Satellite navigation positioning method and device, and storage medium

Abstract

The application discloses a positioning method, device and storage medium of satellite navigation. The positioning method comprises the following steps: acquiring observation parameters collected by a mobile terminal at a preset epoch, wherein the observation parameters comprise GNSS parameters and inertial navigation related parameters, the GNSS parameters comprise GNSS prior parameters and GNSS posterior parameters; inputting the observation parameters into a preset positioning accuracy evaluation model to obtain an accuracy evaluation value, wherein the positioning accuracy evaluation model is obtained based on historical observation parameters collected at multiple epochs; determining whether the positioning information represented by the current observation parameters is accurate based on the accuracy evaluation value; when it is determined that the positioning information is inaccurate, determining a positioning result based on a float solution of the integer ambiguity; and when it is determined that the positioning information is accurate, determining a positioning result based on a fixed solution of the integer ambiguity. The application has the technical effect of improving the positioning accuracy and reliability of the global satellite navigation system.

Description

Technical Field

[0001] This disclosure relates to the field of satellite navigation and positioning technology, and in particular to a satellite navigation positioning method, apparatus, and storage medium. Background Technology

[0002] Global Navigation Satellite Systems (GNSS) are widely used in various fields such as vehicle navigation, autonomous driving, geographic information mapping, and mobile communications due to their ability to provide all-weather, global positioning and navigation services. With the development of artificial intelligence and machine learning technologies, utilizing big data, fusing multi-dimensional features, and learning patterns from historical data through models to automatically determine the state of complex systems has become an important direction for improving the reliability of high-precision positioning in GNSS. Therefore, the issue of positioning accuracy detection deserves attention. Summary of the Invention

[0003] In view of this, embodiments of this application provide a satellite navigation positioning method, apparatus, and storage medium to improve the positioning accuracy of a global satellite navigation system. In a first aspect, a satellite navigation positioning method is provided, comprising: acquiring observation parameters collected by a mobile terminal at a preset epoch, the observation parameters including GNSS parameters and inertial navigation associated parameters, the GNSS parameters including GNSS prior parameters and GNSS posterior parameters; inputting the observation parameters into a preset positioning accuracy evaluation model to obtain an accuracy evaluation value, wherein the positioning accuracy evaluation model is trained based on historical observation parameters collected at multiple epochs; determining whether the positioning information represented by the current observation parameters is accurate based on the accuracy evaluation value; when the positioning information is determined to be inaccurate, determining the positioning result based on a floating-point solution of integer ambiguity; when the positioning information is determined to be accurate, determining the positioning result based on a fixed solution of integer ambiguity.

[0004] The above satellite navigation positioning method can use observation parameters to determine the accuracy evaluation value through a preset positioning accuracy evaluation model, and can determine the accuracy of the positioning information represented by the current observation parameters according to the preset accuracy threshold. Compared with the traditional method that relies on a single empirical threshold, it significantly improves the ability to identify "false fixed solutions" in complex environments, effectively reduces the risk of positioning misjudgment or omission, and improves the reliability and security of the satellite navigation positioning system.

[0005] Optionally, the GNSS posterior parameters include at least one of the following: fixed-solution floating-point solution difference information, ambiguity floating-point solution and fixed-solution difference information, and integer ambiguity variation information.

[0006] Optionally, the inertial navigation correlation parameters include at least one of the following: inertial navigation position and fixed solution difference information, inertial navigation position and floating-point solution difference information, and inertial navigation standard deviation information.

[0007] Optionally, before inputting the observation parameters into the trained positioning accuracy evaluation model, the process may also include: when the inertial navigation correlation parameters meet preset conditions, continuing to input the observation parameters into the trained positioning accuracy evaluation model.

[0008] Optionally, the positioning accuracy evaluation model is trained based on historical observation parameters collected over multiple epochs, including: determining the true value information of each set of historical observation parameters in each epoch based on historical observation parameters and standard observation parameters. The true value information is the difference between the observed navigation coordinates of the mobile terminal and the standard navigation coordinates, and each set of observation parameters corresponds to one true value information. The historical observation parameters and the true value information form the original observation data. The original observation data is input into the initial positioning accuracy evaluation model for training to generate the preset positioning accuracy evaluation model.

[0009] Optionally, the original observation data is input into the initial supervised model for training, including: inputting the original observation data into the initial supervised model based on a preset training ratio, the preset training ratio including a first ratio and a second ratio, the first ratio being greater than or equal to the second ratio; generating a preset positioning accuracy evaluation model, including: training the initial positioning accuracy evaluation model based on historical observation parameters and ground truth information in the original observation data of the first ratio, generating a first positioning accuracy evaluation model; inputting historical observation parameters from the original observation data of the second ratio into the first positioning accuracy evaluation model to determine a first result value; and determining the generated preset positioning accuracy evaluation model based on the first result value and the ground truth information in the original observation data of the second ratio.

[0010] Optionally, it also includes: during the training of the preset positioning accuracy evaluation model, calling the tree model interpreter to determine the parameter contribution values ​​of multiple parameters in the observation parameters; and adjusting the type of historical observation parameters in the training positioning accuracy evaluation model based on the parameter contribution values.

[0011] Optionally, the preset positioning accuracy evaluation model is the extreme gradient boosting model.

[0012] In a second aspect, a satellite navigation positioning device is provided, comprising: a processor for coupling a memory, the memory including instructions, which, when invoked by the processor, cause the processor to execute the positioning method provided in the first aspect.

[0013] Thirdly, a computer-readable storage medium is provided, including instructions stored thereon, wherein when executed by a processor, the instructions are executed as described in the positioning method of the first aspect.

[0014] Fourthly, a satellite navigation positioning device is provided, comprising: an acquisition unit for acquiring observation parameters collected by a mobile terminal at a preset epoch, the observation parameters including GNSS parameters and inertial navigation associated parameters, the GNSS parameters including GNSS prior parameters and GNSS posterior parameters; a determination unit for inputting the observation parameters into a preset positioning accuracy evaluation model to obtain an accuracy evaluation value, wherein the positioning accuracy evaluation model is trained based on historical observation parameters collected at multiple epochs; and a judgment unit for determining the positioning result based on a floating-point solution of integer ambiguity when the positioning information is determined to be inaccurate, and determining the positioning result based on a fixed solution of integer ambiguity when the positioning information is determined to be accurate. Attached Figure Description

[0015] The accompanying drawings used in the description of the embodiments of this disclosure are briefly introduced below:

[0016] Figure 1 A schematic flowchart of a satellite navigation positioning method provided in some embodiments of this application is shown;

[0017] Figure 2 A schematic diagram of the structure of a satellite navigation positioning device provided in some embodiments of this application is shown;

[0018] Figure 3 A schematic diagram of the structure of a satellite navigation positioning device provided in some embodiments of this application is shown. Detailed Implementation

[0019] To more clearly illustrate the technical solutions in the embodiments of this disclosure, examples of implementation methods of this disclosure will be described below with reference to the accompanying drawings. The accompanying drawings described below are merely some embodiments of this disclosure. For those skilled in the art, other drawings and other implementation methods can be obtained based on these drawings without creative effort. Adjustments and improvements made without departing from the concept of this disclosure are all within the protection scope of this disclosure.

[0020] To keep the drawings simple, each figure only schematically shows the parts relevant to the embodiment, and they do not represent the actual structure of the product. In addition, for the sake of clarity and ease of understanding, some figures only schematically show parts of components with the same structure or function, and there may actually be more or fewer components with the same structure or function.

[0021] In this disclosure, unless otherwise expressly specified and limited, ordinal numbers, such as “first”, “second”, etc., are used only to distinguish and describe related objects, and should not be construed as indicating or implying the relative importance or order between related objects; furthermore, they do not represent the quantity of related objects. “Multiple” includes two or more, and other quantifiers are similar. “ / ” is used to describe the relationship between related objects, indicating an “or” relationship between them. “And / or” is used to describe the relationship between related objects, including any combination relationship between them, such as “a and / or b” including: “a alone”, “b alone”, or “a and b”. “One or more” or “at least one” of multiple objects refers to any object or any combination of multiple objects, such as “one or more of a1, a2, a3” or “at least one of a1, a2, a3” including: “a1 alone”, “a2 alone”, “a3 alone”, “a1 and a2”, “a1 and a3”, “a2 and a3”, or “a1, a2 and a3”.

[0022] Global Navigation Satellite System (GNSS) is widely used in vehicle navigation, UAV control, geographic information mapping, mobile communications, and many other fields due to its ability to provide all-weather, global positioning and navigation services. With the increasing demand for high-precision positioning, high-precision solution techniques such as carrier phase observation and integer ambiguity fixing are being widely adopted. Among these, the GNSS "fixed solution" is considered to have centimeter-level positioning accuracy, theoretically capable of meeting the requirements of applications with extremely strict positioning error control. However, in complex environments such as urban canyons, high-rise buildings, and dense vegetation, GNSS signals are easily affected by multipath effects, signal blockage, ionospheric delay, and other factors, leading to a decline in the quality of observation data. As a result, even if the solution result is marked as a "fixed solution," the actual positioning accuracy may be significantly reduced, resulting in "fixed solution misjudgment" or "false fixed solution." In such cases, traditional GNSS positioning systems primarily rely on empirical thresholds or single residuals and confidence levels to determine the reliability of solutions. This often fails to adequately reflect the comprehensive quality of multi-source observation data in complex environments, leading to a high probability of misjudgments and omissions, thus impacting system security and user experience. The satellite navigation positioning method, apparatus, and storage medium provided in this application can improve the accuracy of satellite navigation positioning in complex environments, ensuring the safe and stable operation of the satellite navigation positioning system and its carrier.

[0023] The following description is in conjunction with the accompanying drawings:

[0024] Figure 1 A schematic flowchart of a satellite navigation positioning method provided in some embodiments of this application is shown. This satellite navigation positioning method includes at least the following steps:

[0025] S110: Obtain the observation parameters collected by the mobile terminal at a preset epoch. The observation parameters include GNSS parameters and inertial navigation associated parameters. The GNSS parameters include GNSS prior parameters and GNSS posterior parameters.

[0026] S120: Input the observation parameters into the preset positioning accuracy evaluation model to obtain the accuracy evaluation value. The positioning accuracy evaluation model is trained based on historical observation parameters collected at multiple epochs.

[0027] S130: Determine whether the positioning information represented by the current observation parameters is accurate based on the accuracy evaluation value;

[0028] S140: When the positioning information is determined to be inaccurate, the positioning result is determined based on the floating-point solution of integer ambiguity;

[0029] S150: When the positioning information is determined to be accurate, the positioning result is determined based on the fixed solution of integer ambiguity.

[0030] In the above embodiments of the satellite navigation positioning method, the mobile terminal can be a vehicle-mounted navigation device, a drone, a measuring instrument, or other carrier or device. During its continuous operation, it collects observation parameters at preset time intervals (i.e., preset epochs). The preset epoch can be determined based on the mobile terminal's frequency; for example, at a working frequency of 10Hz, it collects parameters every 0.1 seconds. The observation parameters can include GNSS parameters and inertial navigation associated parameters, and can be collected synchronously in real time by the mobile terminal's GNSS receiver and inertial navigation module. GNSS prior parameters are input information that can be directly obtained from observation geometry, signal quality, ephemeris, etc., before ambiguity fixing or position resolution. GNSS posterior parameters are derived indices obtained after floating-point or fixed-point solutions by calculating residuals, covariance matrices, statistical tests, etc. When determining the accuracy evaluation value using the preset positioning accuracy evaluation model and observation parameters, a correspondence between observation parameters and true values ​​(such as the true distance between the terminal positioning result and the reference point) can be established by collecting historical observation parameters, forming a training sample set. Based on the aforementioned training sample set, a supervised learning model, such as an XGBoost regression model or a neural network model, is used for training and validation to obtain model parameters that can comprehensively judge positioning accuracy based on observation parameters. This positioning accuracy evaluation model can utilize a large amount of historical observation data to learn the complex nonlinear relationship between observation parameters and actual positioning accuracy. Compared with traditional static empirical thresholds or simple criteria, it is more adaptable to the complex environmental changes during GNSS positioning and continuously learns and adaptively adjusts based on actual feedback, constantly improving its discrimination performance and achieving long-term evolution. In the actual detection phase, the observation parameters collected at the current epoch are used as input and passed to the pre-trained positioning accuracy evaluation model, which can output an accuracy evaluation value reflecting the expected error or reliability of the current positioning solution. Furthermore, the accuracy evaluation value is compared with the system-set accuracy threshold (e.g., 0.2 meters) to determine the accuracy of the positioning information. If the accuracy evaluation value is less than or equal to the accuracy threshold, the positioning information represented by the current observation parameters is considered accurate, and the positioning result is determined based on the fixed solution of integer ambiguity. If the accuracy evaluation value is greater than the accuracy threshold, the positioning information is considered inaccurate, and the positioning result is determined based on the floating-point solution of integer ambiguity. The accuracy threshold can be determined based on the maximum tolerance for positioning errors in the application scenario of the positioning accuracy detection method. For example, lane-level navigation may require an accuracy threshold less than or equal to 0.5 meters, while high-precision mapping requires less than or equal to 0.1 meters, thus allowing the end user to set it or according to industry standards. Alternatively, the accuracy threshold can be selected based on historically collected positioning error distribution statistics, choosing a typical value that satisfies a certain confidence interval or cumulative error distribution.For example, by analyzing the actual error data of all previous fixed solution states, the maximum error value that can cover the vast majority of true fixed solutions is selected as the accuracy threshold, without specific limitations. When the positioning information is determined to be inaccurate, the system automatically switches to output the floating-point solution result of that epoch, avoiding mistrusting unreliable fixed solutions and improving overall positioning security. When the positioning information is determined to be accurate, the system directly outputs the fixed solution as the positioning result for applications such as vehicle navigation and surveying. This application can use observation parameters to determine the accuracy evaluation value through a preset positioning accuracy evaluation model, and can determine the accuracy of the positioning information represented by the current observation parameters according to the preset accuracy threshold. Compared with the traditional method that relies on a single empirical threshold, it significantly improves the ability to identify "false fixed solutions" in complex environments, effectively reduces the risk of positioning misjudgment or omission, and improves the reliability and security of the satellite navigation and positioning system. In addition, the accuracy evaluation value can provide a weight reference for the fusion calculation of inertial navigation and GNSS, realize multi-sensor adaptive fusion, and provide reference in various application scenarios such as anomaly warning, map matching degree determination, and trajectory processing optimization, further improving the intelligence and engineering practicality of the overall system. For example, accuracy evaluation values ​​can serve as real-time weights for GNSS observations, allowing inertial navigation systems (INS) to fully trust GNSS solutions when GNSS accuracy is high, and increasing the weight of INS information when accuracy evaluation values ​​increase (i.e., positioning information becomes inaccurate), thereby achieving more stable and adaptive multi-source fusion navigation. As another example, an abnormally large increase in accuracy evaluation values ​​can be considered an "anomaly warning signal" for the positioning system, triggering anomaly detection, applicable to scenarios requiring high-reliability positioning such as autonomous driving and drones. Furthermore, in map matching and location correction, accuracy evaluation values ​​can be combined to give higher confidence to locations with lower accuracy evaluation values ​​and reduce the contribution of locations with higher accuracy evaluation values ​​to trajectory fitting, improving the mobile terminal positioning and map fusion effect.

[0031] In some embodiments of this application, the GNSS posterior parameters include at least one of the following: fixed-solution floating-point solution difference information, ambiguity floating-point solution and fixed-solution difference information, and integer ambiguity change information.

[0032] The fixed-solution vs. floating-point solution difference information describes the coordinate differences between the fixed and floating-point solutions obtained in GNSS positioning. By comparing the three-dimensional position difference between the two, it can reflect whether ambiguity fixing has resulted in a significant offset. The ambiguity floating-point solution vs. fixed-solution difference information measures the difference between the ambiguity estimated in the floating-point solution state and the integer ambiguity solution value in the fixed solution state. When this difference is large, it indicates that the ambiguity fixing result is unstable, thus reducing the reliability of the fixed solution. Integer ambiguity variation information monitors the jumps or fluctuations in integer ambiguity values ​​in consecutive epochs. When frequent changes or discontinuities in ambiguity values ​​are observed, it often indicates the presence of cycle slips, occlusion, or multipath interference, leading to a decrease in the reliability of the fixed solution. Through the above posterior parameters, the consistency between the fixed and floating-point solutions, ambiguity stability, and solution residual level after the solution is completed can be comprehensively reflected, thus providing important input for machine learning models to judge the quality of the fixed solution.

[0033] In some embodiments of this application, the inertial navigation correlation parameters include at least one of the following: inertial navigation position and fixed solution difference information, inertial navigation position and floating-point solution difference information, and inertial navigation standard deviation information. The inertial navigation position and fixed solution difference information is used to represent the spatial difference between the inertial navigation position output and the GNSS fixed solution. If this difference exceeds a reasonable range, it indicates that the fixed solution is abnormal or that the inertial navigation output is drifting. The inertial navigation position and floating-point solution difference information is used to compare the difference between the inertial navigation position and the GNSS floating-point solution. When the fixed solution is unreliable, the difference between the floating-point solution and the inertial navigation position can provide an auxiliary reference to help determine the reliability of the GNSS solution. The inertial navigation standard deviation information is used to describe the magnitude of the uncertainty in the inertial navigation position estimation, such as the standard deviation or covariance matrix of the inertial navigation solution. By combining the standard deviation information, the reliability of the inertial navigation output can be determined, thereby using differentiated weights when comparing GNSS positioning results. By introducing inertial navigation correlation parameters, the short-term accuracy and continuity provided by the inertial navigation system can be utilized to enhance the robustness and integrity of the overall positioning system in cases of insufficient GNSS positioning quality or signal obstruction.

[0034] In some embodiments of this application, before inputting the observation parameters into the preset positioning accuracy evaluation model, the method further includes: when the inertial navigation association parameters meet the preset conditions, continuing to input the observation parameters into the preset positioning accuracy evaluation model.

[0035] In certain positioning scenarios (such as tunnel entrances or areas with severe obstruction), GNSS observation conditions are poor, and the information input to the machine learning model may be incomplete, easily leading to misjudgments. In such cases, if the inertial navigation (INS) discrepancy significantly exceeds a threshold, it's unnecessary to use a preset positioning accuracy evaluation model to determine if inaccurate positioning information exists (i.e., the positioning information at this point is inherently inaccurate), thus avoiding unreliable model outputs. Due to its physical properties, the INS correlation parameter itself can quickly determine whether the current satellite navigation positioning result may be distorted. If the INS can clearly show anomalies, the observation parameters can be removed from the preset positioning accuracy evaluation model for positioning evaluation, thereby saving system power consumption and computational load. However, when the INS correlation parameter meets preset conditions—meaning that the INS correlation parameter alone cannot effectively determine whether distortion exists—then the observation parameters need to continue to be input into the preset positioning accuracy evaluation model for judgment. The preset conditions can be set based on the numerical range or stability of the INS correlation parameter. For example, when the difference between the inertial navigation position and the fixed GNSS solution is less than a preset threshold, the GNSS solution is considered consistent with the INS result, allowing the continued input of observation parameters into the positioning accuracy evaluation model. Alternatively, when the difference between the inertial navigation position and the floating-point solution is within a reasonable range, it indicates that the floating-point solution matches the inertial navigation result, thus meeting the conditions for entering the model judgment. Furthermore, judgment can also be made based on the uncertainty of the inertial navigation itself. For example, when the standard deviation of the inertial navigation is less than a preset upper limit, the inertial navigation result is considered to have sufficient accuracy and reliability, suitable as a reference benchmark for GNSS solution results. In some scenarios, preset conditions can also adopt a combined strategy, such as simultaneously requiring the difference between the inertial navigation and the fixed solution to meet a threshold condition and the standard deviation of the inertial navigation to be within a reasonable range, for comprehensive judgment.

[0036] In some embodiments of this application, the positioning accuracy evaluation model is trained based on historical observation parameters collected over multiple epochs, including: determining the true value information of each group of historical observation parameters in each epoch based on historical observation parameters and standard observation parameters, wherein the true value information is the difference between the observed navigation coordinates of the mobile terminal and the standard navigation coordinates, and each group of observation parameters corresponds to one true value information, and the historical observation parameters and true value information form the original observation data; and inputting the original observation data into the initial supervised model for training to generate the preset positioning accuracy evaluation model.

[0037] For each set of historical observation parameters collected by the mobile terminal across multiple epochs, the corresponding ground truth information is calculated. For example, the difference between the observed navigation coordinates and the standard navigation coordinates can be used to obtain the actual positioning error, which is then used as the ground truth information. In actual calculation, the three-dimensional parameters of the observed and standard navigation coordinates can be converted into one-dimensional parameters for subtraction. The ground truth value can be restricted to less than 1, and greater than 1, and thus set to 1. A ground truth value less than 1 indicates that the vast majority of positioning errors are within 1, and can be retained as labels (e.g., 0.1, 0.3, 0.7, etc.). If the positioning error is particularly large, such as 1.8 meters, 3 meters, or 10 meters, it is uniformly recorded as 1 to avoid these large values ​​increasing the model's loss function and affecting the model's focus. Each set of historical observation parameters corresponds to a ground truth information, and the two together constitute the original observation data sample. Subsequently, all the original observation data samples are input into the initial supervised model, using the historical observation parameters as input and the ground truth information as the output label, for supervised training. Through multiple rounds of iteration and parameter optimization, a preset positioning accuracy evaluation model that can fit the relationship between observed parameters and actual positioning accuracy is finally generated.

[0038] In some embodiments of this application, training an initial positioning accuracy evaluation model by inputting raw observation data includes: inputting raw observation data into the initial positioning accuracy evaluation model based on a preset training ratio, wherein the preset training ratio includes a first ratio and a second ratio, and the first ratio is greater than or equal to the second ratio; generating a preset positioning accuracy evaluation model includes: training the initial positioning accuracy evaluation model based on historical observation parameters and ground truth information in the raw observation data of the first ratio to generate a first positioning accuracy evaluation model; inputting historical observation parameters from the raw observation data of the second ratio into the first positioning accuracy evaluation model to determine a first result value; and determining the generated preset positioning accuracy evaluation model based on the first result value and the ground truth information in the raw observation data of the second ratio.

[0039] In some embodiments of this application, a preset positioning accuracy evaluation model is determined based on the truth information in the first result value and the second proportion of the original observation data. This includes: comparing the truth information in the first result value and the second proportion of the original observation data; under the fixed solution state of the global satellite navigation system, when the absolute value of the difference between the truth information in the first result value and the second proportion of the original observation data is greater than a preset evaluation threshold, continuing to train the first positioning accuracy evaluation model; when the absolute value of the difference between the truth information in the first result value and the second proportion of the original observation data is less than the preset evaluation threshold, using the first positioning accuracy evaluation model as the preset positioning accuracy evaluation model.

[0040] The original observation data is divided to train and validate the initial positioning accuracy evaluation model. Taking an 80% first proportion and a 20% second proportion as an example, 80% of the original observation data is used as the first proportion for training the initial positioning accuracy evaluation model, and 20% is used as the second proportion for validating the model's performance. During the training phase, the observation parameters and their corresponding ground truth information from the first proportion data are input into the initial positioning accuracy evaluation model (e.g., an XGBoost regression model) to obtain a trained first positioning accuracy evaluation model. During the validation phase, the observation parameters from the second proportion data are input into the first positioning accuracy evaluation model to obtain the model's predicted values, which are then compared with the corresponding ground truth information from the second proportion data. If the absolute value of the difference between the predicted value and the ground truth is less than a preset evaluation threshold, the model is considered to have good judgment ability, and the first positioning accuracy evaluation model is adopted as the final preset positioning accuracy evaluation model. If the difference is greater than the evaluation threshold, it indicates that the current model's prediction error is still relatively large, and adjustments can be made to the model parameters, feature terms, or the number of training epochs for further training and optimization. Evaluation thresholds reflect the actual tolerance for positioning errors. For example, in high-precision urban navigation applications, the tolerance for errors is typically less than 0.2 meters, while in general navigation, the allowable deviation may be relaxed to 0.5 to 1 meter. Therefore, the evaluation threshold can be set according to the maximum acceptable error range in a specific use case. Alternatively, the distribution of true errors in a large number of training samples can be statistically analyzed, and a suitable quantile (such as the 95th percentile) can be selected as the threshold, thus balancing the model's recognition ability with the risk of misjudgment. By setting an evaluation threshold to judge the gap between the predicted value and the true accuracy value, the model's acceptable prediction accuracy can be dynamically verified during training, which helps to improve the stability and reliability of the final pre-defined supervised model.

[0041] In some embodiments of this application, GNSS prior parameters include one or more of the following: satellite signal strength information, number of satellites, number of satellite cycle slips, or satellite elevation angle information. GNSS posterior parameters may include: floating-point solution posterior information, including one or more of the following: floating-point solution standard deviation information, scene information, positioning status information, or carrier residual information; fixed solution information, including one or more of the following: fixed carrier residual information, fixed solution floating-point solution gap information, ambiguity floating-point solution and fixed solution gap information, initial number of satellites participating in fixing, or final number of satellites successfully fixed; inertial navigation association parameters may include one or more of the following: inertial navigation position and fixed solution gap information, inertial navigation position and floating-point solution gap information, or inertial navigation standard deviation information. Among these, satellite signal strength information represents the carrier-to-noise ratio of the currently received satellite signals, the number of satellites represents the number of available GNSS satellites in the current epoch, the number of satellite cycle slips represents the number of carrier cycle slips detected in each channel, thus reflecting observation continuity, and satellite elevation angle information represents the elevation angle information of each satellite in the current epoch. The posterior information of the floating-point solution characterizes the statistical properties and residual information of the solution results in floating-point state, and can be used to evaluate the stability of the solution. Examples include floating-point solution standard deviation information (representing the posterior standard deviation of the 3D coordinate floating-point solution), scene information (representing the classification label of the current environment, such as city, mountain, or tunnel), positioning status information (representing the RTK solution status flag, such as floating-point, fixed, or single-point), and carrier residual information (representing the statistical residual of the carrier phase observation for each satellite). Fixed solution information can be used as an indicator of the reliability and effectiveness of ambiguity fixing during the solution process. This includes: fixed carrier residual information (representing the carrier phase residual after fixing integer ambiguity), fixed-solution vs. floating-point solution difference information (representing the difference between the two solutions in coordinates), ambiguity floating-point solution vs. fixed solution difference information (representing the deviation between the floating-point estimate and the fixed integer in ambiguity space), initial number of participating satellites (representing the initial number of satellites used for ambiguity fixing), and final number of successfully fixed satellites (representing the actual number of satellites fixed), which can be used to evaluate the success rate. Inertial navigation information is used to describe the degree of matching between the INS / IMU system solution and the GNSS solution, including: the difference between the inertial navigation position and the fixed solution, the difference between the inertial navigation position and the floating-point solution, and the inertial navigation standard deviation, which are confidence indices of the estimated values ​​of position, velocity, etc., calculated by the inertial navigation system. When training the preset positioning accuracy evaluation model, a vector containing all the above-mentioned observation parameters can be formed using historically collected parameters from a mobile terminal as input samples. For example, 76 feature parameters and their corresponding ground truth values ​​(77 parameters in total) can be selected from the above five types of observation parameters to form a set of training data for a supervised learning model. The model is then trained using the first proportion of training data (original observation data) and validated using the second proportion of training data across multiple epochs, ultimately generating the preset supervised model.In practical use, the observation data collected by the mobile terminal at preset epochs can be input into the supervised model to determine whether the positioning information is accurate.

[0042] In some embodiments of this application, the method further includes: during the training of a preset positioning accuracy evaluation model, calling a tree model interpreter to determine the parameter contribution values ​​of multiple parameters among the observation parameters; and adjusting the type of historical observation parameters in the trained supervised model based on the parameter contribution values.

[0043] Continuing with the example of the 76 feature parameters in the five categories of observation parameters mentioned above, the actual number of observation parameters may not be limited to 76. Therefore, a tree-based model interpreter can be introduced during the training of the preset positioning accuracy evaluation model to help determine the contribution of each observation parameter to the model output and optimize the input feature set accordingly. For example, when using the XGBoost regression model, which is essentially composed of multiple regression trees, it has good nonlinear modeling and feature selection capabilities. After training, a model interpreter such as SHAP (SHapleyAdditiveexPlanations) or the feature_importances_ attribute built into XGBoost can be called to evaluate the importance of features within the model. The split gain, frequency of occurrence, and other indicators of each feature in each regression tree in the positioning accuracy evaluation model can be read, the average contribution of each observation parameter to the prediction result during the entire model training process can be calculated, and the contribution values ​​can be sorted to identify the observation parameters that have the greatest impact on the prediction result. For example, in a specific application scenario, it was found that "carrier residual information" and "difference between inertial navigation and fixed solution" are highly important, while "satellite cycle slip count" contributes almost nothing to the results in most samples. In further training, the "satellite cycle slip count" can be removed or assigned a lower weight to improve the overall model's generalization ability and convergence speed. Through the above implementation method, a feature contribution analysis mechanism can be introduced into the training stage of the preset positioning accuracy evaluation model to remove inefficient features, improve model training efficiency, and clarify the influence of each observation parameter on the positioning information detection results, which helps those skilled in the art to understand the causes of positioning errors.

[0044] In some embodiments of this application, the preset positioning accuracy evaluation model is an extreme gradient boosting model.

[0045] In some embodiments of this application, the preset positioning accuracy evaluation model is configured to be trained based on a regression model from the extreme gradient boosting model library.

[0046] In the above embodiments, the preset positioning accuracy evaluation model can adopt the extreme gradient boosting model XGBoost and call a regression model, such as training based on XGBRegressor in XGBoost, to establish a mapping relationship between observed parameters and actual positioning errors (true values), so that the model can predict the positioning accuracy corresponding to a set of observed parameters after inputting a set of observed parameters. When constructing the extreme gradient model, the number of trees and the maximum depth of each tree can be set as core parameters to reasonably fit the complex relationship of observed parameters during the positioning process and reasonably control the complexity of a single tree to prevent overfitting. Then, the learning rate, sample sampling ratio, and feature sampling ratio are set to adjust the contribution ratio of each new tree in each round and reduce the risk of overfitting. The initial selection of the above core parameters can be set based on experience, and subsequently adjusted continuously based on the performance of the validation set through automatic parameter tuning methods such as grid search, random search, and Bayesian optimization. XGBoost can gradually improve the overall fitting ability by iteratively stacking multiple weak regression trees, with each tree specifically learning the prediction residual of the previous time (i.e., the difference between the model prediction value and the true value). After training, the XGBoost model can be exported to standard file formats (such as JSON, PKL, etc.), providing a foundation for subsequent engineering deployment and integration. Furthermore, the preset localization accuracy evaluation model can be deployed in embedded devices; for example, tools like m2cgen can be used to convert the model into hardware-friendly formats such as C. In practical engineering applications, due to the long model building time and large model size, it is difficult to apply to systems with weak hardware processing capabilities (such as microcontrollers, single-chip microcomputers, etc.). Therefore, by calling `m2cgen.export_to_c`, the preset supervised model can be converted into a lighter machine-oriented language, such as C / C++, generating a model with smaller code and a wider range of applications.

[0047] Figure 2 The diagram illustrates a structural schematic of a satellite navigation positioning device provided in some embodiments of this application. The positioning accuracy detection device 200 includes: an acquisition unit 210, used to acquire observation parameters collected by a mobile terminal at a preset epoch, the observation parameters including GNSS parameters and inertial navigation associated parameters, the GNSS parameters including GNSS prior parameters and GNSS posterior parameters; a determination unit 220, used to input the observation parameters into a preset positioning accuracy evaluation model to obtain an accuracy evaluation value, wherein the positioning accuracy evaluation model is trained based on historical observation parameters collected at multiple epochs; and a judgment unit 230, used to determine the positioning result based on a floating-point solution of integer ambiguity when the positioning information is determined to be inaccurate, and to determine the positioning result based on a fixed solution of integer ambiguity when the positioning information is determined to be accurate.

[0048] In some embodiments, the positioning accuracy detection device 200 further includes: a training unit, used to determine the true value information of each group of historical observation parameters in each epoch based on historical observation parameters and standard observation parameters. The true value information is the difference between the observed navigation coordinates of the mobile terminal and the standard navigation coordinates, and each group of observation parameters corresponds to one true value information. The historical observation parameters and the true value information form the original observation data. The original observation data is input into the initial positioning accuracy evaluation model for training to generate a preset positioning accuracy evaluation model.

[0049] The above division of units is merely a logical functional division. In actual implementation, they can be fully or partially integrated into a single physical entity, or they can be physically separated. Furthermore, these units can be implemented by a processor calling software; for example, a satellite navigation positioning device includes a processor connected to a memory containing instructions. The processor calls the instructions stored in the memory to implement any of the above methods or to realize the functions of each unit. The processor can be, for example, a general-purpose processor, such as a central processing unit (CPU), and the memory can be internal or external to the device. Alternatively, these units can be implemented as hardware circuits. The functions of some or all units can be implemented through the design of the hardware circuit, which can be understood as one or more processors. For example, in some embodiments, the hardware circuit is an application-specific integrated circuit (ASIC), and the functions of some or all units are realized through the design of the logical relationships between the components within the circuit. In another implementation, the hardware circuit can be implemented using a programmable logic device (PLD), which can include a large number of logic gates. The logical relationships between the logic gates are configured through a configuration file, thereby realizing the functions of some or all units. All units of the above devices can be implemented entirely through processor calling programs, or entirely through hardware circuits, or partially through processor calling programs with the remaining parts implemented through hardware circuits.

[0050] Figure 3 The diagram illustrates the structure of a satellite navigation positioning device according to some embodiments of this application. The satellite navigation positioning device 300 includes: a processor 310 coupled to a memory 320, the memory 320 including instructions, which, when invoked by the processor 310, cause the processor 310 to execute the satellite navigation positioning method provided in the above embodiments.

[0051] In the above embodiments, the descriptions of each embodiment have their own emphasis. Parts not described in detail or in a particular embodiment can be referred to in the relevant descriptions of other embodiments. Furthermore, the above embodiments can be freely combined as needed.

Claims

1. A positioning method for satellite navigation, characterized in that, The positioning method includes: The system acquires observation parameters collected by the mobile terminal at a preset epoch. These observation parameters include GNSS parameters and inertial navigation (INS) related parameters. The GNSS parameters include GNSS prior parameters and GNSS posterior parameters. The GNSS prior parameters are input information obtained directly from observation geometry, signal quality, or ephemeris before integer ambiguity fixing or position resolution. The GNSS posterior parameters are derived indices obtained after integer ambiguity floating-point or fixed-point resolution through residual calculation, covariance matrix calculation, or statistical testing. The INS related parameters include at least one of the following: inertial navigation position and fixed-point solution difference information, inertial navigation position and floating-point solution difference information, and inertial navigation standard deviation information. The observation parameters are input into a preset positioning accuracy evaluation model to obtain an accuracy evaluation value, wherein the positioning accuracy evaluation model is trained based on historical observation parameters collected at multiple epochs; Based on the accuracy evaluation value, determine whether the positioning information represented by the current observation parameters is accurate; When it is determined that the positioning information is inaccurate, the positioning result is determined based on the floating-point solution of integer ambiguity; When the positioning information is determined to be accurate, the positioning result is determined based on the fixed solution of the integer ambiguity.

2. The positioning method according to claim 1, characterized in that, The GNSS posterior parameters include at least one of the following: fixed-solution floating-point solution difference information, ambiguity floating-point solution and fixed-solution difference information, and integer ambiguity change information.

3. The positioning method according to claim 1 or 2, characterized in that, Before inputting the observation parameters into the preset positioning accuracy evaluation model, the method further includes: When the inertial navigation correlation parameters meet the preset conditions, the process continues to input the observation parameters into the preset positioning accuracy evaluation model.

4. The positioning method according to claim 1, characterized in that, The positioning accuracy evaluation model is trained based on historical observation parameters collected at multiple epochs, including: Based on the historical observation parameters and standard observation parameters, the truth information of each group of historical observation parameters in each epoch is determined. The truth information is the difference between the observed navigation coordinates of the mobile terminal and the standard navigation coordinates. Each group of observation parameters corresponds to one set of truth information. The historical observation parameters and the truth information form the original observation data. The original observation data is input into the initial positioning accuracy evaluation model for training, thereby generating the preset positioning accuracy evaluation model.

5. The positioning method according to claim 4, characterized in that, The step of inputting the original observation data into the initial supervised model for training includes: The original observation data is input into the initial positioning accuracy evaluation model based on a preset training ratio. The preset training ratio includes a first ratio and a second ratio, wherein the first ratio is greater than or equal to the second ratio. The generation of the preset positioning accuracy evaluation model includes: In the first proportion of raw observation data, the initial positioning accuracy evaluation model is trained based on the historical observation parameters and the true value information to generate the first positioning accuracy evaluation model; Input the historical observation parameters from the second proportion of the original observation data into the first positioning accuracy evaluation model to determine the first result value; Based on the true value information in the first result value and the original observation data of the second ratio, the preset positioning accuracy evaluation model is determined and generated.

6. The positioning method according to claim 5, characterized in that, Also includes: During the training of the preset positioning accuracy evaluation model, the tree model interpreter is invoked to determine the parameter contribution values ​​of multiple parameters among the observation parameters; Based on the contribution value of the parameters, the type of the historical observation parameters in the training of the positioning accuracy evaluation model is adjusted.

7. The positioning method according to claim 1, characterized in that, The preset positioning accuracy evaluation model is an extreme gradient boosting model.

8. A satellite navigation positioning device, characterized in that, Includes: a processor for coupling a memory, the memory including instructions that, when invoked by the processor, cause the processor to perform the positioning method as described in any one of claims 1-7.

9. A computer-readable storage medium, characterized in that, Includes instructions stored thereon, wherein, when the instructions are executed by a processor, the positioning method as described in any one of claims 1-7 is performed.

Images