Vehicle GNSS positioning method based on multi-motion model interaction

Abstract

The invention discloses a vehicle GNSS positioning method based on multi-motion model interaction, and the method comprises the steps: building a PCV model and a PCSAV model of a carrier according to the linear driving posture and the turning driving posture of the carrier, so as to obtain a state estimation vector and a state transition matrix of the carrier at the last moment based on the PCV model and the PCSAV model; an interactive multi-model is introduced to establish a heuristic position-speed filtering HPV-IMM model based on the interactive multi-model, information filtering interaction between a PCV model and a PCSAV model is realized, and a state estimation vector and an error covariance matrix of a carrier at the current moment are obtained, so that the position and the speed of the carrier at the current moment are obtained. The problem that a traditional single kinematic model is low in precision in multi-motion-attitude vehicle positioning is solved, and the method is suitable for multi-attitude vehicle motion dynamic navigation in an urban environment.

Description

Technical Field

[0001] The present invention relates to the technical field of vehicle navigation and positioning, and in particular to a vehicle GNSS positioning method based on multi-motion model interaction. Background Art

[0002] In an open and interference-free environment, the SPP positioning accuracy of ordinary vehicle-mounted navigation receivers can reach the meter level, and the RTD positioning accuracy can reach the sub-meter level. However, the large number of scenes in cities such as urban canyons, tree-lined areas, tunnels, and viaducts can easily cause obstruction and deception of satellite signals. Among them, the serious non-line-of-sight (NLOS) and multipath effects make it impossible to guarantee the availability and continuity of SPP / RTD positioning. In extreme environments, the positioning error may reach tens of meters or even hundreds of meters. In the face of the above difficulties, in order to improve the robustness and anti-interference ability of the navigation system when facing abnormal observation data, there are roughly two types of solutions without relying on external sensor information assistance:

[0003] First, in the observation domain, the representative methods are Fault Detection and Exclusion (FDE) and robust estimation. The core goal of the former is to identify and exclude those observations with significant deviations by executing a series of statistical test procedures, thereby ensuring the accuracy and stability of the system output. The latter focuses on optimizing the matching of observations and their weights to reduce the adverse effects of those observations with large errors and high weights on the overall filtering results. Both of the above methods can be implemented on prior information and a posteriori residuals. FDE and robust estimation based on a posteriori residuals comprehensively consider the prior information and observation data, and can better reflect the distribution of observation residuals at the current moment. However, due to the correlation between observations, some gross errors are assigned to other normal observations, which is easy to cause missed detection and false alarms. Although FDE and robust estimation based on prior information can avoid the negative impact of gross error transfer, they have more stringent requirements on the accuracy of state forecasting.

[0004] Second, in the state domain, at present, most navigation receivers on the market can receive Doppler observation signals, and GNSS multi-system single-point speed measurement can obtain cm / s-level accuracy. Considering that the vehicle's motion state is relatively stable, using the carrier's prior coordinates and Doppler speed measurement information to predict the current coordinates is more robust than the traditional single-point positioning coordinate update method. In the past 20 years, many scholars have discussed the problem of carrier motion models and established various models including the constant velocity (CV) model and the constant acceleration (CA) model to describe the carrier's motion behavior. Among them, the CV model is the most widely used, but it is too ideal and difficult to accurately describe the complex motion state of vehicles driving in the city, especially when the vehicle is in a large curvature turning state. The prediction effect is poor. Summary of the invention

[0005] Purpose of the invention: The purpose of the present invention is to provide a vehicle GNSS positioning method based on multi-motion model interaction, which can accurately locate the multi-posture motion state of the vehicle.

[0006] Technical solution: To achieve the above-mentioned purpose, the vehicle GNSS positioning method based on multi-motion model interaction described in the present invention includes:

[0007] According to the two postures of the carrier in straight-line driving and turning driving, a heuristic position-constant velocity PCV model and a position-constant angular velocity PCSAV model of the carrier are respectively established to obtain the state estimation vector and state transfer matrix of the carrier at the previous moment based on the PCV model and the PCSAV model, wherein the state estimation vector includes the position and velocity of the carrier, and the state transfer matrix is ​​used to transfer the state vector at the previous moment to the next moment;

[0008] An interactive multi-model is introduced, and a heuristic position-velocity filtering HPV-IMM model based on the interactive multi-model is established. The HPV-IMM model predicts and measures the state estimation vector and error covariance matrix of the carrier at the current moment based on the PCV model and PCSAV model according to the state estimation vector and state transfer matrix of the PCV model and PCSAV model at the previous moment, and updates the state vector and error covariance matrix of the carrier of the PCV model and PCSAV model at the current moment after the measurement update; the state vector and error covariance matrix of the carrier of the PCV model and PCSAV model at the current moment after the measurement update are fused to obtain the state estimation vector and error covariance matrix of the carrier at the current moment, and the position and velocity of the carrier at the current moment are obtained based on the state estimation vector of the carrier at the current moment.

[0009] Among them, in the PCV model, the state vector to be estimated by the carrier is expressed as:

[0010] XPCV =[x,y,z,V x ,V y ,V z ];

[0011] The state transfer matrix of the carrier is expressed as:

[0012]

[0013] Where dt represents the time interval between adjacent epochs; (x, y, z) represents the position of the carrier at a certain moment, (V x ,V y ,V z ) represents the velocity of the carrier at position (x, y, z); the state transfer matrix transfers the state vector of the previous moment to the next moment according to the time interval dt.

[0014] Among them, in the PCSAV model, the state vector to be estimated by the carrier can be expressed as:

[0015] X PCSAV =X PCV ,

[0016] Where, X PCV Represents the state vector to be estimated for the carrier in the PCV model;

[0017] The carrier state transfer matrix is ​​expressed as:

[0018]

[0019] Where ω(T) represents the carrier angular velocity at time T; H represents the conversion matrix from the ECEF system to the ENU system.

[0020] The interactive multi-HPV-IMM model includes a state interactive input module, a state prediction module, an innovation outlier detection module, a measurement update module, and an overall output interactive module;

[0021] Among them, the state interaction input module calculates the initial estimated state vector of the i-sub-model carrier at the current k+1 time according to the estimated result of the state vector of the i-sub-model carrier at the previous time k and its error covariance matrix Wherein, when i=1, the i sub-model represents the PCV model, when i=2 or M, the i sub-model represents the PCSAV model, and M=2 represents the number of sub-models;

[0022] The state prediction module is based on the initialization state vector and the error covariance matrix Predict the motion state of the i-sub-model carrier from time k to time k+1, and obtain the state prediction vector of the i-sub-model carrier from time k to time k+1 and the error covariance matrix

[0023] The innovation outlier detection module predicts the vector according to the state of the i-submodel carrier. and the error covariance matrix Calculate the new information sequence of sub-model i at time k+1

[0024] The measurement update module includes measurement update and sub-model probability update. The measurement update is based on the prediction vector of the state of the i sub-model carrier Error covariance matrix Innovation Sequence Perform measurement update to obtain the state vector after measurement update of the i-submodel carrier and the error covariance matrix Sub-model probability update is to calculate the similarity between the i-sub-model and the current carrier motion state, and obtain the posterior probability of the i-sub-model

[0025] The overall interactive output module converts the state estimation vector of the i sub-model carrier into and the error covariance matrix According to the posterior probability Perform weighted fusion to obtain the weighted fusion and the error covariance matrix That is, the state estimation vector and error covariance matrix of the carrier at time k+1;

[0026] The position and velocity of the carrier at time k+1 are obtained based on the state estimation vector of the carrier at time k+1.

[0027] Among them, the i sub-model carrier initializes the state vector and the error covariance matrix It is expressed as:

[0028]

[0029]

[0030] In the formula, Represents the parameters of the i-th sub-model at time k, including the error covariance matrix State transition matrix Process Noise Matrix Coefficient Matrix Observation noise matrix Represents the observation data at time k, including the pseudorange value of the satellite and Doppler measurements

[0031] Indicated by arrive The transition probability of Represents the probability of the sub-model after the interaction in the input interaction module, through the prior transfer probability matrix π and the model posterior probability Calculated, the transition probability matrix is ​​set to

[0032] Among them, the state prediction vector of the i-submodel carrier from time k to time k+1 is and the error covariance matrix It is expressed as:

[0033]

[0034] In the formula, Describe the state transfer matrix of the i-submodel carrier at time k+1, Represents the process noise matrix of the i-th sub-model at time k+1.

[0035] Among them, the new information sequence of the i-submodel at time k+1 is It is expressed as:

[0036]

[0037] In the formula, represents the observed data after eliminating gross errors at time k+1, Represents the coefficient matrix of the i-submodel at time k+1.

[0038] Among them, a multi-dimensional statistical analysis outlier detection algorithm is used to identify and eliminate the observation value gross errors in the observation values, and the observation value gross errors refer to the observation values ​​in which the errors of the observation values ​​are greater than the set threshold.

[0039] Among them, the i-submodel carrier measures the updated state vector and the error covariance matrix It is expressed as:

[0040]

[0041] Where I represents the identity matrix (the main diagonal elements are 1), represents the coefficient matrix of the i-submodel at time k+1; represents the Kalman gain

[0042] The posterior probability of the i submodel It is expressed as:

[0043]

[0044] In the formula, Represents the likelihood function value of the i-submodel at time k+1.

[0045] Among them, the weighted fusion is obtained and the error covariance matrix That is, the state estimation vector and error covariance matrix of the carrier at time k+1 are expressed as:

[0046]

[0047] Beneficial effects: The present invention has the following advantages: The interactive multi-model proposed in the present invention fully considers the importance of the kinematic model for GNSS vehicle navigation and positioning, and based on the principle of total probability, realizes real-time interaction of multiple vehicle kinematic models through the interactive multi-model, solves the problem of low accuracy of the traditional single kinematic model in multi-motion posture vehicle positioning, and is suitable for dynamic navigation of multi-posture vehicle motion in urban environments. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] Figure 1 It is a heuristic PCV model;

[0049] Figure 2 It is the heuristic PCSAV model;

[0050] Figure 3 Schematic diagram of the flow of information filtering interaction between PCV model and PCSAV model for interactive multi-model implementation;

[0051] Figure 4 A schematic diagram of the interactive multi-model realizing the real-time interaction of the PCV model and the PCSAV model in an open scene;

[0052] Figure 5 A schematic diagram showing the comparison of the positioning trajectories of the HPV-IMM model and the prior art;

[0053] Figure 6 This is the positioning effect diagram of the HPV-IMM model in a complex environment before and after the application of the MSA-OD algorithm. DETAILED DESCRIPTION

[0054] The technical solution of the present invention is described in detail below in conjunction with the embodiments and drawings.

[0055] The vehicle GNSS positioning method based on multi-motion model interaction described in the present invention comprises the following steps:

[0056] Step 1: Establish a heuristic PCV model (position-constant velocity model) and a PCSAV model (position-constant angular velocity model) according to the two postures of the carrier (vehicle) in straight-line driving and turning driving respectively;

[0057] The heuristic PCV model is as follows Figure 1 As shown, when the carrier (vehicle) is in a straight-line driving state, it can be constrained to the PCV model, that is, it is considered that the carrier movement speed at time T3 is equal in magnitude and unchanged in direction compared to time T2, and the carrier position at time T3 is predicted by the instantaneous speed at time T2.

[0058] In the PCV model, the clock error and clock rate of the onboard receiver are eliminated by inter-satellite difference, and the state vector to be estimated by the carrier can be expressed as:

[0059] X PCV =[x,y,z,V x ,V y ,V z ] (1);

[0060] The state transfer matrix of the carrier is:

[0061]

[0062] Where dt represents the time interval between adjacent epochs; (x, y, z) represents the position of the carrier at a certain moment, (V x ,V y ,V z ) represents the speed of the carrier at the position (x, y, z); the state vector usually contains the position and speed of the carrier at a certain moment, and the state transfer matrix transfers the state vector of the previous moment to the next moment according to the time interval dt.

[0063] The heuristic PCSAV model is as follows Figure 2 As shown, when the carrier is in a turning state, it is constrained to the PCSAV model, that is, it is considered that the speed of the carrier at time T3 remains unchanged compared to time T2, and the direction is corrected by the angular velocity of the carrier at time T2, and the position of the carrier at time T3 is predicted by the average speed of the carrier from time T2 to time T3.

[0064] In the PCSAV model, the state vector to be estimated by the carrier can be expressed as:

[0065] X PCSAV =X PCV (3);

[0066] The carrier state transfer matrix can be expressed as

[0067]

[0068] Where:

[0069]

[0070] It represents the carrier angular velocity at time T, that is, the rate of change of the carrier heading, which can be derived from the heading angle θ at the current time and the previous time, and the heading angle θ can be derived from the carrier's eastward and northward speeds (V e ,V n ) calculated.

[0071]

[0072] represents the conversion matrix from ECEF system to ENU system, B and L are the initialization state vectors of PCV or PCSAV model respectively. The latitude and longitude of the medium carrier.

[0073]

[0074] Indicates that the horizontal component of the carrier velocity rotates around the origin The rotation matrix of degrees.

[0075] The heuristic PCV and PCSAV models established by the present invention do not rely on data input provided by external sensors. All parameter calculations of the two models only rely on data at the current moment and the previous moment, thereby reducing the computational complexity and significantly improving the real-time performance of the vehicle position calculation process compared to some methods that require motion constraints through sliding window fitting trajectories.

[0076] Step 2: If Figure 3 As shown, an interactive multi-model (HPV-IMM model) is established to realize information filtering interaction between the PCV model and the PCSAV model;

[0077] The HPV-IMM model consists of five modules: state interaction input module, state prediction module, new information outlier detection, measurement update module (including measurement update and sub-model probability update), and overall output interaction module.

[0078] (1) The state interaction input module calculates the initial estimated state of the sub-model carrier at the current k+1 time based on the state estimation of the sub-model (PCV model, PCSAV model) at the previous time (referring to the estimated result of the sub-model state vector), the sub-model posterior probability, and the transition probability matrix π and its error covariance Among them, i=1 represents the PCV model, i=2 or M represents the PCSAV model, and M=2 represents the number of sub-models.

[0079] Different from the extended Kalman filter (EKF) of a single model, in the state prediction process, the interactive multi-model does not directly use the state estimate and error covariance matrix of the previous epoch for the state prediction of the current epoch, but initializes the sub-models at the beginning of each epoch and performs interactive fusion in the state domain.

[0080] According to the total probability theorem, the probability density function (PDF) of the state estimate of each sub-model at time k+1 can be initialized as:

[0081]

[0082] The initialization state vector of each sub-model at time k+1 and its error covariance are expressed as:

[0083]

[0084] In the formula,

[0085]

[0086] Represents the relevant parameters of the i-submodel at time k, including: error covariance matrix State transition matrix Process Noise Matrix Coefficient Matrix Observation noise matrix

[0087]

[0088] represents the observation value data at time k, which is specifically the pseudorange and Doppler measurement value of the satellite in this embodiment.

[0089]

[0090] Indicated by arrive The transition probability can be calculated through the prior transition probability matrix (TransitionProbability Matrix, TPM) Π and the posterior probability of the sub-model.

[0091] The superscript “-” in the above variables indicates the prediction result (prior), “~” indicates the interaction result, and “^” indicates the estimated result after the measurement update (posterior). The posterior probability of the sub-model refers to the probability after the probability update. The transition probability matrix is The subscript k / k above represents the variable at time k, k-1 / k-1 represents the variable at time k-1, and k+1 / k represents the prediction from the variable at time k to time k+1.

[0092] (2) The state prediction module is independent in each sub-model filter (i.e., each sub-model filter has a state prediction module). The initial state vector obtained by the state interaction input module and the error covariance matrix On this basis, the motion state of the sub-model carrier from time k to time k+1 is predicted:

[0093]

[0094] In the formula, Represents the state prediction vector and error covariance matrix of the i-submodel carrier from time k to time k+1 Describe the state transfer matrix of the sub-model carrier at time k+1.

[0095] (3) Outlier detection module, based on the state prediction vector obtained by the state prediction module Based on this, calculate the new information sequence of the sub-model filter at time k+1

[0096]

[0097] In the formula, represents the observed data after eliminating gross errors at time k+1, Represents the coefficient matrix of the i-submodel at time k+1.

[0098] in,

[0099]

[0100] In the formula, represents the pseudorange innovation sequence, represents the Doppler innovation sequence.

[0101] When performing pseudorange single-point positioning and Doppler single-point velocity measurement, the original pseudorange and Doppler observation equations can be expressed as:

[0102]

[0103] Ignore the satellite clock error cdt s , ionospheric delay d ion and the tropospheric delay d trop The modeling error is as follows. From the above observation equation, we can know that the pseudorange innovation sequence Mainly includes receiver clock difference cdt r , satellite-to-ground range ρ deviation caused by forecast position deviation and pseudorange observation noise ε PR . Same pseudorange innovation sequence, Doppler innovation sequence Mainly determined by the receiver clock speed Change rate of satellite-to-ground distance caused by predicted velocity deviation Bias and Doppler observation noise ε DOP composition.

[0104] On the premise of ensuring that the errors of the predicted position and velocity are within an acceptable range, the pseudorange and Doppler news sequences can be regarded as biased but overall stable data sequences. The outliers of the sequence caused by the gross errors of the observations can be effectively identified and eliminated by using the MSA-OD algorithm. It is worth noting that in the process of outlier identification, the existence of inter-system bias (ISB) of satellite systems such as GPS and Beidou requires independent analysis of pseudorange news sequences from different systems. In contrast, the analysis of the Doppler sequence does not need to consider this issue specifically.

[0105] (4) The model measurement update module is included in each independent filter, based on the state prediction vector and its error covariance matrix Innovation Sequence The EKF is used to update the measurement and obtain the state vector after the measurement update. and its error covariance

[0106] Calculate the Kalman gain:

[0107]

[0108] Update status:

[0109]

[0110] Update the error covariance:

[0111]

[0112] Where I represents the identity matrix (the main diagonal elements are 1).

[0113] The posterior probability of the submodel is updated by the current epoch and the model likelihood function value. That is, by calculating the similarity between the submodel and the current carrier motion state, the probability that best fits the current submodel is obtained. The pseudorange and Doppler observations theoretically conform to the normal distribution, and their likelihood function values ​​can be expressed by the following formula:

[0114]

[0115] Among them, det() means calculating the determinant of the square matrix;

[0116]

[0117] is the innovation error covariance calculated by the error propagation law.

[0118] Therefore, according to Bayes' theorem, the submodel posterior probability can be expressed as:

[0119]

[0120] (5) The overall interactive output module is used to convert each sub-model state estimation vector and the error covariance matrix According to the posterior probability Weighted fusion is performed to obtain the final total output of the HPV-IMM model.

[0121]

[0122] According to formula (25), the position and velocity of the carrier at time k+1 are obtained, where the error covariance matrix measures the error of the state vector.

[0123] In the above process, in view of the impact of non-Gaussian distribution observations on the update accuracy of the measurement update module in the urban canyon scenario, a multi-dimensional statistical analysis outlier detection algorithm is used to identify and eliminate the observation value errors, thereby improving the accuracy of the probability of real-time update of the sub-model. The observation value error refers to the observation value with a large error in the observation value, and the observation value corresponds to Equation 12.

[0124] The process of multi-dimensional statistical analysis outlier detection algorithm identifying and eliminating gross errors in observation values ​​is as follows:

[0125] (1) Input the N-dimensional pseudorange or Doppler innovation vector L, the maximum and minimum values ​​of the sequence compared with the mean test threshold K0, the maximum and minimum values ​​of the sequence compared with the median test threshold K1, and the maximum and minimum values ​​of the difference sequence compared with the test threshold K2;

[0126] (2) Initialize the N-dimensional outlier indicator sequence (0: healthy, 1: abnormal), set the list flag = [00...0] T , test cycle number k = 0,

[0127] (3) If Nk<3, end; otherwise, remove the data marked with gross errors in the L sequence;

[0128] (4) Calculation of test quantity:

[0129] Calculate the maximum value element L in L max , minimum element L min and its position i,j in the sequence;

[0130] Calculate the mean Mu and median Med of the L sequence;

[0131] Calculate the maximum value element D in the difference sequence Lmax and the minimum value element D min ;

[0132] (5) Outlier detection: If

[0133] L max -L min -2Mu < K0 && L max -Med < K1 && Med - L min < K1 && D max -D min < K2, that is, there are no outliers, end;

[0134] (6) Outlier marking. If L max -L min -2Mu ≥ K0 || L max -Med ≥ K1 || D max -D min ≥ K2, then flag(i) = 1;

[0135] If L max -L min -2Mu ≤ -K0 || L max -Med ≤ -K1 || D max -D min ≤ -K2, then flag(j) = 1;

[0136] (7) Update k to the new k + 1 and re - execute (3).

[0137] As Figure 4 shown, the HPV - IMM model proposed by the present invention can successfully achieve accurate real - time interaction of two sub - models in an open scenario. The red line represents the probability of the PCV model, and the green line represents the probability of the PCSAV model.

[0138] As Figure 5 shown, the positioning trajectory of the HPV - IMM is smoother and the positioning accuracy is higher than that of the RTD model in terms of positioning. In terms of speed measurement, it solves the problem of "jumping points" that is prone to occur in the SPV. Compared with the traditional PV model, the HPV - IMM model is approximately equivalent to the PV model when the vehicle is in a straight - line driving section, and the positioning and speed - measurement accuracies are comparable. When the vehicle is in a turning driving section, due to the real - time probability update mechanism of the HPV - IMM model, the PCSAV model that is more in line with the current vehicle motion behavior has a greater weight, and the positioning and speed - measurement accuracies are better than those of the PV model. The positioning accuracy can be increased by 27.2%, 50%, and 4.9% in the E, N, and U directions respectively compared with the PV model. The speed - measurement accuracy can be increased by 30%, 40%, and 9.0% respectively. In addition, as Figure 6 shown, after eliminating the gross error of the observed values by the MSA - OD algorithm, the robustness of the HPV - IMM model is effectively improved.

Claims

1. A vehicle GNSS positioning method based on multi-motion model interaction, characterized in that: include: According to the two postures of the carrier in straight-line driving and turning driving, a heuristic position-constant velocity PCV model and a position-constant angular velocity PCSAV model of the carrier are respectively established to obtain the state estimation vector and state transfer matrix of the carrier at the previous moment based on the PCV model and the PCSAV model, wherein the state estimation vector includes the position and velocity of the carrier, and the state transfer matrix is ​​used to transfer the state vector at the previous moment to the next moment; A heuristic position-velocity filtering HPV-IMM model based on interactive multi-models is established. The HPV-IMM model predicts and measures the state estimation vector and error covariance matrix of the carrier at the current moment based on the PCV model and PCSAV model according to the state estimation vector and state transfer matrix of the PCV model and PCSAV model at the previous moment, and updates the state vector and error covariance matrix of the carrier of the PCV model and PCSAV model at the current moment after measurement update; the state vector and error covariance matrix of the carrier of the PCV model and PCSAV model at the current moment after measurement update are fused to obtain the state estimation vector and error covariance matrix of the carrier at the current moment, and the position and velocity of the carrier at the current moment are obtained based on the state estimation vector of the carrier at the current moment.

2. The vehicle GNSS positioning method based on multi-motion model interaction according to claim 1 is characterized in that: In the PCV model, the state vector to be estimated by the carrier is expressed as: X PCV =[x,y,z,V x ,V y ,V z ]; The state transfer matrix of the carrier is expressed as: Where dt represents the time interval between adjacent epochs; (x, y, z) represents the position of the carrier at a certain moment, (V x ,V y ,V z ) represents the velocity of the carrier at position (x, y, z); the state transfer matrix transfers the state vector of the previous moment to the next moment according to the time interval dt.

3. The vehicle GNSS positioning method based on multi-motion model interaction according to claim 1 is characterized in that: In the PCSAV model, the state vector to be estimated by the carrier can be expressed as: X PCSAV =X PCV , Where, X PCV Represents the state vector to be estimated for the carrier in the PCV model; The carrier state transfer matrix is ​​expressed as: Where ω(T) represents the carrier angular velocity at time T; H represents the conversion matrix from the ECEF system to the ENU system.

4. The vehicle GNSS positioning method based on multi-motion model interaction according to claim 1 is characterized in that: The HPV-IMM model includes a state interaction input module, a state prediction module, an innovation outlier detection module, a measurement update module, and an overall output interaction module; Among them, the state interaction input module calculates the initial estimated state vector of the i-sub-model carrier at the current k+1 time according to the estimated result of the state vector of the i-sub-model carrier at the previous time k and its error covariance matrix Wherein, when i=1, the i sub-model represents the PCV model, when i=2 or M, the i sub-model represents the PCSAV model, and M=2 represents the number of sub-models; The state prediction module is based on the initialization state vector and the error covariance matrix Predict the motion state of the i-sub-model carrier from time k to time k+1, and obtain the state prediction vector of the i-sub-model carrier from time k to time k+1 and the error covariance matrix The innovation outlier detection module predicts the vector according to the state of the i-submodel carrier. and the error covariance matrix Calculate the new information sequence of sub-model i at time k+1 The measurement update module includes measurement update and sub-model probability update. The measurement update is based on the state prediction vector of the i sub-model carrier. Error covariance matrix Innovation Sequence Perform measurement update to obtain the state vector after measurement update of the i-submodel carrier and the error covariance matrix Sub-model probability update is to calculate the similarity between the i-sub-model and the current carrier motion state, and obtain the posterior probability of the i-sub-model The overall interactive output module converts the state estimation vector of the i sub-model carrier into and the error covariance matrix According to the posterior probability Perform weighted fusion to obtain the weighted fusion and the error covariance matrix That is, the state estimation vector and error covariance matrix of the carrier at time k+1; The position and velocity of the carrier at time k+1 are obtained based on the state estimation vector of the carrier at time k+1.

5. The vehicle GNSS positioning method based on multi-motion model interaction according to claim 4 is characterized in that: The i sub-model vector initializes the state vector and the error covariance matrix It is expressed as: In the formula, Represents the parameters of the i-th sub-model at time k, including the error covariance matrix State transition matrix Process Noise Matrix Coefficient Matrix Observation noise matrix Represents the observation data at time k, including the pseudorange value of the satellite and Doppler measurements Indicated by arrive The transition probability of Represents the probability of the sub-model after the interaction in the input interaction module, through the prior transfer probability matrix π and the model posterior probability Calculated, the transition probability matrix is ​​set to 6. The vehicle GNSS positioning method based on multi-motion model interaction according to claim 4 is characterized in that: The state prediction vector of the i-submodel carrier from time k to time k+1 and the error covariance matrix It is expressed as: In the formula, Describe the state transfer matrix of the i-submodel carrier at time k+1, Represents the process noise matrix of the i-th sub-model at time k+1.

7. The vehicle GNSS positioning method based on multi-motion model interaction according to claim 4 is characterized in that: The k+1 time i sub-model new information sequence It is expressed as: In the formula, represents the observed data after eliminating gross errors at time k+1, Represents the coefficient matrix of the i-submodel at time k+1.

8. The vehicle GNSS positioning method based on multi-motion model interaction according to claim 7 is characterized in that: A multi-dimensional statistical analysis outlier detection algorithm is used to identify and eliminate observation value gross errors in the observation values, where the observation value gross errors refer to observation values ​​in which the errors are greater than a set threshold.

9. The vehicle GNSS positioning method based on multi-motion model interaction according to claim 4, characterized in that: The i sub-model carrier measures the updated state vector and the error covariance matrix It is expressed as: In the formula, I represents the unit matrix, represents the coefficient matrix of the i-submodel at time k+1; represents the Kalman gain The posterior probability of the i submodel It is expressed as: In the formula, represents the likelihood function value of the i-submodel at time k+1, Represents the probability of the sub-model after the interaction in the input interaction module.

10. The vehicle GNSS positioning method based on multi-motion model interaction according to claim 4, characterized in that: The weighted fusion and the error covariance matrix That is, the state estimation vector and error covariance matrix of the carrier at time k+1 are expressed as:

Images