corresponding state variables X1, X2, X3, and X4 are as follows:

[0018] X1=[φ U ε z ] T (2)

[0019]

[0020]

[0021]

[0022] Where φ E Indicates the deflection angle of the eastward-facing platform, φ N The symbol represents the northward platform deflection angle, φ. U Indicates the tilt angle of the astronautical platform, δV E Indicates the eastward velocity error, δV N Indicates northward velocity error, δV U δL represents the axial velocity error, δλ represents the latitude error, δh represents the altitude error, and ε represents the latitude error. z This indicates z-gyroscope drift. This indicates that the x-accelerometer has zero bias. This indicates that the y-accelerometer has zero bias. Indicates zero bias of the z-accelerometer, l x Indicates the lever arm in the x-direction, l y Indicates the lever arm in the y-direction, l z This indicates the lever arm in the z-direction.

[0023] Furthermore, in step 3, the constructed dimensionality reduction filter includes:

[0024] The system matrix expression for the four state estimation filters is as follows:

[0025]

[0026]

[0027]

[0028]

[0029] Where f U R represents the axial acceleration. x ,R y These represent the radii of the Earth's meridian and circumference, respectively, with L representing the local latitude. This represents the element in the i-th row and j-th column of the INS attitude transformation matrix;

[0030] The expressions for the measurement variables of the four state estimation filters are as follows:

[0031]

[0032] Z1=ψ INS -ψ GNSS (11)

[0033] Z2 = h INS -h GNSS -δp h (12)

[0034] Z3 = L INS -L GNSS -δp L (13)

[0035] Z4=λ INS -λ GNSS -δp λ (14)

[0036] Among them l x0 ,l y0 ,l z0 This represents the lever arm error value obtained during this power-on reading, ψ. GNSS ,ψ INS These represent the dual-antenna headings of GNSS and the headings of INS, respectively. GNSS ,h INS L represents the altitude of GNSS and INS respectively. GNSS ,L INS λ represents the latitude of GNSS and INS respectively. GNSS ,λINS These represent the longitudes of GNSS and INS, respectively.

[0037] Furthermore, in step 3, the constructed dimensionality reduction filter also includes: the measurement matrix expressions for the four state estimation filters are as follows:

[0038] H1 = [1 0] (15)

[0039] H2 = [0 1 0] (16)

[0040]

[0041]

[0042] In this context, the measurement matrix is ​​a one-dimensional variable, and the filter gain is also a one-dimensional variable.

[0043] Furthermore, in step 4, during online Kalman filtering estimation, state estimation filters X3 and X4 share the same state noise matrix Q. k and measurement noise matrix R k Using the same state variable expressions, system matrix, measurement variables, and measurement matrix, the measurements of these four dimensionality reduction filters are all one-dimensional variables and do not require matrix inversion operations, thus simplifying and reducing the amount of computation.

[0044] Furthermore, in step 5, the error compensation for the remaining state variables, including velocity, position, and attitude, is calculated using the following formula:

[0045]

[0046]

[0047]

[0048] in, The compensated eastward, northward, and celestial speeds, For the compensated latitude, longitude and altitude, X is the compensated attitude transformation matrix. i (j) represents the j-th state variable of the i-th filter.

[0049] Furthermore, in step 5, the error compensation for the remaining state variables includes the error compensation for the gyroscope and accelerometer, and the calculation formula is as follows:

[0050]

[0051] in This is to compensate for the z-gyroscope drift and the zero bias of the x, y, and z accelerometers.

[0052] Furthermore, the calculation formula for lever arm error compensation is as follows:

[0053]

[0054] in, This indicates the compensation for the lever arm error in the x-direction. This indicates the compensation for the lever arm error in the y-direction. Indicates the lever arm error compensation in the z-direction; X i (j) represents the j-th state variable of the i-th filter.

[0055] Furthermore, once the lever arm error converges, it can be written into the storage chip. When the integrated navigation system is started again, this error can be read and used to update the l value in the measurement variable expressions of the four state estimation filters. x0 ,l y0 ,l z0 .

[0056] Compared with the prior art, the beneficial effects of this application are as follows:

[0057] (1) The online estimation method for INS / GNSS arm error based on dimension reduction filtering proposed in this application decomposes the arm error and other errors into multiple dimension reduction filters. The mechanism is clear, the model is simplified, the coupling between various errors is reduced, and the online estimation accuracy of arm error is improved.

[0058] (2) The online estimation method for INS / GNSS lever error based on dimensionality reduction filtering proposed in this application has a low dimension of state variables and measurement variables, which significantly improves the calculation speed, reduces the requirements of the integrated navigation algorithm on chip frequency and storage space, is easy to implement, and demonstrates high engineering application value. Attached Figure Description

[0059] Figure 1 A flowchart illustrating the implementation of the online estimation method for lever arm error;

[0060] Figure 2 The simulation estimation curve of the mechanical system x-direction lever arm error in a specific embodiment of this application;

[0061] Figure 3 The simulation estimation curve of the mechanical system's y-direction lever arm error in a specific embodiment of this application;

[0062] Figure 4 The simulation estimation curve of the link arm error in the z-direction of the mechanical system in a specific embodiment of this application;

[0063] Figure 5 This is a motion trajectory diagram of an on-board test according to a specific embodiment of this application;

[0064] Figure 6 This is the true estimation curve of the mechanical system x-direction lever arm error in a specific embodiment of this application;

[0065] Figure 7 This is the true estimation curve of the linkage error in the y-direction of the mechanical system according to a specific embodiment of this application;

[0066] Figure 8 This is the true estimation curve of the lever arm error in the z-direction of the mechanical system according to a specific embodiment of this application;

[0067] Figure 9 This is a comparison curve of the eastward position error in the vehicle-mounted test according to a specific embodiment of this application;

[0068] Figure 10 This is a comparison curve of the northward position error in the vehicle-mounted test according to a specific embodiment of this application;

[0069] Figure 11 The above-ground position error comparison curve is shown in the vehicle-mounted test of a specific embodiment of this application. Detailed Implementation

[0070] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0071] The principle of the online estimation method for lever error in this application is as follows: When filtering and estimating the INS / GNSS lever error based on the extended state approach, the standard error model of the inertial navigation system is expressed as:

[0072]

[0073] The measurement variables for INS / GNSS integrated navigation include 7 dimensions, namely:

[0074]

[0075] If the aforementioned state equations and measurement information are directly used to estimate INS / GNSS boom errors and other errors through Kalman filtering, the state dimension is 18 and the measurement dimension is 7. This high dimensionality leads to severe coupling between state variables, resulting in not only a large computational burden during filtering but also affecting the accuracy of error variable estimation. The technical solution of this application decomposes the 18-dimensional state variables and 7-dimensional measurement variables, while simplifying the error equations. The simplification principles are as follows:

[0076] • In vehicle-mounted environments, the pitch and roll angles of the integrated navigation system are always relatively small, so the heading estimation loop and altitude estimation loop can be separated.

[0077] • Vehicle-mounted INS systems generally use MEMS sensors, which have relatively large zero-bias errors. Therefore, some error terms in the error equation can be ignored, thereby achieving decoupling between the eastward and northward position loops.

[0078] • The drift of the MEMS gyroscope can be directly calculated and the mean subtracted during the initial startup process. After subtraction, the gyroscope drift can be considered small, and there is no need to include the state variable in the filtering estimation.

[0079] Based on this, the standard error model and the 7-dimensional measurement variables of the inertial navigation system can be simplified, resulting in the following set of state equations:

[0080]

[0081]

[0082]

[0083]

[0084] The four state equations mentioned above correspond to the four dimensionality reduction filters proposed in this application. The dimensions of the four state models are now 2D, 3D, 8D, and 8D, respectively, a significant reduction compared to the original 18-dimensional state variables. When the GNSS is in RTK mode, position accuracy can reach the centimeter level, while velocity accuracy is not significantly improved. Therefore, when selecting measurement information, the GNSS velocity information can be ignored, retaining only the four-dimensional measurements of the GNSS dual-antenna heading, latitude, longitude, and altitude, which are precisely distributed across the four filters. Each filter has only one-dimensional measurement variables, thus avoiding the matrix inversion process. After separating the state and measurement variables, the coupling between errors is significantly reduced, thereby improving the estimation speed and accuracy of errors.

[0085] Furthermore, the computational complexity of Kalman filtering depends on the dimensions of the state and measurement variables. According to empirical formulas, the time complexity of Kalman filtering is O(m...). 2.376 +n 2 ), where m is the dimension of the measurement variable and n is the dimension of the state variable. The time complexity of the traditional method can be calculated to be 425.85, while the time complexity of the proposed dimension reduction filtering method is 145.0, reducing the computational load by approximately 65% ​​and significantly improving the computational speed.

[0086] The present application will now be further described with reference to the accompanying drawings.

[0087] Figure 1 A flowchart of the online estimation method for INS / GNSS lever error based on dimensionality reduction filtering, as proposed in this application, is presented. This method decomposes lever error and other errors into multiple dimensionality reduction filters, resulting in a clear mechanism, simplified model, reduced coupling between various errors, and improved online estimation accuracy of lever error. Furthermore, this method includes only five steps, and the designed state and measurement variables have low dimensionality, significantly improving computational speed, reducing the requirements of integrated navigation algorithms on chip clock speed and storage space, and making it easy to implement, demonstrating high engineering application value.

[0088] To verify the correctness of the online wheel speed meter error estimation method proposed in this application, a simulation verification experiment was first designed. An "8" shaped motion trajectory was generated by a trajectory generator. Various errors were added based on the basic performance of the MEMS inertial navigation, GNSS module and wheel speed meter. Then, the simulation estimation results of the INS / GNSS lever arm error were obtained through a simulation program. Figure 2 , Figure 3 and Figure 4 Simulation estimation curves of the arm error in the x, y, and z directions of the carrier coordinate system are presented respectively. The true values ​​of the arm error were set to -0.5m, 0.8m, and 1.0m during the simulation. The simulation results show that the estimated result of the arm in the x direction is -0.49m, the estimated result of the arm in the y direction is 0.80m, and the estimated result of the arm in the z direction is 1.02m. The arm estimation accuracy of INS / GNSS is better than 2cm, and the curve fluctuation is small during the estimation process, which proves that the online estimation method of INS / GNSS arm error proposed in this application is accurate and feasible.

[0089] Figure 5 The motion trajectory of the vehicle-mounted test for the specific application of this application is given. The zero-bias stability of the MEMS gyroscope and accelerometer in the inertial navigation system used in the test is 10° / h and 100ug, respectively. The GNSS module uses the UM482 chip, and the dynamic RTK accuracy can reach 2cm. The total duration of the vehicle-mounted test is approximately 1000 seconds. The driving route is located near the East Fourth Ring Road in Chaoyang District, Beijing. The operation includes straight-line acceleration and deceleration, turning, etc., and the heading motion covers the entire range of 0-360°. The maximum speed is approximately 30m / s.

[0090] Figure 6 , Figure 7 and Figure 8The online estimation results of the INS / GNSS lever arm error obtained from the vehicle-mounted test of this application are presented. The estimated results of the lever arm in the vehicle coordinate system are -0.26m, 0.20m and 1.21m, respectively. The error estimation curve converges quickly and has small fluctuations, which is basically consistent with the simulation results, proving that the online estimation results of the INS / GNSS lever arm error in the vehicle-mounted test are reliable.

[0091] In the vehicle-mounted test specifically applied in this application, a rough value for the INS / GNSS lever arm has already been obtained through prior measurement. To further verify the accuracy of the online estimation result of the INS / GNSS lever arm error in the vehicle-mounted test, the navigation data can be processed offline in two ways: online compensation of the lever arm according to the method of this application and compensation of the lever arm according to the prior measured value. After offline navigation, the position output is obtained and compared with the reference system in the vehicle-mounted test process, thereby verifying the accuracy of the online estimation of the INS / GNSS lever arm. Figure 9 , Figure 10 and Figure 11 The comparison curves of the eastward, northward, and overhead position errors in the vehicle-mounted test of this application are presented respectively. It can be seen that due to the low accuracy of the pre-measurement of the lever arm, the eastward and northward position errors exhibit step errors as the vehicle turns, while the overhead position has a certain constant error. After the online estimation and compensation of the INS / GNSS lever arm error proposed in this application, the eastward, northward, and overhead position errors are relatively stable with small fluctuations. Based on this, the position accuracy can be evaluated by calculating the root mean square of the position error. The root mean squares of the eastward, northward, and overhead position errors obtained by the pre-measurement method of the lever arm are 0.193m, 0.112m, and 0.108m, respectively. After the online estimation and compensation of the lever arm using the method of this application, the root mean squares of the eastward, northward, and overhead position errors are 0.041m, 0.047m, and 0.029m, respectively. The position accuracy is significantly improved, basically reaching the accuracy level of RTK positioning, proving the accuracy of the online estimation and compensation of the INS / GNSS lever arm error proposed in this application.

[0092] Although the illustrative specific embodiments of this application have been described above to enable those skilled in the art to understand this application, it should be understood that this application is not limited to the scope of the specific embodiments. For those skilled in the art, various changes are obvious as long as they are within the spirit and scope of this application as defined and determined by the appended claims, and all inventions utilizing the concept of this application are protected.

Claims

1. An online estimation method for INS / GNSS boom error based on dimensionality reduction filtering, characterized in that, The method includes the following steps: Step 1: Start the INS / GNSS integrated navigation system and complete the initialization of the position and attitude navigation parameters of the integrated navigation system; Step 2: The INS inertial navigation system updates its own navigation parameters, which include velocity, position, and attitude. Step 3: Decompose the error state variables and construct a dimension reduction filter; Step 4: Estimate the various error state variables online using Kalman filtering; Step 5: Compensate for lever arm error and other error state variables; In step 3, the error state variable is broken down, including: The INS / GNSS arm error to be estimated, along with other error variables, is broken down into four state estimation filters, corresponding to the state variables. The expression is: （2） （3） （4） （5） in Indicates the deflection angle of the east-facing platform. This represents the north-facing platform deflection angle. Indicates the tilt angle of the platform. Indicates eastward velocity error, Indicates northbound velocity error, Indicates the upward velocity error, Indicates latitude error, Indicates longitude error, Indicates height error, This indicates z-gyroscope drift. This indicates that the x-accelerometer has zero bias. This indicates that the y-accelerometer has zero bias. This indicates that the z-accelerometer has zero bias. Indicates the lever arm in the x-direction. Indicates the lever arm in the y-direction. This indicates the lever arm in the z-direction.

2. The method according to claim 1, characterized in that, In step 3, the construction of the dimensionality reduction filter involves constructing four state estimation filters, the model of which is expressed as follows: （1） in Corresponding to 4 filters, Represents state variables, Indicates the measurement variable, Represents the system matrix. Represents the measurement matrix; and These are system noise and measurement noise, respectively, both of which are white noise.

3. The method according to claim 1 or 2, characterized in that, In step 3, the constructed dimensionality reduction filter includes: The system matrix expression for the four state estimation filters is as follows: （6） （7） （8） （9） in Indicates celestial acceleration. These represent the radii of the Earth's meridian and circumference, respectively. Indicates the local latitude. The pose transformation matrix of INS is represented by the first... Line number Column elements; The expressions for the measurement variables of the four state estimation filters are as follows: （10） （11） （12） （13） （14） in This indicates the lever arm error value obtained during this power-on reading. These represent the dual-antenna heading of GNSS and the heading of INS, respectively. These represent the altitudes of GNSS and INS, respectively. These represent the latitude of GNSS and INS respectively. These represent the longitudes of GNSS and INS, respectively.

4. The method according to claim 3, characterized in that, In step 3, the constructed dimensionality reduction filter further includes: the measurement matrix expressions for the four state estimation filters are as follows: （15） （16） （17） （18） In this context, the measurement matrix is ​​a one-dimensional variable, and the filter gain is also a one-dimensional variable.

5. The method according to claim 1, characterized in that, In step 4, during online Kalman filtering estimation, state estimation filters X3 and X4 share the same state noise matrix. and measurement noise matrix Using the same state variable expressions, system matrix, measurement variables, and measurement matrix, the measurements of these four dimensionality reduction filters are all one-dimensional variables and do not require matrix inversion operations, thus simplifying and reducing the amount of computation.

6. The method according to claim 1, characterized in that, In step 5, the error compensation for the remaining state variables, including velocity, position, and attitude, is calculated using the following formulas: （19） （20） （21） in, The compensated eastward, northward, and celestial speeds, For the compensated latitude, longitude and altitude, The compensated attitude transformation matrix, Representing the The filter of the nth filter There are several state variables.

7. The method according to claim 1, characterized in that, In step 5, the error compensation for the remaining state variables includes the error compensation for the gyroscope and accelerometer. The calculation formula is as follows: （22） in This is to compensate for the z-gyroscope drift and the zero bias of the x, y, and z accelerometers.

8. The method according to claim 3, characterized in that, The formula for calculating lever arm error compensation is as follows: （23） in, This indicates the compensation for the lever arm error in the x-direction. This indicates the compensation for the lever arm error in the y-direction. This indicates the error compensation of the lever arm in the z-direction; Representing the The filter of the nth filter There are several state variables.

9. The method according to claim 8, characterized in that, Once the lever arm error converges, it can be written into the storage chip. When the integrated navigation system starts up next time, it will be read out and the expression of the measurement variables of the four state estimation filters will be updated. .

Images