Multi-source information fusion navigation method based on variational Bayesian factor graph

Abstract

The invention belongs to the technical field of multi-source information fusion navigation, and particularly relates to a multi-source information fusion navigation method based on a variational Bayesian factor graph, a factor graph model is constructed by combining factor nodes of an inertial navigation system (INS), a scene matching navigation system (IMNS) and a global navigation satellite system (GNSS), and navigation state information is obtained according to a factor graph reasoning method. In order to solve the problem that the performance of a traditional multi-source information fusion navigation algorithm is reduced due to the fact that noise is unknown or suddenly changed, observation noise is modeled into Student-T distribution, then the system state is predicted, and the navigation state and unknown noise parameters are iterated by means of a Gaussian Newton iteration method and variational Bayesian learning. According to the invention, the algorithm is high in universality, has the characteristics of plug and play, and is high in expansibility and robustness.

Description

Technical Field

[0001] The invention belongs to the technical field of multi-source information fusion navigation, and in particular relates to a multi-source information fusion navigation method based on variational Bayesian factor graph. Background Art

[0002] Multi-source information fusion navigation is a complex process. First, it is necessary to build a mathematical model of navigation based on the characteristics of the navigation system itself and sensor noise, and transfer the real-time observation data to the corresponding state vector space through mapping. Then, the data obtained by different navigation modules are merged and processed to output accurate navigation status in real time.

[0003] Multi-source information fusion navigation is usually performed with Kalman filtering as the core, and a large number of improved algorithms have been developed to improve stability and accuracy. However, Kalman filtering requires the use of extrapolation and interpolation methods to ensure the timing consistency of measurements from different sensors, which introduces errors and additional computational costs. In addition, when the effectiveness of a sensor in the integrated navigation system changes, the state equation of the Kalman filter needs to be adjusted accordingly, which undoubtedly poses a challenge to the flexibility of the algorithm.

[0004] As a probabilistic graphical model method, factor graph can encode the probabilistic relationship between navigation state and system measurement, and form a framework graph by interconnecting nodes. Then, Bayesian reasoning is used to estimate the navigation state for these factors connected to the framework graph. The factor graph model has strong flexibility and can realize plug-and-play of sensors. However, since the carrier movement is disturbed by factors such as vibration and temperature, the statistical characteristics of the system noise are unknown or mutated, and the Gaussian noise assumption may make it difficult for traditional factor graph navigation algorithms to converge. With the increasing complexity of application scenarios of multi-source information fusion navigation, a more robust factor graph method is needed to deal with the problem of unknown or time-varying statistical characteristics of noise in multi-source information fusion navigation. Summary of the invention

[0005] 1. Technical issues to be resolved

[0006] The technical problem to be solved by the present invention is: in view of the deficiencies of the prior art, how to provide a multi-source information fusion navigation method based on a variational Bayesian factor graph to improve the robustness of the multi-source information fusion navigation.

[0007] (II) Technical solution

[0008] In order to solve the above technical problems, the present invention provides a multi-source information fusion navigation method based on variational Bayesian factor graph, the method comprising the following steps:

[0009] Step 1: construct a factor graph model using factor nodes of an inertial navigation system, a scene matching navigation system, and a global satellite navigation system;

[0010] Step 2: Use Student-T distribution to model the observation noise of factor nodes;

[0011] Step 3: Combine the Gauss-Newton iterative method and the variational Bayesian method to estimate the joint posterior distribution of the unknown noise parameters of the navigation measurement.

[0012] Wherein, in the step 1, the factor graph model is constructed using the factor nodes of the inertial navigation system (INS), the scene matching navigation system (IMNS), and the global satellite navigation system (GNSS), including the following steps:

[0013] Step 1-1: Set time t k-1 At time t k Attitude, velocity, position increment As a measurement Construction time t k INS factor node;

[0014] Step 1-2: Extract and match the real-time image captured by the carrier camera with the pre-set reference image, and convert the time t k The matching view center of and the azimuth angle Az in the camera coordinate system k and elevation angle El k As a measurement Construction time t k IMNS factor node;

[0015] Step 1-3: Solve the pseudo-range single point positioning and Doppler velocity measurement to obtain the time t k Carrier speed and location As a measurement Construction time t k GNSS factor node;

[0016] Steps 1-4: Integrate the factor nodes of INS, IMNS, and GNSS into the factor graph model.

[0017] Wherein, in step 2, the observation noise of the factor node is modeled using the Student-T distribution, including the following steps:

[0018] Step 2-1: At time t k Measurement noise U for INS, IMNS and GNSS k Perform a first-order Taylor expansion to obtain the Jacobian matrix J k and residual Rk :

[0019]

[0020] Among them, X k is the time series {t1,…,t k} in the navigation state collection, Yes X k An estimate of

[0021] Step 2-2, U k As a rule of thumb, the mean is 0 and the covariance is ∑ k , a random variable with student-t distribution and degrees of freedom ω: U k ~St(U k |0,∑ k ,v).

[0022] Wherein, in step 3, the joint posterior distribution of the navigation state and the unknown noise parameter is estimated by combining the Gauss-Newton iteration method with the variational Bayes method, including the following steps:

[0023] Step 3-1, convert the Student-T distribution into the following integral form:

[0024]

[0025] In the above formula, U follows a Gaussian distribution with mean μ and variance ∑, N(U|μ,∑) represents its probability density function, λ follows a gamma distribution with shape parameter a and scale parameter b, and Gamma(λ|a,b) represents its probability density function;

[0026] Step 3-2: Based on the previous test, give the unknown parameters ∑ k Prior information:

[0027] p(∑ k |c,C)=IW(∑ k |c,C)

[0028] In the above formula, ∑ follows the inverse Wishartf distribution with free parameter c and inverse scaling matrix C, and IW(∑|c,C) represents its probability density function;

[0029] Step 3-3: According to Bayesian theory, we get ΔX k ,∑ k and λ k Joint posterior probability density function:

[0030] p(ΔX k ,∑ k ,λ k |R k ,v,c,C)

[0031] Step 3-4: Use the variation function q(ΔX k ,∑ k ,λ k ) approximates p(ΔX k ,∑ k ,λ k |v,c,C):

[0032] q(ΔX k ,∑ k ,λ k )=q(ΔX k )q(∑ k )q(λ k )

[0033] Step 3-5: Use the expected maximum algorithm to update the approximate value q(ΔX k ),q(∑ k ),q(λ k );

[0034] Step 3-6: Calculate ΔX k ,∑ k and λ k The expected value E(ΔX k )、E(∑ k )、E(λ k );

[0035] Step 3-6: Gauss-Newton iterative method update

[0036] Step 3-7, repeat steps 3-5 and 3-6 until the free energy F(q(ΔX k ,R k ,λ k )) is less than a given threshold in two consecutive iterations.

[0037] (III) Beneficial effects

[0038] Compared with the prior art, the present invention provides a multi-source information fusion navigation method based on variational Bayesian factor graph, which can improve the robustness of multi-source information fusion navigation. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] Figure 1 The overall process of the multi-source information fusion navigation method based on variational Bayesian factor graph;

[0040] Figure 2 is the transformation relationship from the carrier coordinate system to the navigation coordinate system;

[0041] Figure 3 It is the structural diagram of the factor graph model. DETAILED DESCRIPTION

[0042] In order to make the purpose, content, and advantages of the present invention more clear, the specific implementation methods of the present invention are further described in detail below in conjunction with the accompanying drawings and examples.

[0043] In order to solve the above technical problems, the present invention provides a multi-source information fusion navigation method based on variational Bayesian factor graph, the method comprising the following steps:

[0044] Step 1: Use the factor nodes of Inertial Navigation System (INS), Image Matching Navigation System (IMNS), and Global Navigation Satellite System (GNSS) to construct a factor graph model;

[0045] Step 2: Use Student-T distribution to model the observation noise of factor nodes;

[0046] Step 3: Combine the Gauss-Newton iterative method and the variational Bayesian method to estimate the joint posterior distribution of the unknown noise parameters of the navigation measurement.

[0047] Wherein, in the step 1, the factor graph model is constructed using the factor nodes of the inertial navigation system (INS), the scene matching navigation system (IMNS), and the global satellite navigation system (GNSS), including the following steps:

[0048] Step 1-1: Set time t k-1 At time t k Attitude, velocity, position increment As a measurement Construction time t k INS factor node;

[0049] Step 1-2: Extract and match the real-time image captured by the carrier camera with the pre-set reference image, and convert the time t k The matching view center of and the azimuth angle Az in the camera coordinate system k and elevation angle El k As a measurement Construction time t k IMNS factor node;

[0050] Step 1-3: Solve the pseudo-range single point positioning and Doppler velocity measurement to obtain the time t k Carrier speed and location As a measurement Construction time t k GNSS factor node;

[0051] Steps 1-4: Integrate the factor nodes of INS, IMNS, and GNSS into the factor graph model.

[0052] Wherein, in step 2, the observation noise of the factor node is modeled using the Student-T distribution, including the following steps:

[0053] Step 2-1: At time t k Measurement noise U for INS, IMNS and GNSS k Perform a first-order Taylor expansion to obtain the Jacobian matrix J k and residual R k :

[0054]

[0055] Among them, X k is the time series {t1,…,t k} in the navigation state collection, Yes X k An estimate of

[0056] Step 2-2, U k As a rule of thumb, the mean is 0 and the covariance is ∑ k , a random variable with student-t distribution and degrees of freedom ω: U k ~St(U k |0,∑ k ,v).

[0057] Wherein, in step 3, the joint posterior distribution of the navigation state and the unknown noise parameter is estimated by combining the Gauss-Newton iteration method with the variational Bayes method, including the following steps:

[0058] Step 3-1, convert the Student-T distribution into the following integral form:

[0059]

[0060] In the above formula, U follows a Gaussian distribution with mean μ and variance ∑, N(U|μ,∑) represents its probability density function, λ follows a gamma distribution with shape parameter a and scale parameter b, and Gamma(λ|a,b) represents its probability density function;

[0061] Step 3-2: Based on the previous test, give the unknown parameters ∑ k Prior information:

[0062] p(∑ k|c,C)=IW(∑ k |c,C)

[0063] In the above formula, ∑ follows the inverse Wishartf distribution with free parameter c and inverse scaling matrix C, and IW(∑|c,C) represents its probability density function;

[0064] Step 3-3: According to Bayesian theory, we get ΔX k ,∑ k and λ k Joint posterior probability density function:

[0065] p(ΔX k ,∑ k ,λ k |R k ,v,c,C)

[0066] Step 3-4: Use the variation function q(ΔX k ,∑ k ,λ k ) approximates p(ΔX k ,∑ k ,λ k |v,c,C):

[0067] q(ΔX k ,∑ k ,λ k )=q(ΔX k )q(∑ k )q(λ k )

[0068] Step 3-5: Use the expected maximum algorithm to update the approximate value q(ΔX k ),q(∑ k ),q(λ k );

[0069] Step 3-6: Calculate ΔX k ,∑ k and λ k The expected value E(ΔX k )、E(∑ k )、E(λ k );

[0070] Step 3-6: Gauss-Newton iterative method update

[0071] Step 3-7, repeat steps 3-5 and 3-6 until the free energy F(q(ΔX k ,R k ,λ k )) is less than a given threshold in two consecutive iterations.

[0072] Example 1

[0073] The core content of this embodiment is a multi-source information fusion navigation method based on variational Bayesian factor graph, which includes the following three stages: (1) constructing a factor graph model using factor nodes of INS, IMNS, and GNSS; (2) using Student-T distribution to model the observation noise of factor nodes; (3) combining Gauss-Newton iteration method and variational Bayesian method to realize the joint posterior distribution estimation of unknown noise parameters of navigation measurement. The overall process is as follows: Figure 1 As shown:

[0074] (1) Constructing a factor graph model using factor nodes of INS, IMNS, and GNSS

[0075] Step 1: Set time t k-1 to k Inertial navigation attitude, velocity, and position increment at each moment As a measurement Construction time t k INS factor node:

[0076] a) First press Figure 2 As shown, the navigation coordinate system and the carrier coordinate conversion relationship are established, and the navigation coordinate system OX n Y n Z n It made three rotations, the first around OZ n Shaft positive rotation Get OX n(1) Y n(1) Z n , the second time around OX n(1) The shaft rotates forward θ to obtain OX n(1) Y n(2) Z n(1) , the third time around OY n(2) The shaft rotates forward γ to obtain OX b Y b Z b , the corresponding rotation matrix is:

[0077]

[0078]

[0079]

[0080] Then the navigation coordinate system OX n Y n Z n Go to carrier coordinate system OX b Y b Zb The rotation matrix is:

[0081]

[0082]

[0083] b) Solving equations by inertial navigation Get in t k The carrier posture of the moment speed Location information:

[0084]

[0085] In the formula, g n represents the gravity vector;

[0086] c) Establish and navigate state x k-1 、x k Connected INS factor nodes

[0087]

[0088] Step 2: Extract and match the real-time image captured by the carrier camera with the pre-set reference image. k The matching view center of and the azimuth angle Az in the camera coordinate system k and elevation angle El k As a measurement Construction time t k The specific steps are as follows:

[0089] a) Establish the carrier coordinate system OX b Y b Z b Transform to camera coordinate system OX c Y c Z c The rotation matrix is:

[0090]

[0091] Among them, λ and ρ are the horizontal sight angle and the vertical sight angle respectively;

[0092] b) Establish geographic coordinate system OX E Y E Z E Transform to navigation coordinate system OX n Y n Z n The rotation matrix is:

[0093]

[0094] c) Use the carrier camera to obtain the geographic location p k ={φ k ,λ k ,h k} regional scene as a real-time image, and use the ORB algorithm to align the real-time image with the pre-stored reference image;

[0095] d) Confirm the pixel coordinates of the center of the real-time image in the base image, and project these pixel coordinates into the geographic coordinate system to obtain the coordinates of the center of the matching view

[0096] e) Calculate the relative position of the matching positioning center and the carrier in the Earth-centered Earth-fixed coordinate system

[0097]

[0098] f)R N and e are the meridian curvature radius and eccentricity of the ellipsoidal earth model, respectively.

[0099] Step 6: Calculate the relative position vector between the center of the matching view and the carrier in the camera coordinate system:

[0100]

[0101] in,

[0102] g) Establish and k Connected IMNS Factor Nodes

[0103]

[0104] in, Indicates t k The sight angle Az at the moment k and elevation angle El k ,h SMV (x k ) is the measurement equation of IMNS:

[0105]

[0106] Step 3: Combine the carrier velocity obtained by Doppler velocity measurement and pseudo-range single point positioning and location As a measurement Build with x k Connected GNSS Factor Nodes

[0107]

[0108] Among them, h GNSS (x k ) is the measurement equation of GNSS.

[0109] Step 4: Integrate the factor nodes and factor nodes of INS, IMNS, and GNSS into the factor graph model. The specific steps are as follows: Figure 3 As shown:

[0110] a) At time t0, the initial parameter x is set by prior information 0 ;

[0111] b) Time is updated to t1, using the factor nodes of INS Connect x 0 and x 1 ;

[0112] c) Factor nodes using IMNS and GNSS and Connect x 1

[0113] d) As time continues to update, at time t k Forming the factor node set of INS, IMNS, and GNSS Navigation state collection X k ={x 0 ,…,x k} and the measurement set is

[0114] (2) Use Student-T distribution to model the observation noise of factor nodes

[0115] Step 1: At time t k Measurement noise U for INS, IMNS and GNSS k Perform a first-order Taylor expansion to obtain the Jacobian matrix J k and residual R k :

[0116]

[0117] Among them, X k ={x 0 ,…,x k} is the navigation state set before time k, Yes X k An estimate of

[0118] Step 2: Put U k As a rule of thumb, the mean is 0 and the covariance is ∑ k, a random variable of student-t distribution with degrees of freedom ω, satisfies the following formula:

[0119]

[0120] Where n represents U k Dimension.

[0121] (3) Combine the Gauss-Newton iterative method with the variational Bayesian method to estimate the joint posterior distribution of the navigation state and unknown noise parameters

[0122] Step 1: Since it is difficult to obtain a closed-form solution for the maximum likelihood estimation of the Student-T distribution, convert the Student-T distribution into the following integral form:

[0123] St(U k |0,∑ k ,ω)=∫N(U k |0,∑ k / λ k )Gamma(λ k |v / 2,v / 2)dλ k (16)

[0124] In the above formula, U obeys the Gaussian distribution with mean μ and variance ∑, N(U|μ,∑) represents its probability density function, λ obeys the gamma distribution with shape parameter a and scale parameter b, Gamma(λ|a,b) is its probability density function, when v=1, the Student-T distribution is transformed into the Cauchy distribution, when v→∞, the Student-T distribution converges to the Gaussian distribution;

[0125] Step 2: Based on previous experiments and expert experience, unknown parameters ∑ k Prior information:

[0126] p(∑ k |c,C)=IW(∑ k |c,C) (17)

[0127] In the above formula, ∑ follows the inverse Wishartf distribution with free parameter c and inverse scaling matrix C, and IW(∑|c,C) represents its probability density function;

[0128] Step 3: According to Bayesian theory, we get ΔX k ,∑ k and λ k Joint posterior probability density function:

[0129] p(ΔX k ,∑ k ,λ k |R k ,v,c,C)≈

[0130] p(R k |ΔX k ,∑ k ,λ k )IW(∑ k |c,C)Gamma(λ k |v / 2,v / 2) (18)

[0131] Among them, p(R k |ΔX k ,∑ k ,λ k ) is about ΔX k ,∑ k and λ k The likelihood function of :

[0132]

[0133] In the above formula, represents the squared Mahalanobis distance.

[0134] Step 4: Set the number of iterations N and use the variational function q in the i-th iteration (i) (ΔX k ,∑ k ,λ k ) approximates ΔX k ,∑ k and λ k The posterior distribution of :

[0135] a) Get the marginal likelihood term:

[0136] p(R k )=∫∫∫p(R k |ΔX k ,∑ k ,λ k )dΔX k d∑ k dλ k (20)

[0137] b) Logarithmized lnp(R k ) to decompose:

[0138] lnp(R k )=F(q (i) (Θ))+KLD[q (i) (Θ)||p(R k |ΔX k ,∑ k ,λ k )] (twenty one)

[0139] Where Θ = {ΔXk ,∑ k ,λ k},F(q (i) (Θ)) is the free energy, KLD[q (i) (Θ)||p(R k |ΔX k ,∑ k ,λ k )] is the variation function q (i) (Θ) and p(R k |ΔX k ,∑ k ,λ k ). When the Kullback-Leibler divergence reaches 0, q (i) (Θ) is equal to the true posterior probability density function p(R k |ΔX k ,∑ k ,λ k ), free energy F(q (i) (Θ)) reaches its maximum value.

[0140] c) Use the conjugate exponential model to represent q (i) (Θ), decompose it into the product of independent distribution of each parameter:

[0141] q (i) (Θ) = q (i) (ΔX k )q (i) (∑ k )q (i) (λ k ) (twenty two)

[0142] d) Get ΔX k ,∑ k , k Approximate posterior distribution q (i) (ΔX k ),q (i) (∑ k ),q (i) (λ k ):

[0143] q (i) (ΔX k )=N(ΔX k |E(ΔX k ),P k ) (twenty three)

[0144] Among them, P k =((J k ) T E(∑k ) / E(λ k )J k ) -1 ;

[0145]

[0146] in,

[0147]

[0148]

[0149] in,

[0150] Step 5: Calculate ΔX k ,∑ k and λ k The expected value E (i) (ΔX k )、E (i) (∑ k )、E (i) (λ k ):

[0151] E (i) (ΔX k )=P k J k E(∑ k ) / E(λ k )R k (26)

[0152]

[0153]

[0154] Step 6: Update X using Gauss-Newton iteration method k :

[0155]

[0156] Step 7: Use q (i) (ΔX k ),q (i) (∑ k ),q (i) (λ k ) Calculate the free energy F (i) (q(Θ)), calculate the relative difference in free energy between two consecutive iterations tol:

[0157] tol=abs[F (i) (q(Θ))-F (i-1)(q(Θ))] (30)

[0158] If tol is less than a given threshold of 0.01, then stop updating X k ; Otherwise i=i+1, go to step 4.

[0159] The above is only a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the technical principles of the present invention. These improvements and modifications should also be regarded as the scope of protection of the present invention.

Claims

1. A multi-source information fusion navigation method based on variational Bayesian factor graph, characterized in that: The method comprises the following steps: Step 1: construct a factor graph model using factor nodes of an inertial navigation system, a scene matching navigation system, and a global satellite navigation system; Step 2: Use Student-T distribution to model the observation noise of factor nodes; Step 3: Combine the Gauss-Newton iterative method and the variational Bayesian method to estimate the joint posterior distribution of the unknown noise parameters of the navigation measurement.

2. The multi-source information fusion navigation method based on variational Bayesian factor graph according to claim 1, characterized in that: In the step 1, the factor graph model is constructed using the factor nodes of the inertial navigation system (INS), the scene matching navigation system (IMNS), and the global satellite navigation system (GNSS), including the following steps: Step 1-1: Set time t k-1 At time t k Attitude, velocity, position increment As a measurement Construction time t k INS factor node; Step 1-2: Extract and match the real-time image captured by the carrier camera with the pre-set reference image, and convert the time t k The matching view center of and the azimuth angle Az in the camera coordinate system k and elevation angle El k As a measurement Construction time t k IMNS factor node; Step 1-3: Solve the pseudo-range single point positioning and Doppler velocity measurement to obtain the time t k Carrier speed and location As a measurement Construction time t k GNSS factor node; Steps 1-4: Integrate the factor nodes of INS, IMNS, and GNSS into the factor graph model.

3. The multi-source information fusion navigation method based on variational Bayesian factor graph according to claim 2, characterized in that: In step 2, the observation noise of the factor node is modeled using the Student-T distribution, including the following steps: Step 2-1: At time t k Measurement noise U for INS, IMNS and GNSS k Perform a first-order Taylor expansion to obtain the Jacobian matrix J k and residual R k : Among them, X k is the time series {t1,…,t k } in the navigation state collection, Yes X k An estimate of Step 2-2, U k As a rule of thumb, the mean is 0 and the covariance is ∑ k , a random variable with student-t distribution and degrees of freedom ω: U k ~St(U k |0,∑ k ,v).

4. The multi-source information fusion navigation method based on variational Bayesian factor graph according to claim 3, characterized in that: In step 3, the joint posterior distribution of the navigation state and the unknown noise parameter is estimated by combining the Gauss-Newton iteration method with the variational Bayes method, including the following steps: Step 3-1, convert the Student-T distribution into the following integral form: St(U k |0,S k ,v)=∫N(U k |0,∑ k / l k )Gamma(λ k |v / 2,v / 2)dλ k In the above formula, U follows a Gaussian distribution with mean μ and variance Σ, N(U|μ,Σ represents its probability density function, λ follows a gamma distribution with shape parameter a and scale parameter b, and Gamma(λ|a,b) represents its probability density function; Step 3-2: Based on the previous test, give the unknown parameters ∑ k Prior information: p(∑ k |c,C)=IW(∑ k |c,C) In the above formula, ∑ follows the inverse Wishartf distribution with free parameter c and inverse scaling matrix C, and IM(∑|c,C represents its probability density function; Step 3-3: According to Bayesian theory, we get ΔX k ,∑ K and λ k Joint posterior probability density function: p(ΔX k ,∑ k ,λ k |R k ,υ,c,C) Step 3-4: Use the variation function q(ΔX k ,∑ k ,λ k ) approximates p(ΔX k ,∑ k ,λ k |υ,c,C): q(ΔX k ,∑ k ,λ k )=q(ΔX k )q(∑ k )q(λ k ) Step 3-5: Use the expected maximum algorithm to update the approximate value q(ΔX k ),q(∑ k ),q(λ k ); Step 3-6: Calculate ΔX k ,∑ k and λ k The expected value E(ΔX k )、E(∑ k )、E(λ k ); Step 3-6: Update X using Gauss-Newton iteration method k : Step 3-7, repeat steps 3-5 and 3-6 until the free energy F(q(ΔX k ,R k ,λ k )) is less than a given threshold in two consecutive iterations.

5. The multi-source information fusion navigation method based on variational Bayesian factor graph according to claim 4, characterized in that: The method belongs to the technical field of multi-source information fusion navigation.

6. The multi-source information fusion navigation method based on variational Bayesian factor graph according to claim 4, characterized in that: The inertial navigation system is Inertial Navigation System, INS.

7. The multi-source information fusion navigation method based on variational Bayesian factor graph according to claim 4, characterized in that: The scene matching navigation system is Image Matching Navigation System, IMNS.

8. The multi-source information fusion navigation method based on variational Bayesian factor graph according to claim 4, characterized in that: The global satellite navigation system is Global Navigation Satellite System, GNSS.

9. The multi-source information fusion navigation method based on variational Bayesian factor graph according to claim 4, characterized in that: In step 2-1, X k is the time series {t1,…,t k } within the navigation state collection.

10. The multi-source information fusion navigation method based on variational Bayesian factor graph according to claim 4, characterized in that: In the step 2-1, Yes X k The estimated amount.

Images