Mobile positioning method based on recurrent neural network

Abstract

The invention relates to the field of deep learning and communication, in particular to a mobile positioning method based on a recurrent neural network, and the method specifically comprises the steps: S1, building a mobile positioning system, collecting data based on the mobile positioning system, carrying out the normalization processing of the collected data, obtaining a data set, and dividing the data set into a training set and a test set; s2, constructing a recurrent neural network model, and training the recurrent neural network model through the training set to obtain a mobile positioning model; and S3, inputting the test set into the mobile positioning model to obtain positioning data, and carrying out reverse normalization processing on the positioning data to obtain a mobile point position. By means of the high-precision positioning technology, large machinery and constructors can be positioned in real time, the safety distance can be controlled easily, the safety and economical efficiency of construction are improved, and large machinery construction becomes safer and more efficient.

Description

Technical Field

[0001] The present application relates to the field of deep learning and communication, and more specifically, to a mobile positioning method based on a recurrent neural network. Background Art

[0002] With the acceleration of urbanization, high-voltage lines are an important infrastructure for urban power transmission, and construction activities around them are becoming more and more frequent. However, high-voltage line discharge accidents caused by illegal crane operations not only cause serious damage to the power grid, but also seriously threaten the personal safety of construction workers and surrounding residents, becoming a social problem that needs to be solved urgently.

[0003] In order to fundamentally reduce the incidence of construction accidents and protect the safety of power grids and personnel, researchers and engineers have jointly developed a large-scale machinery operation safety control instrument. The device integrates a number of advanced technologies. Its core function is to define safe working areas and monitor the safe distance between machinery and high-voltage lines in real time to ensure that construction activities are carried out in an absolutely safe environment.

[0004] Mobile positioning technology plays a vital role in large-scale mechanical operation safety control instruments. At present, the Global Positioning System (GPS) and its auxiliary technology (A-GPS) are the mainstream positioning means. With their high precision, all-weather and global coverage, they can meet the positioning needs in most cases. However, in complex urban environments, especially in areas with high-rise buildings and severe signal obstruction, the multipath propagation problem of GPS signals becomes particularly prominent, resulting in a significant decrease in positioning accuracy or even complete failure. In addition, the high power consumption characteristics of GPS technology also limit its long-term use on mobile devices.

[0005] As a deep learning model, RNN is particularly good at processing time dependency and long-term dependency in sequence data, which gives it a unique advantage in processing multipath effects and dynamic changes in wireless signal propagation. By training the RNN model, the complex characteristics of the signal propagation path can be learned, so that more accurate positioning can be achieved even when the signal is interfered or blocked.

[0006] In summary, the traditional positioning system is subject to the obstruction of urban buildings and the positioning accuracy decreases. It is urgent to propose a positioning method that is undisturbed and accurate in positioning to alleviate the current state of low positioning accuracy. Summary of the invention

[0007] In view of the fact that the positioning system in the prior art is subject to obstruction by urban buildings and the positioning accuracy decreases, the present application proposes a mobile positioning method based on a recurrent neural network.

[0008] The scheme of this application is as follows:

[0009] A mobile positioning method based on a recurrent neural network comprises the following steps:

[0010] Step S1: Build a mobile positioning system, collect data based on the mobile positioning system, normalize the collected data to obtain a data set, and divide the data set into a training set and a test set;

[0011] Step S2: construct a recurrent neural network model, and train the recurrent neural network model through a training set to obtain a mobile positioning model;

[0012] Step S3: input the test set into the mobile positioning model to obtain positioning data, and perform denormalization on the positioning data to obtain the position of the moving point.

[0013] Preferably, the mobile positioning system includes a mobile station, a mobile positioning server, a data server and a model server; the mobile station is any one of a vehicle-mounted type, a portable type or a handheld type, and is provided with a GPS module and a signal receiving module, the GPS module is used to estimate the GPS coordinates of the mobile station, and the signal receiving module is used to collect network signals, the network signals are RSSIs of neighbor base stations and Wi-Fi wireless access points of heterogeneous networks, and the heterogeneous networks are cellular networks and Wi-Fi networks.

[0014] Preferably, during the training phase, the mobile station collects GPS coordinates and network signals and sends the GPS coordinates and network signals to a mobile positioning server; during the execution phase, the mobile station sends the collected network signals to the mobile positioning server for position estimation.

[0015] Preferably, the data server stores the GPS coordinates and network signals from the mobile station through the mobile positioning server, and the GPS coordinates and network signals can be queried and used to train the recurrent neural network model; the model server saves the trained recurrent neural network model from the mobile positioning server for estimating the position of the mobile point.

[0016] Preferably, the normalization process in step S1 is performed using the minimum and maximum values ​​of the rssi and coordinates, as follows:

[0017] Step S11. Collect network signals and GPS coordinates, and the mobile station i The RSSI from the cellular network can be detected at any time, that is, R in formula (1) c,i , the rssi of the Wi-Fi wireless access point, that is, R in formula (2) w,i and GPS coordinates, i.e. l ​​in formula (3) i , define the tth i The rssi of the jth base station from the cellular network at the moment is rc,j,i , defining the kth wireless access point from the Wi-Fi network at time t i The rssi is r w,k,i ; The network signal t i The rssi data set at the moment is defined as R i (As shown in formula (4)); In addition, position l i Including longitude x,i and latitude l y,i , assuming that a total of m locations are detected, n1 different base stations, and n2 different Wi-Fi wireless access points, if the RSSI of the network signal cannot be detected, it can be set to empty;

[0018]

[0019] l i = {l x,i , l y,i} (3)

[0020] R i = {R c,i , R w,i} (4)

[0021] Step S12: The network signal and GPS coordinates collected in step S11 are normalized using the minimum and maximum values ​​of RSSI and GPS coordinates. According to the minimum and maximum values ​​of RSSI in the cellular network, that is, and Define the jth base station at time t i The normalized rssi is c j,i ; According to the minimum and maximum values ​​of rssi in the Wi-Fi network, that is, and Define the kth Wi-Fi wireless access point in the cellular network at time t i The normalized rssi of k,i ; According to the minimum and maximum values ​​of longitude in GPS coordinates, that is, and Definition i The normalized longitude at the time is x i ; According to the minimum and maximum values ​​of latitude in GPS coordinates, that is, and Definition i The normalized latitude at the time is y i ;

[0022]

[0023]

[0024] Preferably, the recurrent neural network model constructed in step S2 adopts either a recurrent neural network with one timestamp or two timestamps. When a single time series is used as the input of the timestamp, a recurrent neural network with one timestamp is adopted; when two time series are used as the input of the timestamp, a recurrent neural network with two timestamps is adopted.

[0025] Preferably, the recurrent neural network training method for a timestamp is as follows:

[0026] S211. Based on the recurrent neural network consisting of an input layer, a recursive hidden layer and an output layer, the input layer is set to n1 base stations and n2 Wi-Fi access points The output layer is the estimated normalized longitude and latitude and ; The recursive hidden layer contains at least one neuron, and the initial value of the neuron in the recursive hidden layer is defined as h0;

[0027] S212. Based on the above settings, after calculating the RSSI of the first timestamp, the value of the cyclic hidden layer neuron can be updated to h1; c j,i , w k,i The weights of h0 and h1 are α j , β k and v; output and The h1 weights are γ1 and γ2 respectively; the bias of neurons in the hidden layer and output layer is defined as b 1,1 , b 2,1 and b 3,1 ;

[0028] S213. Select the sigmoid function as the activation function of each neuron, then h0, h1, The value of can be calculated by equations (9) to (12) respectively, and the loss function E is defined according to the square of the error as equation (13);

[0029] h0=0(9)

[0030]

[0031] S214. Optimize the recurrent neural network and use the learning rate η and gradient descent method to update each weight and bias, where γ1, γ2, b 2,1 , b 3,1 , α j , β k , v and b 1,1 The update of is proved and calculated by equations (14) to (21) respectively;

[0032]

[0033]

[0034] Preferably, in step S211, when the recursive hidden layer includes more than one neuron, the weights between the neurons are updated by the gradient descent method.

[0035] Preferably, the recurrent neural network training method for the two timestamps is as follows:

[0036] S221. Based on the recurrent neural network consisting of an input layer, a recursive hidden layer and an output layer, the input layer is set to the (n1+n2) standardized rssi (c 1,i , c 2,i , …, and ω 1,i ,ω 2,i , …, and (n1+n2) normalized rssi at the second timestamp (c 1,i+1 , c 2,i+1 , …, and ω 1,i+1 ,ω 2,i+1 , …, ; The output layer is the estimated normalized longitude and latitude at the second timestamp and The recursive hidden layer contains one neuron, and the initial value of the neuron in the recursive hidden layer is defined as h0, as shown in formula (9);

[0037] h0=0(9)

[0038] S222. Based on the above settings, the value of the cyclic hidden layer neuron is updated to h1 in the first timestamp and to h2 in the second timestamp; the weights of the j-th base station and the k-th Wi-Fi wireless access point P in each timestamp are α j and β k Output and The h2 weights are γ1 and γ2 respectively; the weight of the hidden layer neuron in the minimum timestamp loop is defined as v; the deviation of the hidden layer and output layer neurons is defined as b 1,1 , b 2,1 and b 3,1 ;

[0039] S223. Select the sigmoid function as the activation function of each neuron, so h1, h2, The values ​​of can be calculated by equations (22) to (25) respectively; according to the square of the error, the loss function E is defined as equation (26);

[0040]

[0041] S224. For the optimization of a recurrent neural network with two consecutive timestamps, the learning rate η and the gradient descent method are used to update each weight and bias; γ1, γ2, b 2,1 , b 3,1 , α j , β k , v and b 1,1 The update of is proved and calculated by equations (27) to (34) respectively;

[0042]

[0043]

[0044]

[0045] Preferably, the denormalization process in step S3 is:

[0046] For denormalization and estimation, the estimated normalized latitude and longitude are used in equations (35) and (36) respectively. and To retrieve the estimated latitude and longitude and

[0047]

[0048] in, is the minimum value of longitude in GPS coordinates, is the maximum value of longitude in GPS coordinates, is the minimum value of longitude in GPS coordinates, The maximum value of longitude in GPS coordinates.

[0049] Beneficial effects of this application:

[0050] This application constructs a recurrent neural network model and uses a training set for training. The model can learn the complex features and nonlinear relationships in the data, thereby improving the positioning accuracy. In particular, the recurrent neural network can capture the long-term dependencies in time series data, which is particularly important for continuous signal analysis in mobile positioning. The method supports collecting signals from heterogeneous networks (such as cellular networks and Wi-Fi networks), including the RSSI of neighboring base stations of cellular networks and Wi-Fi wireless access points. This multi-source data fusion method can provide richer positioning information and improve the robustness and accuracy of the positioning system. The method provides a flexible model construction method, and a recurrent neural network with a single timestamp or two timestamps can be selected according to actual needs. In addition, the structure and parameters of the model can be adjusted and optimized according to the specific application scenario to adapt to different positioning requirements and environmental changes. Data normalization enables data from different sources and dimensions to be compared and analyzed at the same scale, which helps to improve the training effect and prediction accuracy of the model. At the same time, training the recurrent neural network through optimization algorithms such as gradient descent can further reduce errors and improve the performance of the model. The position of the moving point finally obtained by this application can be denormalized to convert the normalized coordinate value output from the model back to the real longitude and latitude coordinates. With high-precision positioning technology, this application can locate large machinery and construction personnel in real time, which is conducive to controlling the safe distance, improving the safety and economy of construction, and making large-scale machinery construction safer and more efficient. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] Attached Figure 1 This is a structural diagram of a mobile positioning system based on a recurrent neural network in this application;

[0052] Attached Figure 2 A scene graph of network signals and global positioning system (GPS) coordinates collected for this application;

[0053] Attached Figure 3 A diagram of the recurrent neural network structure with a timestamp for this application;

[0054] Attached Figure 4 Diagram of the recurrent neural network structure for two timestamps in this application. DETAILED DESCRIPTION

[0055] In order to further explain the technical means and effects adopted by the present application to achieve the predetermined invention purpose, the specific implementation method, structure, characteristics and effects of the present application are described in detail below in combination with the accompanying drawings and preferred embodiments.

[0056] Example

[0057] This embodiment provides a mobile positioning method based on a recurrent neural network, comprising the following steps:

[0058] Step S1: Figure 1 As shown, a mobile positioning system is built, data is collected based on the mobile positioning system, and the collected data is normalized to obtain a data set, and the data set is divided into a training set and a test set;

[0059] Among them, the mobile positioning system includes a mobile station, a mobile positioning server, a data server and a model server; the mobile station is any one of vehicle-mounted, portable or handheld types, and is equipped with a GPS module and a signal receiving module. The GPS module is used to estimate the GPS coordinates of the mobile station, and the signal receiving module is used to collect network signals. The network signals are the RSSI of neighbor base stations and Wi-Fi wireless access points of heterogeneous networks. The heterogeneous networks are cellular networks and Wi-Fi networks.

[0060] During the training phase, the mobile station collects GPS coordinates and network signals and sends them to the mobile positioning server; during the execution phase, the mobile station sends the collected network signals to the mobile positioning server for position estimation.

[0061] The data server stores the GPS coordinates and network signals from the mobile station through the mobile positioning server. The GPS coordinates and network signals can be queried and used to train the recurrent neural network model; the model server saves the trained recurrent neural network model from the mobile positioning server for estimating the position of the mobile point.

[0062] The normalization process in step S1 uses the minimum and maximum values ​​of rssi and coordinates for normalization, as follows:

[0063] Step S11. Collect network signals and GPS coordinates, and the mobile station i The RSSI from the cellular network can be detected at any time, that is, R in formula (1) c,i , the rssi of the Wi-Fi wireless access point, that is, R in formula (2) w,i and GPS coordinates, i.e. l ​​in formula (3) i , define the tth i The rssi of the jth base station from the cellular network at the moment is r c,j,i , defining the kth wireless access point from the Wi-Fi network at time t i The rssi is r w,k,i ; The network signal t i The rssi data set at the moment is defined as R i (As shown in formula (4)); In addition, position l i Including longitude x,i and latitude l y,i, assuming that a total of m locations are detected, n1 different base stations, and n2 different Wi-Fi wireless access points, if the RSSI of the network signal cannot be detected, it can be set to empty; for example, in Figure 2 The time t i , the mobile station cannot detect the rssi of the Wi-Fi wireless access point, so it will r w,2,i The value of is set to empty;

[0064]

[0065] l i = {l x,i , l y,i} (3)

[0066] R i = {R c,i , R w,i} (4)

[0067] Step S12: The network signal and GPS coordinates collected in step S11 are normalized using the minimum and maximum values ​​of RSSI and GPS coordinates. According to the minimum and maximum values ​​of RSSI in the cellular network, that is, and Define the jth base station at time t i The normalized rssi is c j,i ; According to the minimum and maximum values ​​of rssi in the Wi-Fi network, that is, and Define the kth Wi-Fi wireless access point in the cellular network at time t i The normalized rssi of k,i ; According to the minimum and maximum values ​​of longitude in GPS coordinates, that is, and Definition i The normalized longitude at the time is x i ; According to the minimum and maximum values ​​of latitude in GPS coordinates, that is, and Definition i The normalized latitude at the time is y i ;

[0068]

[0069]

[0070] Step S2: construct a recurrent neural network model, and train the recurrent neural network model through a training set to obtain a mobile positioning model;

[0071] In step S2, the recurrent neural network model is constructed using either a recurrent neural network with one timestamp or two timestamps. When a single time series is used as the input of the timestamp, a recurrent neural network with one timestamp is used; when two time series are used as the input of the timestamp, a recurrent neural network with two timestamps is used.

[0072] Among them, a timestamped recurrent neural network training method, such as Figure 3 As shown, the details are as follows:

[0073] S211. Based on the recurrent neural network consisting of an input layer, a recursive hidden layer and an output layer, the input layer is set to n1 base stations (c 1,i , c 2,i , …, and n2 Wi-Fi wireless access points (ω 1,i ,ω 2,i , …, , the output layer is the estimated normalized longitude and latitude and ; The recursive hidden layer contains at least one neuron, and the initial value of the neuron in the recursive hidden layer is defined as h0;

[0074] S212. Based on the above settings, after calculating the RSSI of the first timestamp, the value of the cyclic hidden layer neuron can be updated to h1; c j,i , w k,i The weights of h0 and h1 are α j , β k and v; output and The h1 weights are γ1 and γ2 respectively; the bias of neurons in the hidden layer and output layer is defined as b 1,1 , b 2,1 and b 3,1 ;

[0075] S213. Select the sigmoid function as the activation function of each neuron, then h0, h1, The value of can be calculated by equations (9) to (12) respectively, and the loss function E is defined according to the square of the error as equation (13);

[0076] h0=0(9)

[0077]

[0078]

[0079] S214. Optimize the recurrent neural network and use the learning rate η and gradient descent method to update each weight and bias, where γ1, γ2, b 2,1 , b 3,1 , αj , β k , v and b 1,1 The update of is proved and calculated by equations (14) to (21) respectively;

[0080]

[0081]

[0082] In step S211, when the recursive hidden layer includes more than one neuron, the weights between the neurons are updated by the gradient descent method.

[0083] Among them, the recurrent neural network training method with two timestamps, such as Figure 4 As shown, the details are as follows:

[0084] S221. Based on the recurrent neural network consisting of an input layer, a recursive hidden layer and an output layer, the input layer is set to the (n1+n2) standardized rssi (c 1,i , c 2,i , …, and ω 1,i ,ω 2,i , …, and (n1+n2) normalized rssi at the second timestamp (c 1,i+1 , c 2,i+1 , …, and ω 1,i+1 ,ω 2,i+1 , …, ; The output layer is the estimated normalized longitude and latitude at the second timestamp and The recursive hidden layer contains one neuron, and the initial value of the neuron in the recursive hidden layer is defined as h0, as shown in formula (9);

[0085] h0=0(9)

[0086] S222. Based on the above settings, the value of the cyclic hidden layer neuron is updated to h1 in the first timestamp and to h2 in the second timestamp; the weights of the j-th base station and the k-th Wi-Fi wireless access point P in each timestamp are α j and β k Output and The h2 weights are γ1 and γ2 respectively; the weight of the hidden layer neuron in the minimum timestamp loop is defined as v; the deviation of the hidden layer and output layer neurons is defined as b 1,1 , b 2,1 and b 3,1 ;

[0087] S223. Select the sigmoid function as the activation function of each neuron, so h1, h2, The values ​​of can be calculated by equations (22) to (25) respectively; according to the square of the error, the loss function E is defined as equation (26);

[0088]

[0089] S224. For the optimization of a recurrent neural network with two consecutive timestamps, the learning rate η and the gradient descent method are used to update each weight and bias; γ1, γ2, b 2,1 , b 3,1 , α j , β k , v and b 1,1 The update of is proved and calculated by equations (27) to (34) respectively;

[0090]

[0091]

[0092]

[0093] Step S3: input the test set into the mobile positioning model to obtain positioning data, and perform denormalization on the positioning data to obtain the position of the moving point.

[0094] Denormalization processing in step S3: For denormalization and estimation, the estimated normalized latitude and longitude are used in equations (35) and (36) respectively. and To retrieve the estimated latitude and longitude and

[0095]

[0096] in, is the minimum value of longitude in GPS coordinates, is the maximum value of longitude in GPS coordinates, is the minimum value of longitude in GPS coordinates, The maximum value of longitude in GPS coordinates.

[0097] The above is only a preferred embodiment of the present application and does not constitute any form of limitation to the present application. Although the present application has been disclosed as a preferred embodiment as above, it is not intended to limit the present application. Any technical personnel in this field can make some changes or modify the technical contents disclosed above into equivalent embodiments without departing from the scope of the technical solution of the present application. However, any brief modifications, equivalent changes and modifications made to the above embodiments based on the technical essence of the present application without departing from the content of the technical solution of the present application are still within the scope of the technical solution of the present application.

Claims

1. A mobile positioning method based on recurrent neural network, characterized in that: The steps include: Step S1. Build a mobile positioning system, collect data based on the mobile positioning system, normalize the collected data to obtain a data set, and divide the data set into a training set and a test set; Step S2. construct a recurrent neural network model, and train the recurrent neural network model through the training set to obtain a mobile positioning model; Step S3: Input the test set into the mobile positioning model to obtain positioning data, and perform denormalization on the positioning data to obtain the position of the moving point.

2. A mobile positioning method based on a recurrent neural network according to claim 1, characterized in that: The mobile positioning system includes a mobile station, a mobile positioning server, a data server and a model server; the mobile station is any one of a vehicle-mounted type, a portable type or a handheld type, and is provided with a GPS module and a signal receiving module, the GPS module is used to estimate the GPS coordinates of the mobile station, and the signal receiving module is used to collect network signals, the network signals are the RSSI of neighbor base stations and Wi-Fi wireless access points of heterogeneous networks, and the heterogeneous networks are cellular networks and Wi-Fi networks.

3. A mobile positioning method based on recurrent neural network according to claim 2, characterized in that: During the training phase, the mobile station collects GPS coordinates and network signals and sends the GPS coordinates and network signals to the mobile positioning server; during the execution phase, the mobile station sends the collected network signals to the mobile positioning server for position estimation.

4. The mobile positioning method based on recurrent neural network according to claim 2 is characterized in that: The data server stores the GPS coordinates and network signals from the mobile station through the mobile positioning server, and the GPS coordinates and network signals can be queried and used to train the recurrent neural network model; the model server saves the trained recurrent neural network model from the mobile positioning server for estimating the position of the mobile point.

5. The mobile positioning method based on recurrent neural network according to claim 1, characterized in that: The normalization process in step S1 uses the minimum and maximum values ​​of rssi and coordinates for normalization, as follows: Step S11. Collect network signals and GPS coordinates, and the mobile station i The RSSI from the cellular network can be detected at any time, that is, R in formula (1) c,i , the rssi of the Wi-Fi wireless access point, that is, R in formula (2) w,i and GPS coordinates, i.e. l ​​in formula (3) i , define the tth i The rssi of the jth base station from the cellular network at the moment is r c,j,i , defining the kth wireless access point from the Wi-Fi network at time t i The rssi is r w,k,i ; The network signal t i The rssi data set at the moment is defined as R i , as shown in formula (4); In addition, the position l i Including longitude x,i and latitude l y,i , assuming that a total of m locations are detected, n1 different base stations, and n2 different Wi-Fi wireless access points, if the RSSI of the network signal cannot be detected, it can be set to empty; l i ＝{l x,i ，l y,i} (3) R i ={R c,i ,R w,i } (4) Step S12: The network signal and GPS coordinates collected in step S11 are normalized using the minimum and maximum values ​​of RSSI and GPS coordinates. According to the minimum and maximum values ​​of RSSI in the cellular network, that is, and Define the jth base station at time t i The normalized rssi is c j,i ; According to the minimum and maximum values ​​of rssi in the Wi-Fi network, that is, and Define the kth Wi-Fi wireless access point in the cellular network at time t i The normalized rssi of k,i ; According to the minimum and maximum values ​​of longitude in GPS coordinates, that is, and Definition i The normalized longitude at the time is x i ; According to the minimum and maximum values ​​of latitude in GPS coordinates, that is, and Definition i The normalized latitude at the time is y i ; 6. The mobile positioning method based on recurrent neural network according to claim 1, characterized in that: In the step S2, the recurrent neural network model is constructed by using a recurrent neural network with one timestamp or two timestamps. When a single time series is used as the input of the timestamp, a recurrent neural network with one timestamp is used; when two time series are used as the input of the timestamp, a recurrent neural network with two timestamps is used.

7. A mobile positioning method based on recurrent neural network according to claim 6, characterized in that: The recurrent neural network training method for a timestamp is as follows: S211. Based on the recurrent neural network consisting of an input layer, a recursive hidden layer and an output layer, the input layer is set to n1 base stations and n2 Wi-Fi wireless access points The output layer is the estimated normalized longitude and latitude The recursive hidden layer contains at least one neuron, and the initial value of the neuron in the recursive hidden layer is defined as h0; S212. Based on the above settings, after calculating the RSSI of the first timestamp, the value of the cyclic hidden layer neuron can be updated to h1; c j,i , w k,i The weights of h0 and h1 are α j , β k and v; output and The h1 weights are γ1 and γ2 respectively; the bias of neurons in the hidden layer and output layer is defined as b 1,1 , b 2,1 and b 3,1 ; S213. Select the sigmoid function as the activation function of each neuron, then h0, h1, The value of can be calculated by equations (9) to (12) respectively, and the loss function E is defined according to the square of the error as equation (13); h0＝0(9) S214. Optimize the recurrent neural network and use the learning rate η and gradient descent method to update each weight and bias, where γ1, γ2, b 2,1 , b 3,1 , α j , β k , v and b 1,1 The update of is proved and calculated by equations (14) to (21) respectively; 8. The mobile positioning method based on recurrent neural network according to claim 7, characterized in that: In step S211, when the recursive hidden layer includes more than one neuron, the weights between the neurons are updated by the gradient descent method.

9. The mobile positioning method based on recurrent neural network according to claim 6, characterized in that: The recurrent neural network training method for the two timestamps is as follows: S221. Based on the recurrent neural network consisting of an input layer, a recursive hidden layer and an output layer, the input layer is set to be the (n1+n2) standardized and (n1+n2) normalized at the second timestamp The output layer is the estimated normalized longitude and latitude at the second timestamp. and The recursive hidden layer contains one neuron, and the initial value of the neuron in the recursive hidden layer is defined as h0, as shown in formula (9); h0＝0(9) S222. Based on the above settings, the value of the cyclic hidden layer neuron is updated to h1 in the first timestamp and to h2 in the second timestamp; the weights of the j-th base station and the k-th Wi-Fi wireless access point P in each timestamp are α j and β k Output and The h2 weights are γ1 and γ2 respectively; the weight of the hidden layer neuron in the minimum timestamp loop is defined as v; the deviation of the hidden layer and output layer neurons is defined as b 1,1 , b 2,1 and b 3,1 ; S223. Select the sigmoid function as the activation function of each neuron, so h1, h2, The values ​​of can be calculated by equations (22) to (25) respectively; according to the square of the error, the loss function E is defined as equation (26); S224. For the optimization of a recurrent neural network with two consecutive timestamps, the learning rate η and the gradient descent method are used to update each weight and bias; γ1, γ2, b 2,1 , b 3,1 , α j , β k , v and b 1,1 The update of is proved and calculated by equations (27) to (34) respectively; 10. The mobile positioning method based on recurrent neural network according to claim 1, characterized in that: In step S3, the denormalization process is as follows: For denormalization and estimation, the estimated normalized latitude and longitude are used in equations (35) and (36) respectively. and To retrieve the estimated latitude and longitude and in, is the minimum value of longitude in GPS coordinates, is the maximum value of longitude in GPS coordinates, is the minimum value of longitude in GPS coordinates, The maximum value of longitude in GPS coordinates.

Images