Beidou and UWB indoor and outdoor high-precision seamless positioning system based on BP neural network

Abstract

The invention discloses a Beidou and UWB indoor and outdoor high-precision seamless positioning system based on a BP neural network. The Beidou and UWB indoor and outdoor high-precision seamless positioning system comprises (1) an architecture module of the Beidou and UWB indoor and outdoor high-precision seamless positioning system, wherein the architecture module comprises a Beidou positioning base station, an UWB positioning system, a positioning terminal, upper computer monitoring software, a cloud server and the like; and (2) a BP neural network-based Beidou and UWB data fusion algorithm module which fuses Beidou BD and UWB data by using the BP neural network so as to realize accurate positioning of the area. The system is based on integrated application of technologies such as a Beidou GNNS, a UWB ultra wide band, an AIoT intelligent terminal, an embedded software technology and a BP neural network data fusion algorithm, and is applied to digital and intelligent management schemes in the fields of intelligent construction and the like, so that cost reduction, efficiency improvement and quality improvement are promoted.

Description

Technical Field

[0001] The patent of this invention relates to a high-precision seamless positioning technology for indoor and outdoor use, and in particular to a high-precision seamless positioning system for indoor and outdoor use based on a BP neural network that integrates Beidou navigation and UWB technology. Background Art

[0002] In the intelligent world of the Internet of Everything, it is meaningless without location-based services (LBS). Location-based services are also called positioning services, which are divided into outdoor positioning and indoor positioning. At present, both indoor and outdoor positioning have achieved phased results. Beidou RTK technology is a real-time high-precision positioning technology based on satellite positioning. The outdoor positioning accuracy of RTK technology can reach plus or minus 2 cm. Indoor positioning often has problems such as building obstruction, multipath effect, and signal interference due to complex spatial environments. UWB has an absolute advantage among many indoor positioning technologies. UWB is essentially an ultra-wideband (3.75~4.25Ghz frequency band) communication technology. Under general industrial environment NLOS non-line-of-sight conditions, the positioning accuracy can reach 5~20 mm.

[0003] Indoor and outdoor positioning technologies are currently in a state of independent development, lacking effective integration with high precision, especially in scenarios such as smart construction sites, which require seamless access and smooth switching between indoor and outdoor positioning areas. The positioning environment is generally divided into three areas: outdoor area, indoor and outdoor multiple positioning signal coexistence area, and indoor area. Figure 1 shown.

[0004] When the positioning area changes, each system needs to switch seamlessly to ensure a smooth positioning trajectory of the moving target and avoid frequent jumps in the positioning results. Beidou outdoor and UWB indoor positioning have high accuracy, but it is difficult to find a clear boundary to determine when to choose which positioning system for positioning. If weighted filtering and other methods are used for combined positioning, it is difficult to find a perfect mathematical model to determine the corresponding weights of the positioning results of the two positioning systems at this moment. Summary of the invention

[0005] Based on the integrated application of Beidou GNNS, UWB ultra-wideband, AIoT smart terminals, embedded software technology, and BP neural network data fusion algorithm, this project proposes to develop a Beidou and UWB high-precision (millimeter level) indoor and outdoor seamless positioning system based on BP neural network, which will be applied in digital management solutions in fields such as intelligent construction to promote cost reduction, efficiency improvement, and quality improvement.

[0006] A BeiDou and UWB indoor and outdoor high-precision seamless positioning system based on BP neural network, comprising: (1) an architecture module of the BeiDou and UWB indoor and outdoor high-precision seamless positioning system;

[0007] The architecture module includes Beidou positioning base station, ultra-wideband positioning system, AIoT positioning terminal, host computer monitoring software, and cloud server;

[0008] Beidou positioning base station, obtain Beidou outdoor positioning and UWB indoor positioning; Beidou outdoor positioning adopts RTK differential positioning, acquired through BD integrated chip and algorithm, to achieve high-speed stability, automatic identification, and long-distance outdoor data collection and transmission; UWB indoor positioning adopts UWB ultra-wideband positioning system, deploys four master-slave base stations to work together, and obtains indoor location information based on TDOA arrival time difference method;

[0009] AIoT positioning terminal realizes seamless access and smooth switching of indoor and outdoor positioning areas; AIoT positioning terminal includes UWB transceiver, BD positioning module, wireless communication module, main control unit, key module, display module. The software of positioning terminal is divided into four layers, namely driver layer, peripheral layer, algorithm layer and application layer. The interdependence between layers is from low to high, and the direct coupling between layers and modules is low.

[0010] The host computer software realizes the monitoring and display of indoor and outdoor and handover area positioning data, personnel base station information statistics, data storage and historical data viewing functions;

[0011] The cloud server deploys and runs the host computer monitoring software. At the same time, it designs the cloud-edge collaboration mechanism and load balancing scheduling strategy to suppress the ping-pong effect of the indoor and outdoor area positioning algorithm, and ensure the real-time data collection of the system and the collaboration of business management.

[0012] (2) The Beidou and UWB data fusion algorithm module based on BP neural network uses BP neural network to fuse the Beidou BD and UWB data to achieve accurate positioning in the area. BRIEF DESCRIPTION OF THE DRAWINGS

[0013] Figure 1 It is a map of outdoor positioning area division.

[0014] Figure 2 It is a high-precision seamless positioning system architecture based on Beidou and UWB indoor and outdoor.

[0015] Figure 3 It is a BP neural network model.

[0016] Figure 4 It is the DB and UWB data fusion algorithm model.

[0017] Figure 5 It is a seamless positioning of BD and UWB data based on the BP neural network structure model.

[0018] Figure 6It is the optimal weight and threshold handover criterion design.

[0019] Figure 7 This is the block diagram of the AIoT terminal hardware system.

[0020] Figure 8 It is the software architecture of embedded positioning terminal.

[0021] Fig. 9 It is a data flow diagram of DB and UWB positioning based on BP neural network.

[0022] Fig.10 It is the upper computer software design framework.

[0023] Fig.11 It is the flow chart of upper computer software design.

[0024] Fig.12 It is to optimize the weight and bias parameters of the BP neural network.

[0025] Fig.13 It is the partial positioning measurement data of the experimental site.

[0026] Fig.14 It is a typical application of the smart construction site system based on Beidou and UWB high-precision seamless indoor and outdoor positioning. DETAILED DESCRIPTION

[0027] The present invention will be further described below with reference to the accompanying drawings and implementations.

[0028] like Figure 2 As shown, the high-precision seamless indoor and outdoor positioning system based on Beidou and UWB includes (1) the architecture module of the high-precision seamless indoor and outdoor positioning system based on Beidou and UWB: the architecture module includes Beidou positioning base station, ultra-wideband positioning system, positioning terminal, host computer monitoring software, cloud server, etc.

[0029] In order to maximize the positioning accuracy, in the networking solution of this system, the outdoor positioning part adopts the RTK differential positioning method, and uses the Zhonghaida VNet series and other brand models to set up Beidou base stations to achieve high-speed stability, automatic identification, and long-distance outdoor data collection and transmission. The indoor positioning adopts the UWB ultra-wideband positioning system, and four master-slave base stations are deployed to work together to obtain indoor location information based on the TDOA arrival time difference method. In order to ensure the accuracy of the ranging between the UWB base station and the tag, the UWB base station should try to avoid the antenna facing the metal at close range during the installation process to prevent the occurrence of multipath effects. In order to expand the positioning range, generally speaking, in a four-base station positioning system, one base station should be 0.5 to 1m higher than the other three base stations, and the other three base stations are at the same height.

[0030] AIoT positioning terminal is a key device to achieve seamless access and smooth switching of indoor and outdoor positioning areas. The terminal hardware system mainly includes UWB transceiver, BD positioning module, software module, wireless communication module, main control unit, key module, display module, etc. The software design adopts layered thinking. The software module of the positioning terminal divides the software into four layers, namely the driver layer, peripheral layer, algorithm layer and application layer. The interdependence between layers is from low to high, and the direct coupling between layers and modules is low.

[0031] The host computer software is mainly responsible for the monitoring and display of indoor and outdoor and handover area positioning data, personnel base station information statistics, data preservation and viewing of historical data, etc., and includes modules such as the main interface, map display area, configuration information bar and personalized settings.

[0032] After the cloud server is set up, the host computer monitoring software is deployed and run. At the same time, the cloud-edge coordination mechanism and load balancing scheduling strategy are designed to suppress the ping-pong effect of the indoor and outdoor interface positioning algorithm, and ensure the real-time data collection of the system and the coordination of business management.

[0033] Design of data fusion algorithm for Beidou and UWB based on BP neural network. Artificial neural network is a mathematical model that simulates the human nervous system. This mathematical model exists in the form of a network. It can divide complex problems and make them divide and conquer. In theory, it can achieve the effect of approximating any function, and can effectively avoid complex mathematical models and make complex problems simple.

[0034] BP neural network is a common neural network model. This model uses error back propagation to train data. It consists of an input layer, a hidden layer, and an output layer. The network structure is as follows: Figure 3 As shown in the figure, after years of development and research, this technology has been widely used in various fields. Each layer is composed of multiple neurons, where the number of input and output layers is 1, and the number of hidden layers is generally one or more, depending on the actual training results.

[0035] According to the basic theory of neural networks, in theory, a three-layer neural network can approximate any continuous function. It can effectively simplify complex problems, and the biggest difference from ordinary curve fitting is that it does not require a complex mathematical model for approximation. It provides ideas for positioning the preparation area when the positioning system switches. Therefore, the BP neural network is used to fuse the Beidou BD and UWB data to achieve accurate positioning of the area. The algorithm flow is as follows: Figure 4 shown.

[0036] Since this algorithm is ultimately concerned with the location information of the target to be measured, the input of the algorithm is easy to determine. Using the indoor coordinate system as the reference coordinate system, the output of the neural network is the two-dimensional coordinates (x, y) of the target to be measured. In order to achieve higher positioning accuracy and better expression of the neural network, the algorithm must reasonably select input neurons. The UWB data are the distance observation values ​​of each base station [d1d2d3d4], and the BD data consists of the longitude and latitude of the differential positioning results and the positioning status indication flag state, that is, [lon lat state]. Research and a large number of experimental verifications have shown that although increasing the number of neural network layers will reduce the number of nodes in each layer, it will complicate the problem. Therefore, this case uses a three-layer neural network structure to perform combined positioning of UWB and BD. Combining the characteristics of BD and UWB positioning signals and the BP neural network structure, a network structure model for seamless indoor and outdoor positioning is established, such as Figure 5 shown.

[0037] The BeiDou (BDS) and ultra-wideband (UWB) data fusion algorithm based on BP neural network mainly includes data preprocessing, feature extraction, neural network structure design, model training and verification, and result output. The following is a detailed algorithm design process:

[0038] 1. Data preprocessing

[0039] Data collection: Collect raw data of BeiDou (BDS) and Ultra-Wideband (UWB). BDS data usually includes longitude, latitude, altitude and timestamp, and UWB data includes distance, direction and timestamp.

[0040] Data cleaning: remove noisy data and invalid data, such as missing values, outliers, etc.

[0041] Data synchronization: Synchronize the BDS and UWB data according to the timestamp to ensure that each data point is the observation value at the same time.

[0042] Assume that we have two datasets: BeiDou dataset B and UWB dataset U, each of which contains a timestamp t and the corresponding observation. The goal of data synchronization is to find the closest matching pair of timestamps in the two datasets.

[0043] Timestamp matching: For each t i ∈B, find t j ∈U, so that |t i -t j ∣Minimum.

[0044] Data normalization: In order to speed up training and improve model performance, the data is normalized and scaled to the [0,1] or [-1,1] interval. The normalization formula is expressed as:

[0045]

[0046] Where x is the original data, x norm is the normalized data, x min and x max are the minimum and maximum values ​​in the data set, respectively.

[0047] 2. Feature extraction

[0048] BDS features: extract longitude, latitude, altitude, etc. as features.

[0049] UWB features: extract distance, direction, etc. as features.

[0050] Fusion features: Combine the features of BDS and UWB into a feature vector as the input of the BP neural network.

[0051] Assume that the BeiDou dataset contains longitude lon, latitude lat and altitude alt, and the UWB dataset contains distance d and direction θ. The feature vector f can be expressed as: f = [lon, lat, alt, d, θ]

[0052] 3. BP neural network structure design

[0053] Input layer: The number of nodes in the input layer is equal to the dimension of the feature vector, which is the sum of BDS and UWB features.

[0054] Hidden layer: Design one or more hidden layers. The number of nodes in each layer can be determined based on experience or experiments. Usually, the number of nodes in the first hidden layer can be set to 1.5 to 2 times the number of nodes in the input layer.

[0055] Output layer: The number of nodes in the output layer is determined according to the specific task. For example, if it is a positioning task, the output layer can be set to 2 nodes (longitude and latitude).

[0056] Activation function: commonly includes Sigmoid, Tanh and ReLU. Select the ReLU activation function, which is defined as: ReLU(x) = max(0,x)

[0057] Loss function: Choose a loss function such as mean squared error (MSE) or cross entropy to measure the difference between the model output and the actual value.

[0058] Where N is the number of samples, y i is the true value, y^ i is the predicted value.

[0059] 4. Model training and validation

[0060] Dataset division: Divide the dataset into training set, validation set and test set. Usually 70% is used for training, 15% for validation and 15% for testing.

[0061] Optimizer: Select an optimizer such as Adam or SGD to update the weights of the model.

[0062] Training process: Use the training set data to train the model and update the weights through the back-propagation algorithm until the set number of training rounds is reached or the loss function converges.

[0063] Gradient descent: In BP neural networks, weight updates are usually implemented using a gradient descent algorithm. The weight update formula is:

[0064] Where η is the learning rate, w ij is the weight from neuron i to neuron j.

[0065] Back propagation: Calculate the gradient of the loss function to the weights through the chain rule. For the hidden layer to the output layer

[0066] The weight of , the gradient is calculated as:

[0067] where δ o is the error term of the output layer, σ′ is the derivative of the activation function, z o is the input of the output layer, a h is the activation value of the hidden layer.

[0068] δ h =(∑ o δ o ·w ho )·σ′(z h )

[0069] For the weights from the input layer to the hidden layer, the gradient is calculated as:

[0070] where δ h is the error term of the hidden layer, x i is the value of the input layer.

[0071] Validation process: Evaluate the performance of the model on the validation set and select the best model parameters.

[0072] 5. Result output and performance evaluation

[0073] Model testing: Evaluate the performance of the final model on the test set and output positioning results or other target values.

[0074] Performance evaluation: Use metrics such as precision, recall, and F1 score to evaluate the performance of the model.

[0075] Data fusion results: According to the output of the BP neural network, the fused BDS and UWB data results are obtained.

[0076] Establish an indoor and outdoor seamless positioning algorithm model based on BP neural network, design activation function and neural network structure, and import the collected samples into the docking system for model training and verification. At the same time, combine the Beidou RTK phase difference algorithm and the UWB three-sphere intersection method to optimize the BP neural network model. Through this process, determine the optimal weight and threshold of the cross-zone handover to ensure the accuracy and applicability of the model; finally, apply the trained BP neural network to indoor and outdoor seamless positioning, and realize high-precision seamless positioning application based on Beidou and UWB.

[0077] Optimal weight and threshold handover criterion design, such as Figure 6 As shown. At the interface between indoor and outdoor, there is a handover of the positioning system. In order to achieve smooth handover, the relative signal strength criterion based on the threshold value is adopted. When transitioning from outdoor to indoor, when the number of available Beidou satellites with signal strength greater than the positioning threshold value is less than 4 and the number of UWB base stations with signal strength greater than the positioning threshold value reaches 4, the system is switched, otherwise it will not be switched; when transitioning from indoor to outdoor, when the number of UWB base stations with signal strength greater than the positioning threshold value is less than 4 and the number of available Beidou satellites with signal strength greater than the positioning threshold value reaches 4, the system is switched, otherwise it will not be switched. This handover criterion avoids positioning drift or even positioning loss caused by unnecessary frequent switching.

[0078] A highly integrated, low-power AIoT positioning terminal based on BeiDou-3. In order to achieve seamless access and smooth switching between indoor and outdoor positioning handover areas, the AIoT positioning terminal is specially developed for BeiDou + UWB seamless data transmission based on the BeiDou-3 signal system's BeiDou DB503D board, UWB module DWM1000, STM32 main control chip and other circuit devices. It adopts an antenna-host integrated design, integrating RDSS antenna, RF transceiver circuit, power amplifier circuit, baseband circuit, etc., with high integration and low power consumption. The preliminary design of the AIoT terminal hardware system block diagram is as follows: Figure 7 shown.

[0079] The embedded software of the positioning terminal in the indoor and outdoor seamless positioning system adopts a layered design concept and is divided into four layers: driver layer, peripheral layer, algorithm layer and application layer. The driver layer includes interface driver modules of SPI, UART, USB and IIC. The peripheral layer includes modules such as OLED display, buttons, NB-IOT, DB501D and DW1000. The algorithm layer includes ranging, positioning area discrimination, Taylor, BP neural network fusion algorithm; the application layer implements different business function applications through different algorithm functions. The interdependence between layers is from low to high, and the direct coupling between layers and modules is low. The software architecture diagram is as follows: Figure 8 shown.

[0080] Beidou outdoor positioning data is obtained through BD integrated chip and algorithm, and indoor positioning data is obtained from UWB ultra-wideband system. In this design, it is mainly used to locate the target to be measured and publish the position data. The software part can be mainly divided into three parts: Beidou positioning data reading, UWB ranging positioning and data publishing. The data flow in the positioning terminal software is as follows: Fig. 9 shown.

[0081] The design and development of the host computer software must have good cross-platform capabilities, and can develop both Linux and Windows applications. The host computer software is mainly responsible for monitoring and displaying indoor and outdoor and handover area positioning data, personnel base station information statistics, data storage and viewing historical data, etc. It includes the main interface, map display area, configuration information bar and personalized settings. The design framework is as follows Fig.10 shown.

[0082] The host computer software designed in this system is developed with a modular concept and uses the signal and slot mechanism to realize the interaction between various components, making software development more intuitive and efficient. The software design flow chart is as follows: Fig.11 shown.

[0083] System verification and accuracy test report, in the BP neural network, the Beidou RTK phase difference algorithm and the UWB three-sphere intersection method are combined to optimize the BP neural network model. The optimal weight and threshold of the handover are determined by adjusting the weight and bias of the network to reduce the error, improve the classification accuracy or accelerate the network convergence. The experimental results of optimizing the BP neural network are as follows Fig.12 shown.

[0084] Based on Beidou and UWB high-precision seamless indoor and outdoor positioning, the outdoor terminal is positioned by Beidou RTK, and the indoor terminal is positioned by UWB. Several UWB base stations need to be installed in the indoor environment, and the installation quantity is designed according to the specific site to ensure the positioning effect.

[0085] Depend on Fig.13From the test data, we can see that the indoor positioning accuracy is mostly around 10cm. When the indoor environment is open and there are fewer obstacles, the positioning accuracy is higher, reaching about 5cm. When the positioning signal is blocked in the transition area between indoor and outdoor, the positioning accuracy can still be guaranteed to be around 15cm, which can meet the needs of high-precision positioning. Most positioning points in the test can be accurately located, but the positioning error of individual points is large, which may be caused by reasons such as severe blocking of the positioning signal. Overall, the positioning results show that the positioning system can locate in real time with high positioning accuracy. The trajectory is smooth when transitioning from indoor to outdoor, and it can achieve seamless connection between indoor and outdoor.

[0086] See Fig.14 , a typical application of the smart construction site system based on Beidou and UWB indoor and outdoor high-precision seamless positioning. The smart construction site industry on the market provides products with coarse positioning services at the decimeter or meter level, and the indoor and outdoor docking positioning areas cannot be automatically switched and seamlessly accessed, lacking accurate and reliable real-time control data sources.

Claims

1. Beidou and UWB indoor and outdoor high-precision seamless positioning system based on BP neural network, characterized by: include: (1) Architecture modules of the high-precision seamless indoor and outdoor positioning system based on Beidou and UWB: The architecture module includes Beidou positioning base station, ultra-wideband positioning system, AIoT positioning terminal, host computer monitoring software, and cloud server; Beidou positioning base station, obtain Beidou outdoor positioning and UWB indoor positioning; Beidou outdoor positioning adopts RTK differential positioning, acquired through BD integrated chip and algorithm, to achieve high-speed stability, automatic identification, and long-distance outdoor data collection and transmission; UWB indoor positioning adopts UWB ultra-wideband positioning system, deploys four master-slave base stations to work together, and obtains indoor location information based on TDOA arrival time difference method; Establish an indoor and outdoor seamless positioning algorithm model based on BP neural network, set the activation function and neural network structure, and import the collected samples into the docking system for model training and verification; at the same time, combine the Beidou RTK phase difference algorithm and the UWB three-sphere intersection method to optimize the BP neural network model, and determine the optimal weight and threshold of the handover through this process to ensure the accuracy and applicability of the model; finally, apply the trained BP neural network to indoor and outdoor seamless positioning, and realize high-precision seamless positioning application based on Beidou and UWB; AIoT positioning terminal realizes seamless access and smooth switching of indoor and outdoor positioning areas; the software modules of AIoT positioning terminal are divided into four layers, namely driver layer, peripheral layer, algorithm layer and application layer. The interdependence between layers is from low to high, and the direct coupling between layers and modules is low; The host computer software realizes the monitoring and display of indoor and outdoor and handover area positioning data, personnel base station information statistics, data storage and historical data viewing functions; The cloud server deploys and runs the host computer monitoring software. It also has a cloud-edge collaboration mechanism and a load balancing scheduling strategy to suppress the ping-pong effect of the indoor and outdoor area positioning algorithm, ensuring the real-time data collection of the system and the collaboration of business management. (2) The Beidou and UWB data fusion algorithm module based on the BP neural network adopts the input layer, hidden layer and output layer neural network structure to perform UWB and BD combined positioning. Combining the characteristics of BD and UWB positioning signals and the BP neural network structure, a network structure model for seamless indoor and outdoor positioning is established to achieve accurate positioning in the area.

2. The Beidou and UWB indoor and outdoor high-precision seamless positioning system based on BP neural network according to claim 1 is characterized in that: The driver layer includes interface driver modules of SPI, UART, USB and IIC; the peripheral layer includes OLED display, buttons, NB-IOT, DB501D and DW1000 modules; the algorithm layer includes ranging, positioning area discrimination, Taylor and BP neural network fusion algorithms; the application layer realizes different business function applications through different algorithm functions.

3. The Beidou and UWB indoor and outdoor high-precision seamless positioning system based on BP neural network according to claim 1 is characterized in that: Design of optimal weight and threshold handover criteria. At the interface between indoor and outdoor areas, there is handover of the positioning system. In order to achieve smooth handover, the relative signal strength criterion based on the threshold is adopted. When transitioning from outdoor to indoor, if the number of available Beidou satellites with signal strength greater than the positioning threshold is less than 4 and the number of UWB base stations with signal strength greater than the positioning threshold reaches 4, the system will be switched, otherwise it will not be switched; When transitioning from indoors to outdoors, if there are less than 4 UWB base stations with signal strength greater than the positioning threshold and the number of available Beidou satellites with signal strength greater than the positioning threshold reaches 4, the system will be switched, otherwise it will not be switched; This handover criterion is used to avoid positioning drift or even positioning loss caused by unnecessary frequent handover.

4. The Beidou and UWB indoor and outdoor high-precision seamless positioning system based on BP neural network according to claim 1 is characterized in that: In the Beidou and ultra-wideband data fusion algorithm module based on BP neural network, this system is used to locate the target to be measured and publish the position data. The software part is divided into three parts: Beidou positioning data reading, UWB ranging positioning and data publishing.

5. The Beidou and UWB indoor and outdoor high-precision seamless positioning system based on BP neural network according to claim 1 is characterized in that: The Beidou and ultra-wideband data fusion algorithm module based on the BP neural network includes the following steps: data preprocessing, feature extraction, neural network structure design, model training and verification, and result output.

6. The Beidou and UWB indoor and outdoor high-precision seamless positioning system based on BP neural network according to claim 5 is characterized by: The data preprocessing includes: Data collection: Collect the original data of Beidou and ultra-wideband. BDS data includes longitude, latitude, altitude and timestamp, and UWB data includes distance, direction and timestamp. Data cleaning: remove noise data and invalid data; Data synchronization: synchronize Beidou and UWB data according to timestamps to ensure that each data point is an observation value at the same time; The BeiDou dataset B and the ultra-wideband dataset U are used, where each dataset contains a timestamp t and the corresponding observation value; the goal of data synchronization is to find the matching pair with the closest timestamps in the two datasets; Timestamp matching: For each t i ∈B, find t j ∈U, so that |t i -t j ∣ minimum; Data normalization: In order to speed up training and improve model performance, the data is normalized and scaled to the [0,1] or [-1,1] interval. The normalization formula is expressed as: Where x is the original data, x norm is the normalized data, x min and x max are the minimum and maximum values ​​in the data set, respectively.

7. The Beidou and UWB indoor and outdoor high-precision seamless positioning system based on BP neural network according to claim 5 is characterized by: The feature extraction comprises: BDS features: extract longitude, latitude, and altitude as features; UWB features: extract distance and direction as features; Fusion features: Combine the features of Beidou and UWB into a feature vector as the input of BP neural network; The Beidou data set contains longitude lon, latitude lat and altitude alt, and the ultra-wideband data set contains distance d and direction θ; the feature vector f can be expressed as: f = [lon, lat, alt, d, θ].

8. The Beidou and UWB indoor and outdoor high-precision seamless positioning system based on BP neural network according to claim 5 is characterized by: The BP neural network structure design includes: Input layer: The number of nodes in the input layer is equal to the dimension of the feature vector, which is the sum of BDS and UWB features; Hidden layer: Design one or more hidden layers. The number of nodes in each layer can be determined based on experience or experiments. Usually, the number of nodes in the first hidden layer can be set to 1.5 to 2 times the number of nodes in the input layer. Output layer: The number of nodes in the output layer is determined according to the specific task; Activation function: including Sigmoid, Tanh and ReLU; select ReLU activation function, which is defined as: ReLU(x) = max(0,x); Loss function: Choose the mean squared error (MSE) or cross entropy loss function to measure the difference between the model output and the actual value; Where N is the number of samples, y i is the true value, y^ i is the predicted value.

9. The Beidou and UWB indoor and outdoor high-precision seamless positioning system based on BP neural network according to claim 5 is characterized by: The model training and verification includes: Dataset division: Divide the dataset into training set, validation set and test set; 70% for training, 15% for validation and 15% for testing; Optimizer: Select Adam or SGD optimizer to update the weights of the model; Training process: Use the training set data to train the model and update the weights through the back-propagation algorithm until the set number of training rounds is reached or the loss function converges; Gradient descent: In BP neural networks, weight updates are usually implemented using a gradient descent algorithm; the weight update formula is: Where η is the learning rate, w ij is the weight from neuron i to neuron j; Back propagation: The gradient of the loss function to the weight is calculated by the chain rule; for the weights from the hidden layer to the output layer, the gradient is calculated as: where δ o is the error term of the output layer, σ′ is the derivative of the activation function, z o is the input of the output layer, a h is the activation value of the hidden layer; For the weights from the input layer to the hidden layer, the gradient is calculated as: d h =(∑ o d o ·w ho )·σ′(z h ) where δ h is the error term of the hidden layer, x i is the value of the input layer; Validation process: Evaluate the performance of the model on the validation set and select the best model parameters.

10. The Beidou and UWB indoor and outdoor high-precision seamless positioning system based on BP neural network according to claim 5 is characterized by: The result output and performance evaluation include: Model testing: Evaluate the performance of the final model on the test set and output positioning results or other target values; Performance evaluation: Use precision, recall, and F1 score metrics to evaluate the performance of the model; Data fusion results: According to the output of the BP neural network, the fused BDS and UWB data results are obtained.

Images