BDSBAS / MSAS fusion positioning method based on full-connection neural network

Through the BDSBAS/MSAS fusion positioning method based on fully connected neural networks, combined with Bayesian regularization algorithm, the problem of seamless connection and limited positioning accuracy in multi-SBAS fusion is solved, and more efficient positioning accuracy and coverage area are achieved.

CN118915110APending Publication Date: 2024-11-08SHENYANG AEROSPACE UNIVERSITY
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202410964745.3
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-07-18
Publication Date
2024-11-08

AI Technical Summary

Technical Problem

The existing multi-SBAS fusion algorithm lacks flexibility and adaptability, making it difficult to achieve seamless connection and fusion in the overlapping areas of BDSBAS and MSAS services, resulting in limited positioning accuracy and coverage areas.

Method used

A BDSBAS/MSAS fusion positioning method based on fully connected neural networks is adopted, and a Bayesian regularization algorithm is introduced to prevent model overfitting. The position coordinates are fusion by constructing a three-layer fully connected neural network model to optimize the positioning effect.

Benefits of technology

The accuracy and stability of BDSBAS and MSAS positioning is improved, the limitations of a separate system are overcome, and better positioning effect and coverage area are achieved.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118915110A_ABST
    Figure CN118915110A_ABST
Patent Text Reader

Abstract

The invention provides a BDSBAS / MSAS fusion positioning method based on a full-connection neural network, and relates to the technical field of satellite-based enhancement system.The method comprises the steps that delay correction information in BDSBAS and MSAS enhancement messages is solved, error correction is conducted on an original measurement pseudo range, and position coordinates corrected through BDSBAS and MSAS are obtained through calculation; fusing the position coordinates of the BDSBAS and the MSAS by using the obtained position coordinates corrected by the BDSBAS and the MSAS through constructing a full-connection neural network model optimized based on a Bayesian regularization algorithm, and outputting the fused position coordinates; and then substituting the BDSBAS position coordinate data, the MSAS position coordinate data and the fused position data into a root mean square error algorithm model, judging the advantages and disadvantages of the fusion positioning effect through comparison, and storing the fused position data.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the field of satellite-based augmentation system positioning technology, and in particular to a BDSBAS / MSAS fusion positioning method based on a fully connected neural network. Background Art

[0002] With the continuous development and popularization of the global satellite navigation system (GNSS), satellite navigation augmentation technology has gradually attracted widespread attention as an important means to improve the quality of navigation and positioning services. In order to meet the accuracy and integrity requirements of high-life safety users such as aviation, the satellite-based augmentation system (SBAS) with GNSS as the augmentation object came into being. However, the SBAS augmentation service is affected by the distribution and accuracy of the ionospheric grid. A single SBAS can only provide services within a limited range, and its service level and coverage area are limited. Multi-SBAS fusion technology is of great significance for achieving seamless integration of each SBAS in the overlapping SBAS service area, and it is also a key technical challenge.

[0003] At present, multi-SBAS fusion algorithms are mainly focused on fusion methods based on arithmetic mean and weighted mean, and there are few related studies. The arithmetic mean simply adds all the values ​​and divides them by the number, while the weighted mean performs weighted summation according to the preset weights. Both assume that there is a linear relationship between variables, lacking flexibility and adaptability. Machine learning-based models have been experiencing vigorous development in recent years, and neural networks have shown great potential in the field of positioning. Compared with traditional multi-SBAS fusion algorithms, the fusion method based on neural networks has stronger generalization ability and adaptability, and can learn feature information from a large amount of data, thereby improving the accuracy and stability of positioning. The present invention constructs a BDSBAS and MSAS fusion positioning system based on a fully connected neural network model, and develops a scheme for determining the improvement effect of multi-SBAS positioning to meet the diversified needs of users and realize the navigation performance challenges caused by SBAS correction in the overlapping area of ​​BDSBAS and MSAS services. Summary of the invention

[0004] In view of the shortcomings of the prior art, the present invention provides a BDSBAS / MSAS fusion positioning method based on a fully connected neural network, which improves the positioning effects of BDSBAS and MSAS, and introduces a Bayesian regularization algorithm to prevent overfitting of the fully connected neural network model, thereby improving the generalization ability of the model.

[0005] A BDSBAS / MSAS fusion positioning method based on a fully connected neural network comprises the following steps:

[0006] Step 1: By solving the delay correction information in the BDSBAS and MSAS enhanced messages, the error of the original measured pseudorange is corrected, and the position coordinates (x BDSBAS ,y BDSBAS ,z BDSBAS ) and (x MSAS ,y MSAS ,z MSAS );

[0007] By solving the delay correction information in the SBAS enhancement message, the observed pseudorange is obtained, that is, the pseudorange value ρ of the satellite after the enhancement information correction is obtained. correected :

[0008] ρ correected =ρ measured +PRC(t of )+PRC(t of )×(tt of )+d iono +d tropo +d resive +d mul

[0009] Where ρ measured is the geometric distance from the receiver antenna to the satellite, PRC(t of ) is the satellite fast-changing correction number, t is the current observation time of the receiver, t of is the satellite clock correction number, d iono d tropo d resive d mul are the ionosphere, troposphere, receiver and multipath delay corrections respectively, and t is the current observation time of the receiver;

[0010] Assuming that there are i visible stars providing pseudorange observations at the current moment, the position coordinates (x, y, z) corrected by SBAS are obtained by the pseudorange positioning principle:

[0011]

[0012] Among them, (x i ,y i ,z i ) is the position coordinate of the i-th satellite, δt u is the receiver clock error, ρ corrected,i is the pseudorange value of the ith satellite.

[0013] The position coordinates (x BDSBAS ,y BDSBAS ,zBDSBAS ) and (x MSAS ,y MSAS ,z MSAS );wherein the BDSBAS and MSAS service overlapping area should be selected for the original measurement pseudorange measurement reference point, which is used to limit the reception of BDSBAS and MSAS enhanced messages at the reference point for position coordinate solution.

[0014] Step 2: Use the position coordinates (x BDSBAS ,y BDSBAS ,z BDSBAS )、(x MSAS ,y MSAS ,z MSAS ), by constructing a fully connected neural network model optimized based on the Bayesian regularization algorithm to fuse the BDSBAS and MSAS position coordinates, the fused position coordinates (x Fusion ,y Fusion ,z Fusion );

[0015] Step 2.1: Build a fully connected neural network model;

[0016] The fully connected neural network model has three layers, namely, input layer, hidden layer and output layer; the input layer has 6 neurons, the number of neurons matches the dimension of the input data, and the hidden layer and output layer have 3 neurons respectively;

[0017] In the fully connected neural network model, for each neuron, its input is the output of all neurons in the previous layer. The fully connected neural network neuron consists of an input part x, a weight w, a bias b, an activation function h(·) and an output part a.

[0018] The output of each neuron is expressed as:

[0019]

[0020] in, is the output of the jth neuron in the lth layer; N (l-1) is the number of neurons in the l-1th layer, and L is the total number of layers; is the weight connecting the i-th neuron in the l-1th layer to the j-th neuron in the lth layer; is the output of the i-th neuron in the l-1th layer; is the bias of the jth neuron in the lth layer; h(·) is the activation function.

[0021] The fully connected neural network model performs prediction or classification tasks by calculating the output of each neuron and passing it from the input layer to the output layer.

[0022] Step 2.2: Introduce the Bayesian regularization algorithm to optimize the fully connected neural network model.

[0023] In the Bayesian regularization algorithm, the regularization term is the sum of the squares of the weights, and its formula is:

[0024]

[0025] Among them, λ represents the regularization parameter; N (l) is the number of neurons in layer l; is the weight connecting the i-th neuron to the j-th neuron in the l-1th layer.

[0026] The loss function combined with the Bayesian regularization algorithm is expressed as:

[0027]

[0028] Where m is the number of training samples, y i is the true value of the i-th sample, is the predicted value of the ith sample.

[0029] During the back-propagation process, the error of the output layer is first calculated:

[0030] Z (L) =A (L) -Y

[0031] Among them, A (L) is the predicted value of the output layer, and Y is the true value;

[0032] Next, based on the error of the output layer, the gradient of the weights and biases of each layer is calculated; for the output layer, the weight gradient dw of the Lth layer is (L) and bias gradient db (L) The calculation formula is:

[0033]

[0034] Where m is the number of training samples, A (L-1) is the activation value of the previous layer, A (L-1)T is the transpose of the activation value of the previous layer, w (L) is the weight of the Lth layer, b (L) is the bias of the Lth layer.

[0035] Then, backpropagation transfers the error to the hidden layer; for the lth layer, the formula for error propagation is:

[0036] Z (l) =(w (l+1) ) T ·dZ (l+1) ·h′(Z(l) )

[0037] Among them, (w (l+1) ) T is the transpose of the weights of the l+1th layer, dZ (l+1) is the error of the l+1th layer, h′(Z (l) ) is the derivative of the activation function of the lth layer. Through this formula, the error is passed back from the output layer layer by layer, affecting the gradient calculation of the previous layer;

[0038] For each hidden layer, the weight gradient dw of the lth layer (l) and bias gradient db (l) The calculation formula is:

[0039]

[0040] Finally, the gradient descent method is used to update the parameters; the formula for parameter update is:

[0041]

[0042] Among them, α is the learning rate, which determines the step size of each update; w (l) and b (l) are the weights and biases of the lth layer respectively; by adjusting the updated weights and bias Substituting into the traditional fully connected neural network neuron output formula, the output of each neuron is:

[0043]

[0044] in, is the output of the jth neuron in the lth layer after Bayesian regularization training; is the weight connected to the i-th neuron in the l-1th layer to the j-th neuron in the lth layer after Bayesian regularization training; is the output of the i-th neuron in the l-1th layer after Bayesian regularization training; is the bias of the jth neuron in the lth layer after Bayesian regularization training;

[0045] Step 2.3: Convert the position coordinates (x BDSBAS ,y BDSBAS ,z BDSBAS )、(x MSAS ,y MSAS ,z MSAS ), as input, trained and tested through the fully connected neural network model, and obtained the fused position coordinates (x Fusion ,y Fusion ,z Fusion );

[0046] When using the fully connected neural network model for fusion, set the learning rate to 0.01, the number of iterations to 1000, and the maximum number of failures to 10. Take the set proportion of the corrected position coordinate data obtained in step 1 as the training set to train the fully connected neural network model and obtain the BDSBAS and MSAS fusion positioning model based on the fully connected neural network.

[0047] After obtaining the trained BDSBAS and MSAS fusion positioning model based on the fully connected neural network, the remaining BDSBAS position coordinate data (x′ BDSBAS ,y′ BDSBAS ,z′ BDSBAS ) and MSAS position coordinate data (x′ MSAS ,y′ MSAS ,z′ MSAS ), we can get the fused position data (x Fusion ,y Fusion ,z Fusion ).

[0048] Step 3: Calculate the remaining BDSBAS position coordinate data (x′ BDSBAS ,y′ BDSBAS ,z′ BDSBAS ), MSAS position coordinate data (x′ MSAS ,y′ MSAS ,z′ MSAS ), and the fused position data (x Fusion ,y Fusion ,z Fusion ) relative to the true position (x true ,y true ,z true ) and compare them to determine the quality of the fusion positioning effect and save the fusion position data;

[0049] Step 3.1: Use the root mean square error as the evaluation method for comparison; let (x, y, z) be the coordinate data obtained after correction or fusion of BDSBAS and MSAS, (x true ,y true ,z true ) is the real position value obtained after processing; (x, y, z) and (x true ,y true ,z true ) to obtain the difference (Δx, Δy, Δz):

[0050]

[0051] Then the root mean square error (RMS) of the (Δx, Δy, Δz) parameters is Δx 、RMS Δy 、RMS Δz ) is calculated as follows:

[0052]

[0053] Where N is the number of measurement epochs;

[0054] The total positioning root mean square error RMS is:

[0055]

[0056] Step 3.2: Substitute the BDSBAS position coordinate data, MSAS position coordinate data and fused position data into the total positioning root mean square error RMS calculation formula, and judge the quality of the fusion positioning effect by comparison;

[0057] Except for the training set, the remaining BDSBAS position coordinate data (x′ BDSBAS ,y′ BDSBAS ,z′ BDSBAS ), MSAS position coordinate data (x′ MSAS ,y′ MSAS ,z′ MSAS ) and the fused position data (x Fusion ,y Fusion ,z Fusion ) are substituted into the total positioning root mean square error RMS calculation formula to obtain RMS BDSBAS 、RMS MSAS 、RMS Fusion ;

[0058] If RMS Fusion <RMS BDSBAS And RMS Fusion <RMS MSAS , it means that the positioning effect after fusion of the fully connected neural network is better than that of BDSBAS and MSAS, and the fused position coordinate data (x Fusion ,y Fusion ,z Fusion ), otherwise continue to return to step 2.3 to train the fully connected neural network model.

[0059] The beneficial effects of adopting the above technical solution are:

[0060] The present invention provides a BDSBAS / MSAS fusion positioning method based on a fully connected neural network, which aims to study the BDSBAS and MSAS fusion positioning algorithm, solve the challenge of limited performance of a single SBAS service and lack of seamless connection between the overlapping areas of BDSBAS and MSAS. Through the application of a fully connected neural network model, the present invention can effectively improve positioning accuracy and overcome the limitations of each system when used alone. BRIEF DESCRIPTION OF THE DRAWINGS

[0061] Figure 1 The overall design flow chart provided for the embodiment of the present invention;

[0062] Figure 2 A flow chart of BDSBAS and MSAS enhanced message data processing provided by an embodiment of the present invention;

[0063] Figure 3 A flowchart for designing a fully connected neural network based on Bayesian regularization is provided for an embodiment of the present invention. DETAILED DESCRIPTION

[0064] The specific implementation of the present invention is further described in detail below in conjunction with the accompanying drawings and examples. The following examples are used to illustrate the present invention, but are not intended to limit the scope of the present invention.

[0065] A BDSBAS / MSAS fusion positioning method based on a fully connected neural network, such as Figure 1 As shown, the following steps are included:

[0066] Step 1: By solving the delay correction information in the BDSBAS and MSAS enhanced messages, the error of the original measured pseudorange is corrected, and the position coordinates (x BDSBAS ,y BDSBAS ,z BDSBAS ) and (x MSAS ,y MSAS ,z MSAS );

[0067] SBAS continuously tracks and observes navigation satellites through ground monitoring stations, and transmits the observation data to the corresponding master control station for processing. After processing, the master control station generates differential corrections and integrity parameters, and compiles them into enhanced messages, which are broadcast to users within the service range through geosynchronous orbit satellites. Users use the differential corrections in the enhanced messages to improve positioning accuracy, and use integrity parameters to ensure the integrity of service performance. The data processing flow of BDSBAS and MSAS enhanced messages is as follows: Figure 2 shown.

[0068] By solving the delay correction information in the SBAS enhancement message, the observed pseudorange is obtained, that is, the pseudorange value ρ of the satellite after the enhancement information correction is obtained. correected :

[0069] ρ correected =ρ measured +PRC(t of )+PRC(t of )×(tt of )+d iono +d tropo +d resive +d mul

[0070] Where ρ measured is the geometric distance from the receiver antenna to the satellite, PRC(t of ) is the satellite fast-changing correction number, t is the current observation time of the receiver, t of is the satellite clock correction number, d iono d tropo d resive d mul are the ionosphere, troposphere, receiver and multipath delay corrections respectively, and t is the current observation time of the receiver;

[0071] Assuming that there are i visible stars providing pseudorange observations at the current moment, the position coordinates (x, y, z) corrected by SBAS are obtained by the pseudorange positioning principle:

[0072]

[0073] Among them, (x i ,y i ,z i ) is the position coordinate of the i-th satellite, δt u is the receiver clock error, ρ corrected,i is the pseudorange value of the ith satellite.

[0074] The position coordinates (x BDSBAS ,y BDSBAS ,z BDSBAS ) and (x MSAS ,y MSAS ,z MSAS );wherein the BDSBAS and MSAS service overlapping area should be selected for the original measurement pseudorange measurement reference point, which is used to limit the reception of BDSBAS and MSAS enhanced messages at the reference point for position coordinate solution.

[0075] Step 2: Use the position coordinates (xBDSBAS ,y BDSBAS ,z BDSBAS )、(x MSAS ,y MSAS ,z MSAS ), by constructing a fully connected neural network model optimized based on the Bayesian regularization algorithm to fuse the BDSBAS and MSAS position coordinates, the fused position coordinates (x Fusion ,y Fusion ,z Fusion );

[0076] Step 2.1: Build a fully connected neural network model.

[0077] The structure diagram of the fully connected neural network model in this embodiment is as follows Figure 3 The fully connected neural network model has three layers, namely, input layer, hidden layer and output layer; the input layer has 6 neurons, the number of neurons matches the dimension of the input data, and the hidden layer and output layer have 3 neurons respectively;

[0078] In the fully connected neural network model, for each neuron, its input is the output of all neurons in the previous layer. The fully connected neural network neuron consists of an input part x, a weight w, a bias b, an activation function h(·) and an output part a.

[0079] The functions of each part in the network model are as follows:

[0080] x: Input part, which is the output of n neurons in the previous layer.

[0081] w: weight, indicating the importance of each input. In a fully connected layer, each neuron has a set of weights connected to all neurons in the previous layer.

[0082] b: Bias, which represents the connection bias term, used to adjust the ease with which neurons are activated.

[0083] h(·): activation function, which is a nonlinear function that maps the weighted summation result to a nonlinear output. Common activation functions include ReLU, Sigmoid, Tanh, etc. In this embodiment, Sigmoid is selected as the activation function.

[0084] a: The output of the neuron, which is the weighted sum of the activation function.

[0085] The output of each neuron is expressed as:

[0086]

[0087] in, is the output of the jth neuron in the lth layer; N(l-1) is the number of neurons in the l-1th layer, and L is the total number of layers; is the weight connecting the i-th neuron in the l-1th layer to the j-th neuron in the lth layer; is the output of the i-th neuron in the l-1th layer; is the bias of the jth neuron in the lth layer; h(·) is the activation function.

[0088] The fully connected neural network model performs prediction or classification tasks by calculating the output of each neuron and passing it from the input layer to the output layer.

[0089] Step 2.2: Introduce the Bayesian regularization algorithm to optimize the fully connected neural network model.

[0090] In order to prevent the model from overfitting, this embodiment introduces a Bayesian regularization training algorithm. Under the Bayesian regularization training algorithm, the training process of the fully connected neural network optimizes the model through steps such as forward propagation, loss calculation, back propagation and parameter update.

[0091] In the Bayesian regularization algorithm, the regularization term is the sum of the squares of the weights, and its formula is:

[0092]

[0093] Among them, λ represents the regularization parameter; N (l) is the number of neurons in layer l; is the weight connecting the i-th neuron to the j-th neuron in the l-1th layer.

[0094] The loss function combined with the Bayesian regularization algorithm is expressed as:

[0095]

[0096] Where m is the number of training samples, y i is the true value of the i-th sample, is the predicted value of the ith sample.

[0097] During the back-propagation process, the error of the output layer is first calculated:

[0098] Z (L) =A (L) -Y

[0099] Among them, A (L) is the predicted value of the output layer, and Y is the true value;

[0100] Next, based on the error of the output layer, the gradient of the weights and biases of each layer is calculated; for the output layer, the weight gradient dw of the Lth layer is (L) and bias gradient db(L) The calculation formula is:

[0101]

[0102] Where m is the number of training samples, A (L-1) is the activation value of the previous layer, A (L-1)T is the transpose of the activation value of the previous layer, w (L) is the weight of the Lth layer, b (L) is the bias of the Lth layer.

[0103] Then, backpropagation transfers the error to the hidden layer; for the lth layer, the formula for error propagation is:

[0104] Z (l) =(w (l+1) ) T ·dZ (l+1) ·h′(Z (l) )

[0105] Among them, (w (l+1) ) T is the transpose of the weights of the l+1th layer, dZ (l+1) is the error of the l+1th layer, h′(Z (l) ) is the derivative of the activation function of the lth layer. Through this formula, the error is passed back from the output layer layer by layer, affecting the gradient calculation of the previous layer;

[0106] For each hidden layer, the weight gradient dw of the lth layer (l) and bias gradient db (l) The calculation formula is:

[0107]

[0108] Finally, the gradient descent method is used to update the parameters; the formula for parameter update is:

[0109]

[0110] Among them, α is the learning rate, which determines the step size of each update; w (l) and b (l) are the weights and biases of the lth layer respectively; by adjusting the updated weights and bias Substituting into the traditional fully connected neural network neuron output formula, the output of each neuron is:

[0111]

[0112] in, is the output of the jth neuron in the lth layer after Bayesian regularization training; is the weight connected to the i-th neuron in the l-1th layer to the j-th neuron in the lth layer after Bayesian regularization training; is the output of the i-th neuron in the l-1th layer after Bayesian regularization training; is the bias of the jth neuron in the lth layer after Bayesian regularization training;

[0113] Step 2.3: Convert the position coordinates (x BDSBAS ,y BDSBAS ,z BDSBAS )、(x MSAS ,y MSAS ,z MSAS ), as input, trained and tested through the fully connected neural network model, and obtained the fused position coordinates (x Fusion ,y Fusion ,z Fusion );

[0114] When using the fully connected neural network model for fusion, the learning rate is set to 0.01, the number of iterations is 1000, the maximum number of failures is set to 10 times, and the corrected position coordinate data obtained in step 1 is taken as a set ratio as a training set to train the fully connected neural network model to obtain a BDSBAS and MSAS fusion positioning model based on the fully connected neural network; in this embodiment, the corrected position coordinate data obtained in step 1 is divided into 50% and 50%, half of which is used for training and the other half is used for subsequent fusion;

[0115] After obtaining the trained BDSBAS and MSAS fusion positioning model based on the fully connected neural network, the remaining 50% of the BDSBAS position coordinate data (x′ BDSBAS ,y′ BDSBAS ,z′ BDSBAS ) and MSAS position coordinate data (x′ MSAS ,y′ MSAS ,z′ MSAS ), we can get the fused position data (x Fusion ,y Fusion ,z Fusion ).

[0116] Step 3: Calculate the BDSBAS position coordinate data (x′) of the remaining 50% of the training set BDSBAS ,y′ BDSBAS ,z′ BDSBAS ), MSAS position coordinate data (x′ MSAS ,y′ MSAS ,z′ MSAS ), and the fused position data (x Fusion ,y Fusion,z Fusion ) relative to the true position (x true ,y true ,z true ) and compare them to determine the quality of the fusion positioning effect and save the fusion position data;

[0117] Step 3.1: Use the root mean square error as the evaluation method for comparison; let (x, y, z) be the coordinate data obtained after correction or fusion of BDSBAS and MSAS, (x true ,y true ,z true ) is the actual position value obtained after processing; (x, y, z) and (x true ,y true ,z true ) to obtain the difference (Δx, Δy, Δz):

[0118]

[0119] Then the root mean square error (RMS) of the (Δx, Δy, Δz) parameters is Δx 、RMS Δy 、RMS Δz ) is calculated as follows:

[0120]

[0121] Where N is the number of measurement epochs;

[0122] The total positioning root mean square error RMS is:

[0123]

[0124] Step 3.2: Substitute the BDSBAS position coordinate data, MSAS position coordinate data and fused position data into the total positioning root mean square error RMS calculation formula, and judge the quality of the fusion positioning effect by comparison;

[0125] Except for the remaining 50% of the training set, the BDSBAS position coordinate data (x′ BDSBAS ,y′ BDSBAS ,z′ BDSBAS ), MSAS position coordinate data (x′ MSAS ,y′ MSAS ,z′ MSAS ) and the fused position data (x Fusion ,y Fusion ,z Fusion ) are substituted into the total positioning root mean square error RMS calculation formula to obtain RMS BDSBAS 、RMS MSAS 、RMS Fusion ;

[0126] If RMS Fusion <RMS BDSBAS And RMS Fusion <RMS MSAS , it means that the positioning effect after fusion of the fully connected neural network is better than that of BDSBAS and MSAS, and the fused position coordinate data (x Fusion ,y Fusion ,z Fusion ), otherwise continue to return to step 2.3 to train the fully connected neural network model.

[0127] The above description is only a preferred embodiment of the present disclosure and an explanation of the technical principles used. Those skilled in the art should understand that the scope of the invention involved in the embodiments of the present disclosure is not limited to the technical solutions formed by a specific combination of the above-mentioned technical features, but should also cover other technical solutions formed by any combination of the above-mentioned technical features or their equivalent features without departing from the above-mentioned inventive concept. For example, the above-mentioned features are replaced with the technical features with similar functions disclosed in the embodiments of the present disclosure (but not limited to) to form a technical solution.

Claims

1. A BDSBAS / MSAS fusion positioning method based on a fully connected neural network, characterized in that: The following steps are involved: Step 1: By solving the delay correction information in the BDSBAS and MSAS enhanced messages, the error of the original measured pseudorange is corrected, and the position coordinates (x BDSBAS ,y BDSBAS ,z BDSBAS ) and (x MSAS ,y MSAS ,z MSAS ); Step 2: Use the position coordinates (x BDSBAS ,y BDSBAS ,z BDSBAS )、(x MSAS ,y MSAS ,z MSAS ), by constructing a fully connected neural network model optimized based on the Bayesian regularization algorithm to fuse the BDSBAS and MSAS position coordinates, the fused position coordinates (x Fusion ,y Fusion ,z Fusion ); Step 3: Calculate the remaining BDSBAS position coordinate data (x′ BDSBAS ,y′ BDSBAS ,z′ BDSBAS ), MSAS position coordinate data (x′ MSAS ,y′ MSAS ,z′ MSAS ), and the fused position data (x Fusion ,y Fusion ,z Fusion ) relative to the true position (x true ,y true ,z true ) is compared to determine the quality of the fusion positioning effect and save the fusion position data.

2. A BDSBAS / MSAS fusion positioning method based on a fully connected neural network according to claim 1, characterized in that: The step 1 is specifically as follows: by solving the delay correction information in the SBAS enhancement message, the observed pseudorange is obtained, that is, the pseudorange value ρ of the satellite after the enhancement information correction is obtained. correected : ρ correected =ρ measured +PRC(t of )+PRC(t of )×(tt of )+d iono +d tropo +d resive +d mul Where ρ measured is the geometric distance from the receiver antenna to the satellite, PRC(t of ) is the satellite fast-changing correction number, t is the current observation time of the receiver, t of is the satellite clock correction number, d iono d tropo d resive d mul are the ionosphere, troposphere, receiver and multipath delay corrections respectively, and t is the current observation time of the receiver; Assuming that there are i visible stars providing pseudorange observations at the current moment, the position coordinates (x, y, z) corrected by SBAS are obtained by the pseudorange positioning principle: Among them, (x i ,y i ,z i ) is the position coordinate of the i-th satellite, δt u is the receiver clock error, ρ corrected,i is the pseudorange value of the i-th satellite; The position coordinates (x BDSBAS ,y BDSBAS ,z BDSBAS ) and (x MSAS ,y MSAS ,z MSAS );wherein the BDSBAS and MSAS service overlapping area should be selected for the original measurement pseudorange measurement reference point, so as to limit the receiving of both BDSBAS and MSAS enhanced messages at the reference point for position coordinate solution.

3. The BDSBAS / MSAS fusion positioning method based on a fully connected neural network according to claim 1, characterized in that: The step 2 comprises the following steps: Step 2.1: Build a fully connected neural network model; The fully connected neural network model has three layers, namely, input layer, hidden layer and output layer; the input layer has 6 neurons, the number of neurons matches the dimension of the input data, and the hidden layer and output layer have 3 neurons respectively; In the fully connected neural network model, for each neuron, its input is the output of all neurons in the previous layer; the fully connected neural network neuron is composed of an input part x, a weight w, a bias b, an activation function h(·) and an output part a; The output of each neuron is expressed as: in, is the output of the jth neuron in the lth layer; N (l-1) is the number of neurons in the l-1th layer, and L is the total number of layers; is the weight connecting the i-th neuron in the l-1th layer to the j-th neuron in the lth layer; is the output of the i-th neuron in the l-1th layer; is the bias of the jth neuron in the lth layer; h(·) is the activation function; The fully connected neural network model performs prediction or classification tasks by calculating the output of each neuron and passing it from the input layer to the output layer; Step 2.2: Introduce the Bayesian regularization algorithm to optimize the fully connected neural network model; In the Bayesian regularization algorithm, the regularization term is the sum of the squares of the weights, and its formula is: Among them, λ represents the regularization parameter; N (l) is the number of neurons in layer l; is the weight connecting the i-th neuron to the j-th neuron in the l-1th layer; The loss function combined with the Bayesian regularization algorithm is expressed as: Where m is the number of training samples, y i is the true value of the i-th sample, is the predicted value of the i-th sample; During the back-propagation process, the error of the output layer is first calculated: dZ (L) =A (L) -AND Among them, A (L) is the predicted value of the output layer, and Y is the true value; Next, based on the error of the output layer, the gradient of the weights and biases of each layer is calculated; for the output layer, the weight gradient dw of the Lth layer is (L) and bias gradient db (L) The calculation formula is: Where m is the number of training samples, A (L-1) is the activation value of the previous layer, A (L-1)T is the transpose of the activation value of the previous layer, w (L) is the weight of the Lth layer, b (L) is the bias of the Lth layer; Then, backpropagation transfers the error to the hidden layer; for the lth layer, the formula for error propagation is: dZ (l) =(in (l+1) ) T ·dZ (l+1) h′(Z (l) ) Among them, (w (l+1) ) T is the transpose of the weights of the l+1th layer, dZ (l+1) is the error of the l+1th layer, h′(Z (l) ) is the derivative of the activation function of the lth layer; through this formula, the error is passed back from the output layer layer by layer, affecting the gradient calculation of the previous layer; For each hidden layer, the weight gradient dw of the lth layer (l) and bias gradient db (l) The calculation formula is: Finally, the gradient descent method is used to update the parameters; the formula for parameter update is: Among them, α is the learning rate, which determines the step size of each update; w (l) and b (l) are the weights and biases of the lth layer respectively; by adjusting the updated weights and bias Substituting into the traditional fully connected neural network neuron output formula, the output of each neuron is: in, is the output of the jth neuron in the lth layer after Bayesian regularization training; is the weight connected to the i-th neuron in the l-1th layer to the j-th neuron in the lth layer after Bayesian regularization training; is the output of the i-th neuron in the l-1th layer after Bayesian regularization training; is the bias of the jth neuron in the lth layer after Bayesian regularization training; Step 2.3: Convert the position coordinates (x BDSBAS ,y BDSBAS ,z BDSBAS )、(x MSAS ,y MSAS ,z MSAS ), as input, trained and tested through the fully connected neural network model, and obtained the fused position coordinates (x Fusion ,y Fusion ,z Fusion ); When using the fully connected neural network model for fusion, set the learning rate to 0.01, the number of iterations to 1000, and the maximum number of failures to 10. Take the set proportion of the corrected position coordinate data obtained in step 1 as the training set to train the fully connected neural network model and obtain the BDSBAS and MSAS fusion positioning model based on the fully connected neural network. After obtaining the trained BDSBAS and MSAS fusion positioning model based on the fully connected neural network, the remaining BDSBAS position coordinate data (x′ BDSBAS ,y′ BDSBAS ,z′ BDSBAS ) and MSAS position coordinate data (x′ MSAS ,y′ MSAS ,z′ MSAS ), we can get the fused position data (x Fusion ,y Fusion ,z Fusion ).

4. The BDSBAS / MSAS fusion positioning method based on a fully connected neural network according to claim 1, characterized in that: The step 3 comprises the following steps: Step 3.1: Use the root mean square error as the evaluation method for comparison; let (x, y, z) be the coordinate data obtained after correction or fusion of BDSBAS and MSAS, (x true ,y true ,z true ) is the actual position value obtained after processing; (x, y, z) and (x true ,y true ,z true ) to obtain the difference (Δx, Δy, Δz): Then the root mean square error (RMS) of the (Δx, Δy, Δz) parameters is Δx 、RMS Δy 、RMS Δz ) is calculated as follows: Where N is the number of measurement epochs; The total positioning root mean square error RMS is: Step 3.2: Substitute the BDSBAS position coordinate data, MSAS position coordinate data and fused position data into the total positioning root mean square error RMS calculation formula, and judge the quality of the fusion positioning effect by comparison; Except for the training set, the remaining BDSBAS position coordinate data (x′ BDSBAS ,y′ BDSBAS ,z′ BDSBAS ), MSAS position coordinate data (x′ MSAS ,y′ MSAS ,z′ MSAS ) and the fused position data (x Fusion ,y Fusion ,z Fusion ) are substituted into the total positioning root mean square error RMS calculation formula to obtain RMS BDSBAS 、RMS MSAS 、RMS Fusion ; If RMS Fusion <RMS BDSBAS And RMS Fusion <RMS MSAS , it means that the positioning effect after fusion of the fully connected neural network is better than that of BDSBAS and MSAS, and the fused position coordinate data (x Fusion ,y Fusion ,z Fusion ), otherwise continue to return to step 2.3 to train the fully connected neural network model.