AI-based geographic information precise navigation method and system

Abstract

The invention discloses a geographic information precise navigation method and system based on AI, and relates to the technical field of satellite sensing measurement. Performing branch processing on a time feature set and a space feature set of sliding selection based on a hierarchical LSTM structure, constructing a dynamic time delay observation correction model, fusing spatial and temporal features into the dynamic time delay observation correction model, and embedding physical constraints; and the fusion algorithm maintains the positioning precision, and the extreme weather navigation positioning is optimized. Four-level redundant tight coupling of a satellite navigation GNSS, a ground IMU, a visual sensor and a ground sensor is adopted, a dynamic time delay observation correction model and the satellite navigation GNSS are fused, the problem of long-term drifting of the ground IMU in extreme weather is solved, deep fusion with a physical model and a tight coupling navigation algorithm is achieved, sub-meter real-time positioning precision in extreme weather is achieved, and the real-time positioning precision of the satellite navigation GNSS is improved. Multi-source data deep collaboration, dynamic environment adaptive modeling and computing resource optimization are realized, and a systematic solution path is provided for high-precision positioning in extreme weather.

Description

Technical Field

[0001] The present invention relates to the technical field of satellite sensing measurement, and particularly to a precise navigation method and system for geographic information based on AI. Background Art

[0002] With the popularization of smart phones and mobile devices and the development of artificial intelligence technology, precise navigation methods and systems for geographic information based on AI have become the forefront development direction of modern navigation technology. Traditional navigation systems mainly rely on the global positioning system and inertial navigation system. Although these systems provide basic positioning services, in some specific environments, such as areas with dense high-rise buildings, underground environments, adverse weather conditions, and areas where GPS signals are interfered, their accuracy and reliability are often greatly affected. AI technologies, especially deep learning, machine learning, image recognition, sensor fusion, and big data analysis, can significantly improve the accuracy, robustness, and real-time performance of geographic information navigation systems. The navigation method based on AI can effectively overcome the deficiencies in traditional navigation systems by integrating multi-source data and combining intelligent algorithms, and provide more precise, real-time, and reliable navigation services.

[0003] Currently, the Chinese invention patent with the application number CN202211648354.8 discloses a precise navigation method and system, which generates a precise navigation path information according to the user's needs and actual situation to provide parking navigation services for users; combines the destination location that the user actually needs to reach, and generates a detailed and precise three-dimensional navigation information based on three-dimensional data to provide precise destination navigation for users, avoiding poor signals, and based on the actual environment and user needs, thus achieving efficient and precise navigation; however, the existing technology uses single satellite navigation, static navigation models, fixed parameter ground correction, and cannot improve the navigation accuracy in extreme weather, cannot perform in-depth cooperation of multi-source data, dynamic environment adaptive modeling, and computing resource optimization, and cannot provide a systematic solution path for high-precision positioning in extreme weather. Summary of the Invention

[0004] The technical problem solved by the present invention is that the existing technology uses single satellite navigation, static navigation models, fixed parameter ground correction, and cannot improve the navigation accuracy in extreme weather, cannot perform in-depth cooperation of multi-source data, dynamic environment adaptive modeling, and computing resource optimization, and cannot provide a systematic solution path for high-precision positioning in extreme weather.

[0005] To solve the above technical problem, the present invention provides the following technical solution: A precise navigation method for geographic information based on AI, comprising the following steps:

[0006] Step S1: Obtain data through sensors and establish a sensor database;

[0007] Step S2: Based on the sensor database, perform branch processing on the time feature set and space feature set selected by sliding based on the hierarchical LSTM structure, construct a dynamic time-delay observation correction model, fuse the spatio-temporal features into the dynamic time-delay observation correction model and embed physical constraints;

[0008] Step S3: Maintain the positioning accuracy through a fusion algorithm for the dynamic time-delay observation correction model and optimize the navigation positioning in extreme weather.

[0009] Preferably, the sensors include satellite navigation GNSS, fiber Bragg grating (FBG) strain sensors, ground IMU, and vision systems. A low-cost dual-antenna satellite navigation GNSS receiver is used to obtain the signal strength and original observations of the L1 or L2 frequency band in real time. An FBG strain sensor network is deployed to monitor the micro-deformation of the ground surface. The continuous operation data of the ground IMU is collected. The continuous operation data includes the data of the accelerometer, gyroscope, and temperature sensor, including the drift characteristic data under normal conditions and extreme weather. The extreme weather includes heavy rain and high temperature. The integrated ground IMU is used as an inertial measurement unit to track the dynamic attitude of the target object. A binocular vision system is used to obtain high-resolution terrain matching data;

[0010] Maintaining the time synchronization and space alignment of the data obtained by the sensors includes:

[0011] Through a unified clock system and data synchronization mechanism, accurately align the data obtained from each slave sensor in time. Use the Network Time Protocol (NTP) to perform the operation of synchronizing the clocks of the sensors that do not have the direct ability to receive satellite navigation GNSS signals. Give all sensors a unified trigger signal to unify the data acquisition moments of all sensors to the same time point;

[0012] Add timestamps to the data collected by each sensor respectively. The logic of adding timestamps includes: giving a timestamp to each obtained data record respectively. The timestamp comes from the same clock source, and record the spatio-temporal reference, accuracy index, and real-time working state of the sensor corresponding to each data. The spatio-temporal reference includes the reference coordinate system and the timestamp. The accuracy index includes the positioning accuracy and the sensor sensitivity. The real-time working state includes the configuration information, operation state, health state, and performance index of the sensor;

[0013] Through the spatial registration algorithm between sensors, perform spatial alignment on the positioning data of satellite navigation GNSS, vision data, ground IMU data, and FBG sensor data. The spatial alignment includes making the reference coordinates of the positioning data of satellite navigation GNSS, vision data, ground IMU data, and FBG sensor data fall on the same reference coordinate system;

[0014] Optimize and fuse the time - synchronized and spatially aligned data using Kalman filtering, and integrate the optimized and fused data set into a sensor database.

[0015] Preferably, the step S2 includes the following sub - steps:

[0016] Step S21: Using the TOPSIS multi - criterion decision - making algorithm, select initial correction factors from the sensor database based on data timeliness, spatial resolution, and reliability. Use a sliding window to screen the initial correction factors to obtain optimal correction factors, and divide the initial correction factors into a time feature set and a spatial feature set;

[0017] Step S22: Based on the sensor database, calculate the prior dry delay and wet delay using the Saastamoinen model;

[0018] Step S23: Use an ID convolutional layer to extract local meteorological gradients and construct a dynamic time - delay observation correction model.

[0019] Preferably, the step S21 includes the following sub - steps:

[0020] Step S211: Use the TOPSIS multi - criterion decision - making algorithm to select initial correction factors from the sensor database;

[0021] Step S212: Calculate the time delay and update time interval of the initial correction factors, measure the standard deviation of the data update time interval, and use the standard deviation as an evaluation index for timeliness;

[0022] Step S213: Extract the time feature set and the spatial feature set according to the sliding window, and construct a decision matrix including:

[0023] Obtain the time feature set based on a 5 - minute interval and an overall 1 - hour sliding window, and obtain the spatial feature set based on a 10 - minute interval and an overall 1 - hour sliding window. The performance of each correction factor on each feature will be used as a row of the decision matrix, and the features will be used as columns. The features are random eigenvalue in the time feature set and the spatial feature set. The time feature set includes data timeliness, and the spatial feature set includes spatial resolution and reliability;

[0024] Step S214: Normalize the decision matrix, and respectively obtain the positive ideal solution and the negative ideal solution according to the normalized decision matrix. The positive ideal solution is the maximum value of each column, and the negative ideal solution is the minimum value of each column;

[0025] Step S215: Calculate the Euclidean distances between each correction factor and the positive ideal solution and the negative ideal solution respectively. Calculate the relative closeness between each correction factor and the positive ideal solution according to the Euclidean distance, and select the correction factor with the largest relative closeness as the optimal correction factor. Discard the numerical nodes corresponding to the features of the non-optimal correction factor from the time feature set and the space feature set to obtain the final time feature set and space feature set.

[0026] Preferably, step S23 includes the following sub-steps:

[0027] Step S231: Use the ID convolutional layer to extract the local meteorological gradient, where the local meteorological gradient includes the temperature spatial change rate, and branch-process the time feature set and the space feature set based on the hierarchical LSTM structure;

[0028] Step S232: Construct a dynamic time-delay observation correction model, and perform adversarial training on the dynamic time-delay observation correction model using gradient penalty.

[0029] Preferably, step S231 includes:

[0030] Process the time feature set and the space feature set through the first layer of the hierarchical LSTM structure for high-frequency interval data to capture turbulent fluctuations, input the captured turbulent fluctuations into the second layer of the hierarchical LSTM structure for intermediate-frequency aggregation processing to extract the eigenvalues of the weather system evolution, and input the eigenvalues of the weather system evolution into the third layer of the hierarchical LSTM structure for low-frequency learning to learn the seasonal pattern features with a daily unit cycle component.

[0031] Preferably, step S232 includes:

[0032] Take the wet delay output by Saastamoinen as the prior value, adaptively learn the residual term of the wet delay using a neural network, take the atmospheric thermodynamic equation constraint as the physical loss function, and add MC-Dropout variational inference to the output layer to predict the mean and standard deviation of the wet delay;

[0033] Embed the wet delay predicted by the LSTM into the GNSS pseudorange equation to obtain a tightly coupled observation equation, and its mathematical expression is:

[0034]

[0035] where P is the GNSS pseudorange coordinate, ρ is the received signal distance, c is the speed of light, δt is the delay correction error between the sensor and the original coordinate point of the current coordinate system, and the delay correction error between the sensor and the original coordinate point of the current coordinate system is the error value between the initial value and the predicted value of the coordinate parameter calculated by the least squares method iteration, T dry is the dry delay, The wet delay is predicted by LSTM, and ε is noise;

[0036] Set the weights of the dynamic time-delay observation correction model to be calculated by a dynamic uncertainty formula, and its mathematical expression is:

[0037]

[0038] where γ is the weight, is the standard deviation of the wet delay, and σ GNSS is the observation weight of satellite navigation GNSS.

[0039] Preferably, according to the step S21, set the sliding window to an overall 1-hour sliding window with a 10-minute interval, and extract the initial correction factor from the continuous operation data and dynamic attitude obtained by the ground IMU from the sensor database to obtain the temperature drift amount correlation feature quantity and the angular velocity integral error accumulation rate feature quantity;

[0040] Add the Euler dynamics equation as a regular term to the loss function of the dynamic time-delay observation correction model;

[0041] Take the difference between the predicted values of the output temperature drift amount correlation feature quantity and the angular velocity integral error accumulation rate feature quantity and the temperature drift amount correlation feature quantity and the angular velocity integral error accumulation rate feature quantity calculated per actual unit time as the prior error, and establish a ground IMU error state vector. The ground IMU error state vector is obtained by the convolution of the prior error and the state transition matrix. The state transition matrix is derived from the kinematics of the ground IMU. The kinematic derivation formula of the ground IMU includes a position update formula, a velocity update formula, and a rotation matrix update formula.

[0042] Preferably, when obtaining data based on extreme weather, take the reciprocal of the negative indicators in the sensor database to normalize the data in the sensor database, and adjust the signal strength and original observation values of the L1 or L2 frequency band output by satellite navigation GNSS in real time to construct an evaluation matrix. The index functions of the evaluation matrix include GNSS signal-to-noise ratio, visual feature matching rate, and ground IMU confidence;

[0043] Calculate the index weights of the evaluation matrix based on the RSR method, obtain the optimal correction factor through step S21, perform weighted addition of the optimal correction factor and the index weights to obtain the fusion weights of each sensor, use the fusion weights as the extreme weather weight combination of the dynamic time-delay observation correction model, iteratively train the dynamic time-delay observation correction model based on the real-time updated sensor database to obtain the optimal extreme weather weight combination, analyze the historical data output by the ground IMU using the hierarchical LSTM model to identify abnormal data, use the abnormal data as drift mode data, input the drift mode data into the dynamic time-delay observation correction model, use the dynamic equation as the discrete loss function for real-time data correction, and achieve error compensation through the Kalman filter.

[0044] An AI-based precise geographic information navigation system includes a sensor module, a model establishment module, and an extreme optimization module:

[0045] The sensor module includes obtaining data through sensors and establishing a sensor database;

[0046] The model establishment module includes, based on the sensor database, performing branch processing on the sliding-selected time feature set and space feature set based on the hierarchical LSTM structure, constructing a dynamic time-delay observation correction model, integrating spatio-temporal features into the dynamic time-delay observation correction model and embedding physical constraints;

[0047] The extreme optimization module includes maintaining the positioning accuracy through a fusion algorithm for the dynamic time-delay observation correction model and optimizing the navigation positioning in extreme weather.

[0048] The beneficial effects of the present invention: Adopt four-level redundant tight coupling of satellite navigation GNSS, ground IMU, vision, and ground sensors, dynamically correct Saastamoinen through the hierarchical LSTM structure to adapt to meteorological mutations, and through the sliding window branch feature quantity, embed physical constraints to construct a dynamic time-delay observation correction model, include more extreme weather samples in the training data, improve the generalization ability of the dynamic time-delay observation correction model, and at the same time can predict wet delay and other factors affecting positioning, such as ionospheric delay, thereby overall improving the navigation accuracy. Use LSTM to correct the residuals of the traditional model instead of completely replacing it, combining the advantages of both. Integrate the dynamic time-delay observation correction model with satellite navigation GNSS to solve the long-term drift problem of the ground IMU in extreme weather, and deeply integrate with physical models and tight-coupling navigation algorithms to achieve sub-meter-level real-time positioning accuracy in extreme weather, realizing deep coordination of multi-source data, dynamic environment adaptive modeling, and optimization of computing resources, providing a systematic solution path for high-precision positioning in extreme weather. Description of the Drawings

[0049] Figure 1Schematic diagram of the basic process of a method for precise navigation of geographic information based on AI provided by an embodiment of the present invention. Detailed implementation manners

[0050] To make the above objects, features, and advantages of the present invention more obvious and understandable, the following will describe the detailed implementation manners of the present invention in conjunction with the drawings of the specification. Obviously, the described embodiments are part of the embodiments of the present invention, rather than all embodiments.

[0051] Refer to Figure 1 , which is an embodiment of the present invention, and provides a method for precise navigation of geographic information based on AI, including the following steps:

[0052] Step S1: Obtain data through sensors and establish a sensor database;

[0053] Step S2: Based on the sensor database, perform branch processing on the time feature set and space feature set selected by sliding based on the hierarchical LSTM structure, construct a dynamic time-delay observation correction model, fuse the spatio-temporal features into the dynamic time-delay observation correction model and embed physical constraints;

[0054] Step S3: Maintain the positioning accuracy through a fusion algorithm for the dynamic time-delay observation correction model and optimize the navigation positioning in extreme weather.

[0055] The sensors include satellite navigation GNSS, fiber Bragg grating (FBG) strain sensors, ground IMU, and vision systems. A low-cost dual-antenna satellite navigation GNSS receiver is used to obtain the signal strength and raw observations of the L1 or L2 frequency band in real time. An FBG strain sensor network is deployed to monitor the micro-deformation of the ground surface. The continuous operation data of the ground IMU is collected. The continuous operation data includes the data of the accelerometer, gyroscope, and temperature sensor, including the drift feature data under normal conditions and extreme weather. Extreme weather includes heavy rain and high temperature. The integrated ground IMU is used as an inertial measurement unit to track the dynamic attitude of the target object, and a binocular vision system is used to obtain high-resolution terrain matching data;

[0056] Maintaining the time synchronization and spatial alignment of the data obtained through sensors includes:

[0057] Through a unified clock system and data synchronization mechanism, the data obtained from each slave sensor is accurately aligned in time to avoid data errors caused by the acquisition time difference of different sensors. The network time protocol (NTP) is used to synchronize the clocks of sensors that do not have the ability to directly receive satellite navigation GNSS signals. Through network connection, the sensors can regularly obtain the current accurate time from the central time server to ensure that the clocks of all sensors are always synchronized. Give all sensors a unified trigger signal to unify the data acquisition moments of all sensors to the same time point;

[0058] Precisely add timestamps to each sensor during data acquisition. The logic for adding timestamps includes: giving each acquired data record a timestamp, where the timestamp is from the same clock source, and recording the spatio-temporal reference, accuracy metrics, and real-time working status of the sensor corresponding to each data. The spatio-temporal reference includes the reference coordinate system and the timestamp, the accuracy metrics include the positioning accuracy and the sensor sensitivity, and the real-time working status includes the configuration information, operating status, health status, and built-in performance metrics of the sensor;

[0059] Through the spatial registration algorithm between sensors, spatially align the positioning data of satellite navigation GNSS, visual data, ground IMU data, and FBG sensor data. The spatial alignment includes making the reference coordinates of the positioning data of satellite navigation GNSS, visual data, ground IMU data, and FBG sensor data fall on the same reference coordinate system;

[0060] Use Kalman filtering to optimize and fuse the time-synchronized and spatially aligned data, and integrate the optimized and fused data set into the sensor database to improve the accuracy and reliability of the data. It can utilize the data of multiple sensors to complement each other in the presence of noise or uncertainty, thereby obtaining more accurate real-time data.

[0061] Step S2 includes the following sub-steps:

[0062] Step S21: Using the TOPSIS multi-criteria decision-making algorithm, select the initial correction factors from the sensor database based on data timeliness, spatial resolution, and reliability. Use a sliding window to screen the initial correction factors to obtain the optimal correction factors, and divide the initial correction factors into a time feature set and a spatial feature set;

[0063] Step S22: Based on the sensor database, use the Saastamoinen model to calculate the prior dry delay and wet delay;

[0064] Step S23: Use the ID convolutional layer to extract the local meteorological gradient and construct a dynamic time-delay observation correction model.

[0065] By deeply integrating LSTM with physical models and tightly coupled navigation algorithms, sub-meter-level real-time positioning accuracy can be achieved in extreme weather, which is 3 - 5 times higher than traditional methods, providing reliable positioning guarantees for scenarios such as unmanned driving and disaster emergency.

[0066] Step S21 includes the following sub-steps:

[0067] The time feature set and the spatial feature set are used to capture the laws of data variation over time;

[0068] Step S211: Select the initial correction factor from the sensor database using the TOPSIS multi-criteria decision-making algorithm;

[0069] Step S212: Calculate the time delay and update time interval of the initial correction factor, and measure the standard deviation of the measurement data update time interval, which is used as the evaluation index of timeliness;

[0070] Step S213: Extract the time feature set and spatial feature set according to the sliding window, and construct a decision matrix including:

[0071] Obtain the time feature set based on a 5-minute interval and an overall 1-hour sliding window, and obtain the spatial feature set based on a 10-minute interval and an overall 1-hour sliding window. The performance of each correction factor on each feature will be used as the rows of the decision matrix, and the features will be used as the columns. The features are random eigenvalue in the time feature set and spatial feature set. The time feature set includes data timeliness, and the spatial feature set includes spatial resolution and reliability;

[0072] The decision matrix is shown in the following table:

[0073] Correction factor Timeliness Spatial resolution Reliability Correction factor 1 0.9 0.85 0.8 Correction factor 2 0.8 0.9 0.9 Correction factor 3 0.85 0.8 0.85 ... ... ... ...

[0074] Step S214: Normalize the decision matrix, and obtain the positive ideal solution and negative ideal solution respectively according to the normalized decision matrix. The positive ideal solution is the maximum value of each column, and the negative ideal solution is the minimum value of each column;

[0075] Step S215: Calculate the Euclidean distances between each correction factor and the positive ideal solution and the negative ideal solution respectively, calculate the relative closeness of each correction factor to the positive ideal solution according to the Euclidean distance, and select the correction factor with the largest relative closeness as the optimal correction factor. Discard the numerical nodes corresponding to the features of the non-optimal correction factor from the time feature set and spatial feature set to obtain the final time feature set and spatial feature set.

[0076] Step S23 includes the following sub-steps:

[0077] Step S231: Extract the local meteorological gradient using the ID convolutional layer. The local meteorological gradient includes the temperature spatial change rate, and branch-process the time feature set and spatial feature set based on the hierarchical LSTM structure;

[0078] Step S232: Construct a dynamic time-delay observation correction model, and perform adversarial training on the dynamic time-delay observation correction model using gradient penalty to improve the generalization ability of the model when data is missing.

[0079] Step S231 includes:

[0080] Process the time feature set and the spatial feature set through the first layer of the hierarchical LSTM structure for high-frequency interval data to capture turbulent fluctuations. Input the captured turbulent fluctuations into the second layer of the hierarchical LSTM structure for medium-frequency aggregation processing to extract the eigenvalues of weather system evolution. Input the eigenvalues of weather system evolution into the third layer of the hierarchical LSTM structure for low-frequency learning to learn seasonal pattern features with a daily unit cycle component.

[0081] Step S232 includes:

[0082] Take the wet delay output by Saastamoinen as a prior value, adaptively learn the residual term of the wet delay using a neural network, use the atmospheric thermodynamic equation constraint as a physical loss function, and add MC-Dropout variational inference to the output layer to predict the mean and standard deviation of the wet delay;

[0083] Embed the wet delay predicted by LSTM into the GNSS pseudorange equation to obtain a tightly coupled observation equation, and its mathematical expression is:

[0084]

[0085] Where P is the GNSS pseudorange coordinate, ρ is the received signal distance, c is the speed of light, δt is the delay correction error between the sensor and the original coordinate point of the current coordinate system, and the delay correction error between the sensor and the original coordinate point of the current coordinate system is the error value between the initial value and the predicted value of the coordinate parameter calculated by the least squares method iteration, T dry is the dry delay, is the wet delay predicted by LSTM, and ε is the noise;

[0086] Set the weight of the dynamic time delay observation correction model to be calculated by a dynamic uncertainty formula, and its mathematical expression is:

[0087]

[0088] Where γ is the weight, is the standard deviation of the wet delay, σ GNSS is the observation weight of the satellite navigation GNSS, and the observation weight of the satellite navigation GNSS is manually set.

[0089] According to step S21, set a 1-hour sliding window with a 10-minute interval for the sliding window, and extract the initial correction factor from the continuous operation data and dynamic attitude obtained through the ground IMU from the sensor database to obtain the temperature drift amount correlation feature quantity and the angular velocity integral error accumulation rate feature quantity;

[0090] Add the Euler dynamics equation as a regular term to the loss function of the dynamic time delay observation correction model to ensure that the predicted value conforms to physical laws;

[0091] The difference between the predicted values of the output temperature drift amount correlation feature quantity and the angular velocity integral error accumulation rate feature quantity and the temperature drift amount correlation feature quantity and the angular velocity integral error accumulation rate feature quantity calculated per actual unit time is used as the prior error to establish a ground IMU error state vector. The ground IMU error state vector is obtained through the convolution of the prior error and the state transition matrix, and the state transition matrix is derived from the kinematics of the ground IMU. The kinematic derivation formula of the ground IMU includes a position update formula, a velocity update formula, and a rotation matrix update formula.

[0092] When obtaining data based on extreme weather, take the reciprocal of the negative indicators in the sensor database to normalize the data in the sensor database, and adjust the signal intensity and original observation values of the L1 or L2 frequency bands output by the satellite navigation GNSS in real time to avoid the impact of signal error accumulation on positioning accuracy, and construct an evaluation matrix. The index functions of the evaluation matrix include GNSS signal-to-noise ratio, visual feature matching rate, and ground IMU confidence;

[0093] The GNSS signal-to-noise ratio, visual feature matching rate, and ground IMU confidence are calculated from the data in the sensor database through the formulas of existing technologies.

[0094] Calculate the index weights of the evaluation matrix based on the RSR method, obtain the optimal correction factor through step S21, perform weighted addition of the optimal correction factor and the index weights to obtain the fusion weights of each sensor, use the fusion weights as the extreme weather weight combination of the dynamic time delay observation correction model, and iteratively train the dynamic time delay observation correction model based on the real-time updated sensor database to obtain the optimal extreme weather weight combination. Use the hierarchical LSTM model to analyze the historical data output by the ground IMU, identify abnormal data, use the abnormal data as drift mode data, input the drift mode data into the state time delay observation correction model, use the dynamic equation as the discrete loss function to perform real-time data correction, and achieve error compensation through the Kalman filter.

[0095] An AI-based precise geographic information navigation system includes a sensor module, a model establishment module, and an extreme optimization module:

[0096] The sensor module includes obtaining data through sensors and establishing a sensor database;

[0097] The model establishment module includes, based on the sensor database, performing branch processing on the sliding-selected time feature set and space feature set based on the hierarchical LSTM structure, constructing a dynamic time delay observation correction model, and integrating the spatio-temporal features into the dynamic time delay observation correction model and embedding physical constraints;

[0098] The extreme optimization module includes maintaining positioning accuracy through a fusion algorithm for the dynamic time delay observation correction model and optimizing extreme weather navigation positioning.

[0099] The present invention adopts a four-level redundant tight coupling of satellite navigation GNSS, ground IMU, vision and ground sensors, dynamically corrects Saastamoinen through a hierarchical LSTM structure to adapt to meteorological mutations, and constructs a dynamic time-delay observation correction model by embedding physical constraints through sliding window branch feature quantities. By including more samples of extreme weather in the training data, the generalization ability of the dynamic time-delay observation correction model is improved. At the same time, it can predict wet delay and other factors affecting positioning, such as ionospheric delay, thereby improving the navigation accuracy as a whole. The LSTM is used to correct the residuals of the traditional model instead of completely replacing it, combining the advantages of both. By fusing the dynamic time-delay observation correction model with satellite navigation GNSS, the problem of long-term drift of the ground IMU under extreme weather is solved, and it is deeply integrated with the physical model and the tight coupling navigation algorithm to achieve sub-meter real-time positioning accuracy under extreme weather, realizing deep cooperation of multi-source data, dynamic environment adaptive modeling and computing resource optimization, and providing a systematic solution path for high-precision positioning under extreme weather.

[0100] Those skilled in the art should understand that the embodiments of the present invention may be provided as a method, a system or a computer program product. Therefore, the present invention may take the form of a complete hardware embodiment, a complete software embodiment or an embodiment combining software and hardware aspects. Moreover, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media containing computer-usable program code therein. Among them, the storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (Static Random Access Memory, abbreviated as SRAM), electrically erasable programmable read-only memory (Electrically Erasable Programmable Read-Only Memory, abbreviated as EEPROM), erasable programmable read-only memory (Erasable Programmable Read Only Memory, abbreviated as EPROM), programmable read-only memory (Programmable Red-Only Memory, abbreviated as PROM), read-only memory (Read-Only Memory, abbreviated as ROM), magnetic memory, flash memory, magnetic disk or optical disc. These computer program instructions can also be stored in a computer-readable memory capable of guiding a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer-readable memory produce a manufactured article including an instruction device, and the instruction device implements the functions specified in one process Figure 1 a process or multiple processes and / or blocks Figure 1 the functions specified in a block or multiple blocks.

[0101] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit them. Although the present invention has been described in detail with reference to the preferred embodiments, those of ordinary skill in the art should understand that the technical solutions of the present invention can be modified or equivalently replaced without departing from the spirit and scope of the technical solutions of the present invention, and all of them should be covered by the scope of the claims of the present invention.

Claims

1. A geographic information precise navigation method based on AI, characterized in that: The following steps are involved: Step S1: Acquire data through sensors and establish a sensor database; Step S2: Based on the sensor database, the time feature set and the space feature set of the sliding selection are branched based on the hierarchical LSTM structure, a dynamic time delay observation correction model is constructed, the time and space features are integrated into the dynamic time delay observation correction model and physical constraints are embedded; Step S3: Maintaining positioning accuracy by fusing the dynamic time delay observation correction model and optimizing extreme weather navigation positioning.

2. The AI-based geographic information precise navigation method according to claim 1, characterized in that: The sensors include satellite navigation GNSS, fiber Bragg grating (FBG) strain sensors, ground IMU and vision systems. A low-cost dual-antenna satellite navigation GNSS receiver is used to obtain L1 or L2 frequency band signal strength and original observation values ​​in real time. A fiber Bragg grating (FBG) strain sensor network is deployed to monitor surface micro-deformation. The ground IMU continuous operation data is collected. The continuous operation data includes data from accelerometers, gyroscopes and temperature sensors, including drift characteristic data under normal conditions and extreme weather conditions. The extreme weather conditions include heavy rain and high temperature. The ground IMU is integrated as an inertial measurement unit to track the dynamic posture of the target object. A binocular vision system is used to obtain high-resolution terrain matching data. Maintaining temporal synchronization and spatial alignment of data acquired by sensors involves: Through a unified clock system and data synchronization mechanism, the data obtained from each sensor is accurately aligned in time. The network time protocol NTP is used to synchronize the clocks of sensors that do not directly receive satellite navigation GNSS signals. A unified trigger signal is given to all sensors, unifying the data collection time of all sensors to the same time point. Adding a timestamp to each sensor during data collection, the logic of adding the timestamp includes: recording a timestamp for each acquired data, the timestamp is derived from the same clock source, and recording the spatiotemporal reference, accuracy index and real-time working status of the sensor corresponding to each data, the spatiotemporal reference includes a reference coordinate system and a timestamp, the accuracy index includes positioning accuracy and sensor sensitivity, and the real-time working status includes configuration information, operating status, health status and performance index of the sensor; The positioning data, visual data, ground IMU data and FBG sensor data of the satellite navigation GNSS are spatially aligned by a spatial registration algorithm between sensors, wherein the spatial alignment includes placing the reference coordinates of the positioning data, visual data, ground IMU data and FBG sensor data of the satellite navigation GNSS on the same reference coordinate system; Kalman filtering is used to optimize the fusion of time-synchronized and space-aligned data, and the optimized fused data are integrated into a sensor database.

3. The AI-based geographic information precise navigation method according to claim 2, characterized in that: The step S2 comprises the following sub-steps: Step S21: using the TOPSIS multi-criteria decision algorithm, selecting an initial correction factor from the sensor database based on data timeliness, spatial resolution and reliability, using a sliding window to screen the initial correction factor to obtain an optimal correction factor, and dividing the initial correction factor into a time feature set and a spatial feature set; Step S22: Calculate a priori dry delay and wet delay using the Saastamoinen model based on the sensor database; Step S23: Use the ID convolution layer to extract the local meteorological gradient and construct a dynamic time-delay observation correction model.

4. The AI-based geographic information precise navigation method according to claim 3, characterized in that: The step S21 includes the following sub-steps: Step S211: Selecting an initial correction factor from the sensor database using the TOPSIS multi-criteria decision algorithm; Step S212: Calculate the time delay and update time interval of the initial correction factor, measure the standard deviation of the data update time interval, and use the standard deviation as an evaluation index of timeliness; Step S213: extracting the temporal feature set and the spatial feature set according to the sliding window, and constructing a decision matrix includes: The time feature set is obtained based on a sliding window with a 5-minute interval and a 1-hour overall sliding window, and the spatial feature set is obtained based on a sliding window with a 10-minute interval and a 1-hour overall sliding window. The performance of each correction factor on each feature will be used as the row of the decision matrix, and the feature will be used as the column. The feature is the random feature value in the time feature set and the spatial feature set. The time feature set includes data timeliness, and the spatial feature set includes spatial resolution and reliability; Step S214: normalizing the decision matrix, and obtaining a positive ideal solution and a negative ideal solution according to the normalized decision matrix, wherein the positive ideal solution is the maximum value of each column, and the negative ideal solution is the minimum value of each column; Step S215: Calculate the Euclidean distance between each correction factor and the positive ideal solution and the negative ideal solution respectively, calculate the relative proximity between each correction factor and the positive ideal solution based on the Euclidean distance, select the correction factor with the largest relative proximity as the optimal correction factor, discard the numerical nodes corresponding to the features corresponding to the non-optimal correction factors from the time feature set and the space feature set, and obtain the final time feature set and the space feature set.

5. The AI-based geographic information precise navigation method according to claim 4, characterized in that: The step S23 includes the following sub-steps: Step S231: extracting a local meteorological gradient using an ID convolution layer, wherein the local meteorological gradient includes a temperature spatial change rate, and performing branching processing on the temporal feature set and the spatial feature set based on a hierarchical LSTM structure; Step S232: construct a dynamic time delay observation correction model, and use gradient penalty to perform adversarial training on the dynamic time delay observation correction model.

6. The AI-based geographic information precise navigation method according to claim 5, characterized in that: The step S231 includes: The time feature set and the spatial feature set are processed with high-frequency interval data through the first layer of the hierarchical LSTM structure to capture turbulent fluctuations. The captured turbulent fluctuations are input into the second layer of the hierarchical LSTM structure for medium-frequency aggregation processing to extract the weather system evolution characteristic values. The weather system evolution characteristic values ​​are input into the third layer of the hierarchical LSTM structure for low-frequency learning, and the seasonal pattern characteristics are learned with daily periodic components.

7. The AI-based geographic information precise navigation method according to claim 6, characterized in that: The step S232 includes: The wet delay output by Saastamoinen is used as a priori value, and the residual term of wet delay is adaptively learned using a neural network. The atmospheric thermodynamic equation constraint is used as the physical loss function, and MC-Dropout variational inference is added to the output layer to predict the mean and standard deviation of the wet delay. The wet delay predicted by LSTM is embedded in the GNSS pseudorange equation to obtain the tightly coupled observation equation, whose mathematical expression is: Where P is the GNSS pseudo-range coordinate, ρ is the received signal distance, c is the speed of light, δt is the delay correction error between the sensor and the original coordinate point of the current coordinate system, and the delay correction error between the sensor and the original coordinate point of the current coordinate system is the error between the initial value and the predicted value of the coordinate parameter iteratively calculated by the least squares method, T dry For dry delay, is the wet delay predicted by LSTM, ε is the noise; The weight of the dynamic time delay observation correction model is set to be calculated by a dynamic uncertainty formula, and its mathematical expression is: Among them, γ is the weight, is the standard deviation of wet delay, σ GNSS is the observation weight of satellite navigation GNSS.

8. The AI-based geographic information precise navigation method according to claim 7, characterized in that: According to step S21, the sliding window is set to an overall 1-hour sliding window with a 10-minute interval, and an initial correction factor is extracted from the continuous operation data and dynamic posture obtained by the ground IMU in the sensor database to obtain a temperature drift correlation feature value and an angular velocity integral error accumulation rate feature value; Adding the Euler dynamics equation as a regular term into the loss function of the dynamic time-delay observation correction model; The difference between the output temperature drift correlation characteristic quantity and the angular velocity integral error accumulation rate characteristic quantity predicted value and the temperature drift correlation characteristic quantity and the angular velocity integral error accumulation rate characteristic quantity calculated per actual unit time is taken as a priori error, and a ground IMU error state vector is established. The ground IMU error state vector is obtained by convolution of the priori error and a state transfer matrix. The state transfer matrix is ​​derived by ground IMU kinematics. The ground IMU kinematics derivation formula includes a position update formula, a velocity update formula and a rotation matrix update formula.

9. The AI-based geographic information precise navigation method according to claim 8, characterized in that: When acquiring data based on extreme weather, the negative index in the sensor database is taken inversely to normalize the data in the sensor database, and the L1 or L2 frequency band signal strength and the original observation value output by the satellite navigation GNSS are adjusted in real time to construct an evaluation matrix, wherein the index function of the evaluation matrix includes GNSS signal-to-noise ratio, visual feature matching rate and ground IMU confidence; The indicator weights of the evaluation matrix are calculated based on the RSR method, and the optimal correction factor is obtained through step S21. The optimal correction factor and the indicator weight are weightedly added to obtain the fusion weights of each sensor, and the fusion weights are used as the extreme weather weight combination of the dynamic time delay observation correction model. The dynamic time delay observation correction model is iteratively trained based on a real-time updated sensor database to obtain the optimal extreme weather weight combination. The hierarchical LSTM model is used to analyze the historical data output by the ground IMU, and abnormal data is identified. The abnormal data is used as drift mode data, and the drift mode data is input into the state delay observation correction model. The dynamic equation is used as a discrete loss function to perform real-time data correction, and error compensation is achieved through a Kalman filter.

10. An AI-based geographic information precision navigation system, characterized in that: Includes sensor module, model building module and extreme optimization module: The sensor module includes acquiring data through sensors and establishing a sensor database; The model building module includes branching the time feature set and the space feature set of the sliding selection based on the sensor database and the hierarchical LSTM structure, building a dynamic time delay observation correction model, integrating the time and space features into the dynamic time delay observation correction model and embedding physical constraints; The extreme optimization module includes maintaining positioning accuracy by fusing the dynamic time delay observation correction model with an algorithm to optimize extreme weather navigation positioning.

Images