A method for satellite positioning in complex urban environments assisted by radio signals

Abstract

The present invention proposes a method for radio signal-assisted satellite positioning in a complex urban environment, comprising: step 1, WiFi fingerprint positioning based on a convolutional neural network (CNN) model with an attention mechanism, comprising: constructing a convolutional neural network (CNN) model based on an attention mechanism, and performing offline training; using the trained convolutional neural network (CNN) model based on the attention mechanism, solving a positioning result online, and obtaining a WiFi fingerprint positioning result; step 2, WiFi fingerprint positioning assisting a global navigation satellite system (GNSS) positioning, comprising: inverting a pseudorange according to a position obtained from a WiFi fingerprint positioning result; detecting the influence of non-line-of-sight (NLOS) signal reception; ensuring that a selected satellite in the global navigation satellite system (GNSS) positioning satisfies a geometric precision factor (GDOP); and using a weighted least squares method to perform position solution to complete positioning, i.e., completing radio signal-assisted satellite positioning in a complex urban environment.

Description

Technical Field

[0001] The invention relates to a satellite positioning method, in particular to a method for satellite positioning in a complex urban environment assisted by radio signals. Background Art

[0002] With the popularity of smart phones, radio signals such as WiFi, UWB and Bluetooth are used for positioning. In addition, WiFi positioning technology has the advantages of low hardware equipment cost, low computing overhead and relatively high positioning accuracy, and is widely used in indoor positioning. WiFi positioning methods can be divided into triangulation positioning method and fingerprint positioning method. The latter calculates the position by matching the collected RSS data with the fingerprint in the fingerprint database. Due to the influence of non-line-of-sight signals and multipath effects, it is difficult to map RSS to distance, so more research is based on fingerprint database positioning.

[0003] In urban environments, satellite signals are easily blocked. If only GNSS is used for positioning, the number of available satellites will be affected, resulting in reduced positioning accuracy. In addition, WiFi signals are less affected by NLOS, but are mostly used for indoor positioning. Among the methods for detecting and eliminating NLOS, there are few studies on using WiFi signals to detect NLOS-assisted satellite positioning.

[0004] When using traditional WiFi fingerprint positioning methods, mismatching often occurs because the matching algorithm is relatively complex and requires a lot of computing time and resources. At present, machine learning methods are mostly used to optimize the WiFi fingerprint matching process. Since the data volume of the fingerprint database is not large, if the model complexity used is high, it is easy to cause data overfitting, increase the computing burden, and is not conducive to the extraction of fingerprint data features. Summary of the invention

[0005] Purpose of the invention: The technical problem to be solved by the present invention is to provide a method for satellite positioning in complex urban environments assisted by radio signals in view of the deficiencies in the prior art.

[0006] In order to solve the above technical problems, the present invention discloses a method for satellite positioning in a complex urban environment assisted by radio signals, comprising the following steps:

[0007] Step 1: WiFi fingerprint positioning based on the convolutional neural network (CNN) model with attention mechanism;

[0008] Step 1-1: Build a convolutional neural network (CNN) model based on the attention mechanism and perform offline training. The specific steps are as follows:

[0009] Step 1-1-1, data preprocessing, the specific method is as follows:

[0010] The received signal strength RSS is preprocessed, that is, standardized as follows:

[0011]

[0012] Among them, RSS` is the data obtained after preprocessing, min(RSS) is the minimum value of the collected WiFi signal strength, and max(RSS) is the maximum value of the WiFi signal strength;

[0013] The fingerprint data of the WiFi signal is converted into a square matrix. If there are z fingerprint points of the observed samples, the data of the z fingerprint points are integrated into a matrix, and the missing parts of the matrix are filled with 0.

[0014] Step 1-1-2, build a convolutional neural network CNN model based on the attention mechanism. The specific method is as follows:

[0015] The convolutional neural network (CNN) model structure based on the attention mechanism is composed of two convolutional layers of size 3×3, two pooling layers of size 2×2, an attention layer and a fully connected layer; the CBAM attention mechanism is adopted, and the CBAM attention mechanism combines the channel and spatial attention mechanism modules and is integrated into the convolutional neural network (CNN) structure;

[0016] The ReLU function is used as the activation function; a dropout layer is added before the fully connected layer; the dropout parameter is set to 0.5, and the learning rate is selected to be 0.0001; the number of convolution kernels in the first convolution layer is set to 100, and the number of convolution kernels used in the second convolution layer is 300;

[0017] The convolutional neural network (CNN) model based on the attention mechanism is trained offline using a pre-collected data set to obtain a fixed feature weight corresponding to the data of each fingerprint point; the coordinates of the fingerprint point and the corresponding fixed feature weight are stored in a WiFi fingerprint database.

[0018] Step 1-2, use the trained convolutional neural network (CNN) model based on the attention mechanism to solve the positioning result online and obtain the WiFi fingerprint positioning result. The specific method is as follows:

[0019] After preprocessing the detected WiFi signal strength data according to the method in step 1-1-1, the input data X is obtained and input into the pre-trained convolutional neural network CNN model based on the attention mechanism to obtain the feature weights; the probability P of the observed position, i.e. the user position, at the jth fingerprint point is calculated. j , the method is as follows:

[0020]

[0021] Among them, σ x is the variance of the input data X, ax is the variance parameter of the input data X, Y j is the output data at fingerprint point j, Y v is the output data at fingerprint point v, z is the number of fingerprint points, and v is the index variable used to traverse all fingerprint points;

[0022] Select the first M fingerprint points with the highest probability as neighboring points, and calculate the user location coordinates S by weighted average, and the weight G of the kth neighboring point k The calculation method is as follows:

[0023]

[0024]

[0025] Among them, P k is the probability that the user's location is at the kth fingerprint point, P q is the probability that the user's position is at the qth fingerprint point, q is the index variable used to traverse all neighboring points, S is the coordinate of the observed position, that is, the positioning result obtained by the solution, S k is the coordinate of the kth fingerprint point selected.

[0026] Step 2: WiFi fingerprint positioning aids global navigation satellite system GNSS positioning;

[0027] Step 2-1, inverting the pseudo-range based on the position obtained from the WiFi fingerprint positioning result, the specific method includes:

[0028] The user location coordinates obtained by WiFi fingerprint positioning based on the attention mechanism CNN model are (X r , Y r , Z r ) Assuming that the received satellite signal is a line-of-sight signal, the pseudo-range ρ of the i-th receivable satellite is inverted si for:

[0029]

[0030] Among them, (X si， Y si , Z si ), i∈(1, 2, ..., n) is the coordinate of the i-th satellite, the number of satellite signals received at this location is n, τ r is the satellite signal receiver clock error, τ si is the clock error of the ith satellite, c is the speed of light in vacuum, V ion is the ionospheric delay, V trop It is the tropospheric delay.

[0031] Step 2-2, detecting the impact of non-line-of-sight reception of NLOS signals, includes the following steps:

[0032] Step 2-2-1: Subtract the pseudorange observation value from the inversion value. The specific method includes:

[0033] At the user's location, the pseudorange observation values ​​of the n visible satellites are subtracted from the inverted pseudorange values, and the pseudorange observation value of the i-th satellite is and the inverted pseudorange value ρ si Make the difference and take the absolute value to get the absolute value C of the difference between the pseudorange observation value and the inversion value of the i-th satellite si ;

[0034] Step 2-2-2, by judging the influence of non-line-of-sight reception of NLOS signals, selecting satellites to participate in position solution, the specific method includes:

[0035] Set the threshold T 1 , if the absolute value of the difference between the pseudorange observation value and the inversion value of the i-th satellite is C si ≤T 1 , then select the satellite to participate in the position solution; all visible satellites at the user's location are compared and saved in the satellite dataset.

[0036] Step 2-3, making the selected satellites in the global navigation satellite system GNSS positioning meet the geometric precision factor GDOP, the specific method includes:

[0037] According to the pseudorange observation equation of the selected satellite, calculate the current geometric precision factor GDOP value; set the threshold T 2 , when GDOP<T 2 When GDOP≥T 2 When T is adjusted upward 1 value, repeat step 2-2-2 until the geometric precision factor GDOP meets the above conditions.

[0038] Step 2-4, use weighted least squares method to solve the position and complete the positioning, that is, complete the radio signal assisted satellite positioning in complex urban environments. The specific method is as follows:

[0039] The weight coefficient ω of the i-th satellite si It is expressed as:

[0040]

[0041] Among them, σ 1 is the mean square error associated with the satellite, which is solved by the user ranging factor N.

[0042]

[0043] The weighted least square mathematical model of satellite pseudorange single point positioning is:

[0044] WGΔX=Wb

[0045] Where W is the weight matrix, W = diag(ω s1 ,ω s2 ,…,ω sf ), f is the number of satellites selected to be stored in the satellite data set; G is the direction cosine matrix of the satellite, ΔX is the calculated three-dimensional position correction value and the receiver clock error, and b is composed of the difference between the pseudorange observation value of each satellite and the pseudorange inverted from the calculated receiver position; solve the equation ΔX = (G T W T WG) - 1 G T W T Wb, and then through repeated iterations, the coordinate position of the receiver is solved to complete the positioning.

[0046] Beneficial effects:

[0047] In view of the problem that GNSS positioning is easily affected by NLOS in urban environments and has large pseudorange measurement errors. The present invention proposes to use WiFi signals to alleviate the impact of NLOS in GNSS positioning solutions, and invert the pseudoranges of the receivable satellites based on the positioning results of the WiFi fingerprint database. In addition, the inverted pseudorange value is subtracted from the pseudorange observation value of the GNSS at that location, and the satellites that are less affected by NLOS are compared and selected, and it is determined that the selected satellites meet the GDOP requirements. Using the weighted least squares method, the NLOS signal is downweighted based on the difference between the pseudorange inversion value and the observation value to perform position solution.

[0048] Traditional WiFi fingerprint database positioning methods often have problems such as mismatching, making the positioning results unreliable. In order to improve the accuracy of subsequent inversion pseudorange, an accurate user location should be obtained first. The machine learning theory is introduced into the WiFi fingerprint matching problem, and the CNN model based on the attention mechanism is used to train the WiFi fingerprint data. Adding the attention mechanism to the CNN network makes it easier to train parameters, and by giving high weights to important fingerprint information. When the input data is large, it can reduce the amount of calculation and increase the running speed, which improves the performance of the CNN model. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] The present invention will be further described in detail below in conjunction with the accompanying drawings and specific embodiments, and the above and / or other advantages of the present invention will become more clear.

[0050] Figure 1 It is a schematic diagram of the process of the present invention.

[0051] Figure 2Schematic diagram of the convolutional neural network (CNN) model structure of the attention mechanism in the present invention. DETAILED DESCRIPTION

[0052] The core content of the present invention is a method for inverting pseudorange based on WiFi fingerprint positioning results, assisting GNSS (Global Navigation Satellite System, GNSS) to detect NLOS (Non line of sight, NLOS) impact and weighted position solution. It includes the following two stages: (1) Obtain the user's coordinate position through WiFi fingerprint positioning based on the CNN (Convolutional Neural Network, CNN) model of the attention mechanism. (2) Invert the pseudorange at the positioning position, detect the NLOS impact, and then use the weighted least squares algorithm to reduce the weight of the NLOS signal for position solution. The overall process is as follows: Figure 1 As shown:

[0053] (1) WiFi fingerprint positioning method based on CNN model of attention mechanism

[0054] 1) Offline training of network model

[0055] Step 1: Data Preprocessing

[0056] In order to facilitate subsequent data processing, RSS (Received Signal Strength, RSS) is standardized:

[0057]

[0058] Among them, RSS` is the data obtained after preprocessing, min(RSS) is the minimum value of the collected WiFi signal strength, and max(RSS) is the maximum value of the WiFi signal strength.

[0059] Before model training, the fingerprint data should be converted into a square matrix. Assuming there are z observed sample fingerprint points, the data of z fingerprint points can be integrated into a matrix. In order to ensure that the rows and columns of the matrix are consistent, the missing parts should be filled with 0.

[0060] Step 2: Build a CNN model based on attention mechanism

[0061] The present invention uses a CNN model structure based on the attention mechanism as follows Figure 2As shown, it consists of two 3×3 convolutional layers, two 2×2 pooling layers, an attention layer and a fully connected layer. The present invention adopts the CBAM attention mechanism (reference: Cbam: Convolutional block attention module [C] / / Proceedings of the European conference on computer vision (ECCV). 2018: 3-19.), which combines the channel and spatial attention mechanism modules, and as a lightweight general module can be seamlessly integrated into the CNN structure, which improves the performance of the CNN model.

[0062] The activation function uses the ReLU function to effectively solve the gradient explosion and gradient vanishing problems in the process of gradient descent and back propagation. In order to prevent overfitting, a dropout layer is added before the fully connected layer. The dropout parameter is set to 0.5, and the learning rate is selected as 0.0001. The number of convolution kernels in the first convolution layer is set to 100, and the number of convolution kernels used in the second convolution layer is 300.

[0063] When the training is completed, the fixed feature weight corresponding to each fingerprint point data can be obtained and stored in the database.

[0064] 2) Online solution of positioning results

[0065] In the online stage, the observed WiFi signal strength data is preprocessed and then input into the pre-trained network to obtain feature weights and outputs. The probability of the observed position being at the jth fingerprint point is calculated as:

[0066]

[0067] Among them, σ x is the variance of the input data X, a x is the variance parameter of the input data X, Y j is the output data at fingerprint point j, Y v is the output data at fingerprint point v, z is the number of fingerprint points, and v is the index variable used to traverse all fingerprint points.

[0068] Select the first M fingerprint points with the highest probability as neighboring points, calculate the user location coordinates S by weighted average, and the weight G of the kth neighboring point k The calculation method is as follows:

[0069]

[0070]

[0071] Among them, Pk is the probability that the user's location is at the kth fingerprint point, P q is the probability that the user's position is at the qth fingerprint point, q is the index variable used to traverse all neighboring points, S is the coordinate of the observed position, that is, the positioning result obtained by the solution, S k is the coordinate of the kth fingerprint point selected.

[0072] (2) WiFi fingerprint positioning assisted GNSS positioning

[0073] 1) WiFi fingerprint positioning position inversion pseudorange

[0074] The user location coordinates located by the WiFi fingerprint database are (X r , Y r , Z r ). Assuming that the received satellite signal is a LOS (Line-of-sight) signal, the pseudorange ρ of the i-th receivable satellite is inverted si for:

[0075]

[0076] Where (X si , Y si , Z si ), i∈(1, 2, ..., n) is the coordinate of the i-th satellite, the number of satellite signals that can be received at this location is n, τ r is the receiver clock error, τ si is the clock error of the ith satellite, c is the speed of light in vacuum, V ion is the ionospheric delay, V trop It is the tropospheric delay.

[0077] 2) Detecting the impact of NLOS signals

[0078] Step 1: Difference between pseudorange observations and inversion values

[0079] At the user's location, the difference between the pseudorange observation values ​​of the n visible satellites and the inverted pseudorange values ​​is calculated, such as the pseudorange observation value of the i-th satellite and the inverted pseudorange value ρ si Make a difference and then take the absolute value to get C si .

[0080] Step 2: Select satellites to participate in position calculation by judging the NLOS impact

[0081] In an urban environment, the difference between the pseudorange observation value and the inversion value can reflect the degree to which the satellite signal is affected by NLOS to a certain extent. 1 , if the satellite's C si ≤T1 , then select the satellite. All n visible satellites at the user's position are compared, and the selected satellite is stored in the data set. 1 Set to 15.

[0082] 3) Make the selected satellite meet the GDOP (Geometric Dilution of Precision, GDOP) requirements

[0083] According to the pseudo-range observation equation of the selected satellite, the current GDOP value is calculated (reference: A satellite selection method for a multi-mode GNSS receiver [P]. Beijing: CN103954980A, 2014-07-30.), and the satellite with less NLOS influence is selected for position solution through the threshold T1, but it may make the satellite geometry layout worse. In order to ensure the positioning accuracy, the threshold T is set. 2 , when GDOP<T 2 When GDOP≥T 2 When T is adjusted upward 1 The satellite selection process for position solution is repeated until the calculated GDOP meets the conditions. 2 is 6.

[0084] 4) Weighted least squares method for position calculation

[0085] The weight coefficient of the i-th satellite can be expressed as:

[0086]

[0087] Among them, C si is the absolute value of the difference between the pseudorange value observed by the ith satellite and the simulated pseudorange value inverted, σ 1 is the mean square error associated with the satellite and can be solved by the user ranging factor N.

[0088]

[0089] The weighted least square mathematical model of satellite pseudorange single point positioning is:

[0090] WGΔX=Wb

[0091] Where W is the weight matrix, W = diag(ω s1 ,ω s2 ,…,ω sf ), f is the number of satellites selected to be stored in the data set; G is the direction cosine matrix of the satellite; ΔX is the calculated three-dimensional position correction value and the receiver clock error; b is the difference between the pseudorange observation value of each satellite and the pseudorange inverted from the calculated receiver position. Solve the equation ΔX = (GT W T WG) -1 G T W T Wb, and then the coordinate position of the receiver is solved after multiple iterations.

[0092] In a specific implementation, the present application provides a computer storage medium and a corresponding data processing unit, wherein the computer storage medium can store a computer program, and when the computer program is executed by the data processing unit, the invention content of the method for satellite positioning in a complex urban environment assisted by a radio signal provided by the present invention and some or all of the steps in each embodiment can be executed. The storage medium can be a disk, an optical disk, a read-only memory (ROM) or a random access memory (RAM), etc.

[0093] Those skilled in the art can clearly understand that the technical solutions in the embodiments of the present invention can be implemented by means of computer programs and their corresponding general hardware platforms. Based on this understanding, the technical solutions in the embodiments of the present invention are essentially or partly contributed to the prior art can be embodied in the form of a computer program, i.e., a software product, which can be stored in a storage medium and includes several instructions for enabling a device including a data processing unit (which can be a personal computer, a server, a single-chip microcomputer, a MUU or a network device, etc.) to execute the methods described in various embodiments of the present invention or certain parts of the embodiments.

[0094] The present invention provides a method and idea of ​​a method for satellite positioning in a complex urban environment assisted by radio signals. There are many methods and ways to implement the technical solution. 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 principle of the present invention. These improvements and modifications should also be regarded as the protection scope of the present invention. All components not specified in this embodiment can be implemented by existing technologies.

Claims

1. A method for satellite positioning in a complex urban environment assisted by radio signals, characterized in that: It includes the following steps: Step 1, WiFi fingerprint positioning based on a convolutional neural network (CNN) model with an attention mechanism; Step 1-1, construct a convolutional neural network (CNN) model with an attention mechanism and perform offline training; Step 1-2, use the trained convolutional neural network (CNN) model with an attention mechanism to solve the positioning result online and obtain the WiFi fingerprint positioning result; Step 2, WiFi fingerprint positioning assisted by the Global Navigation Satellite System (GNSS); Step 2-1, invert the pseudorange according to the position obtained from the WiFi fingerprint positioning result; Step 2-2, detect the influence of non-line-of-sight (NLOS) received signals; Step 2-3, make the selected satellites in the Global Navigation Satellite System (GNSS) positioning satisfy the Geometric Dilution of Precision (GDOP); Step 2-4, perform position calculation using the weighted least squares method to complete the positioning, that is, complete the satellite positioning assisted by radio signals in complex urban environments; Among them, the specific method for inverting the pseudorange according to the position obtained from the WiFi fingerprint positioning result in Step 2-1 includes: The user location coordinates obtained by WiFi fingerprint positioning based on the attention mechanism CNN model are (X r ,Y r ,Z r ) Assuming that the received satellite signal is a line-of-sight signal, the pseudo-range ρ of the i-th receivable satellite is inverted si for: Among them, (X si ,Y si ,Z si ), i∈(1,2,…,n) is the coordinate of the i-th satellite, the number of satellite signals received at this location is n, τ r is the satellite signal receiver clock error, τ si is the clock error of the ith satellite, c is the speed of light in vacuum, V ion is the ionospheric delay, V trop It is the tropospheric delay; The detection of the influence of non-line-of-sight (NLOS) received signals in Step 2-2 includes the following steps: Step 2-2-1, take the difference between the pseudorange observation value and the inversion value; Step 2-2-2, select the satellites participating in the position calculation by judging the influence of non-line-of-sight (NLOS) received signals; The specific method for taking the difference between the pseudorange observation value and the inversion value in Step 2-2-1 includes: At the user's location, the pseudorange observation values ​​of the n visible satellites are subtracted from the inverted pseudorange values, and the pseudorange observation value of the i-th satellite is and the inverted pseudorange value ρ si Make the difference and take the absolute value to get the absolute value C of the difference between the pseudorange observation value and the inversion value of the i-th satellite si ; The specific method for selecting the satellites participating in the position calculation by judging the influence of non-line-of-sight (NLOS) received signals in Step 2-2-2 includes: Set the threshold T1, if the absolute value C of the difference between the pseudorange observation value and the inversion value of the i-th satellite si ≤T1, the satellite is selected to participate in the position solution; all visible satellites at the user's position are compared and saved in the satellite data set; The specific method for making the selected satellites in the Global Navigation Satellite System (GNSS) positioning satisfy the Geometric Dilution of Precision (GDOP) in Step 2-3 includes: According to the pseudorange observation equation of the selected satellites, calculate the current Geometric Dilution of Precision (GDOP) value; set a threshold T2. When GDOP < T2, the selected satellites are used for position calculation; when GDOP ≥ T2, then adjust the value of T1 upward and repeat Step 2-2-2 until the Geometric Dilution of Precision (GDOP) satisfies the condition of GDOP < T2; The specific method for performing position calculation using the weighted least squares method in Step 2-4 is as follows: The weight coefficient ω of the i-th satellite si It is expressed as: Among them, σ1 is the mean square error related to the satellite, which is solved by the user range factor N; The weighted least squares mathematical model for single-point positioning of satellite pseudorange is: WGΔX = Wb Where W is the weight matrix, W = diag(ω s1 ,ω s2 ,…,ω sf ), f is the number of satellites selected to be stored in the satellite data set; G is the direction cosine matrix of the satellite, ΔX is the calculated three-dimensional position correction value and the receiver clock error, and b is composed of the difference between the pseudorange observation value of each satellite and the pseudorange inverted from the calculated receiver position; solve the equation ΔX = (G T W T WG) -1 G T W T Wb, and then through repeated iterations, the coordinate position of the receiver is solved to complete the positioning.

2. The method for satellite positioning in complex urban environments assisted by radio signals according to claim 1, characterized in that: The specific steps for constructing a convolutional neural network (CNN) model with an attention mechanism and performing offline training in Step 1-1 are as follows: Step 1-1-1, data preprocessing; Step 1-1-2, construct a convolutional neural network (CNN) model with an attention mechanism.

3. The method of radio signal assisted satellite positioning in complex urban environments according to claim 2, characterized in that: The specific method for data preprocessing in Step 1-1-1 is as follows: Preprocess the Received Signal Strength (RSS), that is, standardize it as follows: Among them, RSS` is the data obtained after preprocessing, min(RSS) is the minimum value of the collected WiFi signal strength, and max(RSS) is the maximum value of the WiFi signal strength; The fingerprint data of the WiFi signal is converted into a square matrix. If there are z fingerprint points of the observed samples, the data of the z fingerprint points are integrated into a matrix, and the missing parts of the matrix are filled with 0.

4. The method for satellite positioning in complex urban environments assisted by radio signals according to claim 3, characterized in that: The specific method of building a convolutional neural network CNN model based on the attention mechanism described in step 1-1-2 is as follows: The convolutional neural network (CNN) model structure based on the attention mechanism is composed of two convolutional layers of size 3×3, two pooling layers of size 2×2, an attention layer and a fully connected layer; the CBAM attention mechanism is adopted, and the CBAM attention mechanism combines the channel and spatial attention mechanism modules and is integrated into the convolutional neural network (CNN) structure; The ReLU function is used as the activation function; a dropout layer is added before the fully connected layer; the dropout parameter is set to 0.5, and the learning rate is selected to be 0.0001; the number of convolution kernels in the first convolution layer is set to 100, and the number of convolution kernels used in the second convolution layer is 300; The convolutional neural network (CNN) model based on the attention mechanism is trained offline using a pre-collected data set to obtain a fixed feature weight corresponding to the data of each fingerprint point; the coordinates of the fingerprint point and the corresponding fixed feature weight are stored in a WiFi fingerprint database.

5. The method for satellite positioning in a complex urban environment assisted by radio signals according to claim 4, characterized in that: The specific method for solving the positioning results online in steps 1-2 is as follows: After preprocessing the detected WiFi signal strength data according to the method in step 1-1-1, the input data X is obtained and input into the pre-trained convolutional neural network CNN model based on the attention mechanism to obtain the feature weights; the probability P of the observed position, i.e. the user position, at the jth fingerprint point is calculated. j , the method is as follows: Among them, σ x is the variance of the input data X, a x is the variance parameter of the input data X, Y j is the output data at fingerprint point j, Y v is the output data at fingerprint point v, z is the number of fingerprint points, and v is the index variable used to traverse all fingerprint points; Select the first M fingerprint points with the highest probability as neighboring points, calculate the user location coordinates S by weighted average, and the weight G of the kth neighboring point k The calculation method is as follows: Among them, P k is the probability that the user's location is at the kth fingerprint point, P q is the probability that the user's position is at the qth fingerprint point, q is the index variable used to traverse all neighboring points, S is the coordinate of the observed position, that is, the positioning result obtained by the solution, S k is the coordinate of the kth fingerprint point selected.

Images