Navigation satellite positioning result identification method and device, equipment and storage medium

Abstract

A navigation satellite positioning result identification method, device and equipment and a storage medium are disclosed. The method comprises: obtaining target positioning data, the target positioning data comprising a positioning state flag; when the positioning state flag is a fixed solution, inputting the target positioning data into an anomaly detection model to obtain an anomaly score of the target positioning data, the anomaly score being used to represent a deviation degree of a feature data distribution of the fixed solution of the target positioning data from a feature data distribution of a fixed solution of reference positioning data; when it is determined that the anomaly score satisfies an anomaly score screening condition, determining an error overrun confidence of the target positioning data based on a supervised classification model, wherein the error overrun confidence is used to represent a deviation degree of a positioning error of the fixed solution of the target positioning data from a preset positioning error threshold; and determining an error anomaly state of the fixed solution of the target positioning data based on the error overrun confidence. The present application has the technical effect of improving the positioning result identification accuracy of a navigation satellite system.

Description

Technical Field

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

[0002] With the widespread application of Global Navigation Satellite System (GNSS), in practical navigation and positioning, positioning data is typically extracted by analyzing the messages output by the GNSS receiver to determine the positioning result of the GNSS receiver carrier. However, when facing complex and ever-changing real-world operating environments, positioning data is often affected by multipath effects and non-line-of-sight interference, such as urban canyons or tree-lined obstructions. This makes it easy for the GNSS receiver to output anomalies where the positioning status appears fixed, but the actual positioning error is significant. Blindly accepting these anomalies can easily affect the positioning results of the GNSS system, and even lead to navigation failure. Therefore, the reliability of GNSS positioning results deserves attention. Summary of the Invention

[0003] In view of this, the present disclosure provides a method, apparatus, device and storage medium for identifying navigation satellite positioning results, in order to improve the accuracy of identifying positioning results of global navigation satellite systems.

[0004] Firstly, a method for identifying navigation satellite positioning results is provided, comprising: acquiring target positioning data, the target positioning data including a positioning status flag; when the positioning status flag is a fixed solution, inputting the target positioning data into an anomaly detection model to obtain an anomaly score for the target positioning data, wherein the anomaly score is used to characterize the degree of deviation of the feature data distribution of the fixed solution of the target positioning data from the feature data distribution of the fixed solution of the reference positioning data; when it is determined that the anomaly score meets the anomaly score screening conditions, determining the error over-limit confidence level of the target positioning data based on a supervised classification model, wherein the error over-limit confidence level is used to characterize the degree to which the positioning error of the fixed solution of the target positioning data deviates from a preset positioning error threshold; and determining the error anomaly state of the fixed solution of the target positioning data based on the error over-limit confidence level.

[0005] The above method for identifying navigation satellite positioning results uses an anomaly detection model to identify anomalous scores in the current fixed solution data that deviate from the normal distribution, intercepting data that affects the final positioning result. Then, a supervised classification model is used to determine the error exceeding the confidence level of the current anomalous score fixed solution. Finally, combined with a pre-set confidence threshold, this method determines the possible deviation between the positioning result represented by the current positioning data's fixed solution and the actual positioning result. This identification logic, which progressively advances from data distribution deviation to physical error probability, avoids the enormous computational cost of complex calculations on all data and identifies anomalous fixed solutions that affect the reliability of the positioning result. By outputting the error anomaly status, it effectively prevents navigation drift or path control errors caused by blindly adopting low-confidence fixed solutions when determining subsequent positioning results, thus improving the positioning reliability and safety of the carrier equipped with the navigation satellite positioning system in complex environments.

[0006] Optionally, after determining the error anomaly state of the fixed solution of the positioning data based on the error excess confidence level, the method further includes: when the error anomaly state is abnormal, reducing the weight of the fixed solution of the positioning data in positioning fusion based on the Global Navigation Satellite System.

[0007] Optionally, it also includes: when the error anomaly status of the current epoch is abnormal, maintaining the positioning result of the Global Navigation Satellite System in the previous epoch in the current epoch.

[0008] Optionally, the anomaly detection model is constructed based on the following steps: acquiring sample positioning data and corresponding true positioning error data; marking the fixed solutions in the sample positioning data with positioning errors based on the true positioning error data to determine normal fixed solution data; training and establishing the anomaly detection model based on the feature data distribution of normal fixed solution samples; the supervised classification model is constructed based on the following steps: inputting the fixed solutions in the sample positioning data into the anomaly detection model to determine the score anomaly data that meets the anomaly score screening conditions; training and establishing the supervised classification model based on the score anomaly data and the positioning error markings of the score anomaly data.

[0009] In the embodiments of the above navigation satellite positioning result identification method, the extracted normal fixed solution samples are used to train the anomaly detection model. This model can fit the distribution boundary of normal data, thereby possessing the ability to detect abnormal distributions of fixed solutions caused by various complex environments compared to historical normal fixed solutions. Furthermore, the sample positioning data is filtered through the pre-trained anomaly detection model, and the extracted scoring anomaly data is used as the training set for the supervised classification model. Then, the supervised classification model is trained in conjunction with real positioning error labels, guiding the supervised classification model to establish the ability to identify the error judgment between the physical positioning results represented by the anomaly fixed solutions and the actual physical positioning.

[0010] Optionally, the fixed solution in the sample positioning data is marked with positioning error based on the real positioning error data, including: extracting the first time synchronization information of the sample positioning data and the second time synchronization information of the real positioning error data; aligning the sample positioning data with the real positioning error data in time based on the first time synchronization information and the second time synchronization information to determine the fixed solution corresponding to the real positioning error data; and marking the fixed solution with positioning error based on the time-aligned real positioning error data.

[0011] Optionally, acquiring target positioning data includes: acquiring initial positioning data, parsing the initial positioning data, and extracting target positioning data related to positioning quality. The target positioning data includes at least one of the following: positioning status flags, number of satellites participating in positioning, position accuracy attenuation factor, root mean square error, standard deviation of major and minor axes of the error ellipsoid, and standard deviation of position error.

[0012] Optionally, after extracting the target positioning data related to positioning quality, the method further includes: preprocessing the target positioning data, wherein the preprocessing includes at least one of the following operations: outlier removal, data smoothing, and missing value completion; and inputting the target positioning data into the anomaly detection model, including: inputting the preprocessed target positioning data into the anomaly detection model.

[0013] Secondly, a device for identifying navigation satellite positioning results is provided, comprising: an acquisition unit for acquiring target positioning data, the target positioning data including a positioning status flag; an anomaly scoring determination unit for inputting the target positioning data into an anomaly detection model to obtain an anomaly score for the target positioning data when the positioning status flag is a fixed solution, wherein the anomaly score is used to characterize the degree of deviation of the feature data distribution of the fixed solution of the target positioning data from the feature data distribution of the fixed solution of the reference positioning data; an error exceeding confidence level determination unit for determining the error exceeding confidence level of the target positioning data based on a supervised classification model when the anomaly score is determined to meet the anomaly score screening conditions, wherein the error exceeding confidence level is used to characterize the degree of deviation of the positioning error of the fixed solution of the target positioning data from a preset positioning error threshold; and an anomaly state determination unit for determining the error anomaly state of the fixed solution of the target positioning data based on the error exceeding confidence level.

[0014] Thirdly, a computer device is provided, including a memory and a processor, wherein when the memory is read by the processor, the method for identifying navigation satellite positioning results provided in the first aspect above is executed.

[0015] Fourthly, a computer-readable storage medium is provided, including instructions stored thereon, wherein when the instructions are executed by a processor, the method for recognizing navigation satellite positioning results provided in the first aspect above is executed. Attached Figure Description

[0016] The accompanying drawings used in the description of the embodiments of this disclosure are briefly introduced below: Figure 1 A flowchart illustrating a method for identifying navigation satellite positioning results provided in some embodiments of this application is shown. Figure 2 The diagram shows a flowchart illustrating a training method for an anomaly detection model provided in some embodiments of this application. Figure 3 The diagram shows a flowchart illustrating a training method for a supervised classification model provided in some embodiments of this application. Figure 4 The diagram shows a schematic representation of a navigation satellite positioning result identification device provided in some embodiments of this application. Detailed Implementation

[0017] 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.

[0018] 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.

[0019] 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.

[0020] Global Navigation Satellite Systems (GNSS), as a crucial space information infrastructure, are widely used in vehicle navigation, autonomous driving, precision agriculture, and surveying and mapping. During GNSS positioning calculations, the results are typically categorized into single-point solutions, floating-point solutions, and fixed solutions. Fixed solutions, due to the successful fixation of carrier phase ambiguity, are generally considered to have the highest positioning accuracy. However, in complex real-world environments, such as densely populated urban areas, under overpasses, tunnel entrances, or scenes obstructed by trees, GNSS signals are susceptible to multipath effects, non-line-of-sight propagation, and localized electromagnetic interference. In these scenarios, even if the GNSS solution status is identified as a fixed solution, the positioning results may still exhibit significant deviations. For example, in urban canyon environments, reflected signals can cause systematic errors in pseudorange observations, resulting in a horizontal positioning deviation of more than one meter for the fixed solution. Current technologies typically use the corresponding positioning results directly based on the fixed solution status marker, lacking a further mechanism for evaluating the internal quality of the fixed solution. Some methods rely solely on empirical judgments based on single indicators such as the Dilution of Precision (DOP) or the number of available satellites, making it difficult to accurately identify fixed solutions with large errors. This can easily lead to abrupt changes in positioning trajectories or decreased navigation stability. If traditional fixed solutions are still relied upon, the carrier equipped with the GNSS receiver will be unable to identify and adapt the authenticity of fixed solutions under different obstruction environments. To address this, this application proposes a method, apparatus, device, and storage medium for identifying navigation satellite positioning results. By acquiring positioning data obtained from the Global Navigation Satellite System (GNSS), and based on a two-stage anomaly detection model and a supervised classification model for joint evaluation, anomaly identification can be specifically performed on the fixed solutions of the raw positioning data output by the GNSS receiver, thereby improving the reliability of GNSS positioning results.

[0021] The following description is in conjunction with the accompanying drawings: Figure 1 The diagram illustrates a flowchart of a method for identifying navigation satellite positioning results provided in some embodiments of this application. This identification method includes at least the following steps: S110: Obtain target positioning data, which includes positioning status flags; S120: When the positioning status flag is a fixed solution, the target positioning data is input into the anomaly detection model to obtain the anomaly score of the target positioning data. The anomaly score is used to characterize the degree of deviation of the feature data distribution of the fixed solution of the target positioning data from the feature data distribution of the fixed solution of the reference positioning data. S130: When it is determined that the abnormal score meets the abnormal score screening conditions, the error over-limit confidence level of the target positioning data is determined based on the supervised classification model. The error over-limit confidence level is used to characterize the degree to which the positioning error of the fixed solution of the target positioning data deviates from the preset positioning error threshold. S140: Error anomaly state of fixed solution for target positioning data determined based on error over-limit confidence level.

[0022] In practical navigation and positioning scenarios, GNSS receivers continuously receive and process satellite signals to output positioning data. Global Navigation Satellite Systems (GNSS) can acquire this positioning data and use it as the basis for determining the final positioning result. In this application, the subsequent identification process can be triggered when the positioning status flag of the received target positioning data explicitly indicates that the positioning data has a fixed solution. For positioning data with inherently low accuracy, such as floating-point solutions or single-point positioning, conventional logic can be applied, for example, not using it as the basis for determining the positioning result. After confirming that the solution result of the current positioning data is a fixed solution, the positioning data can be input into a pre-trained anomaly detection model. This anomaly detection model can be trained based on fixed solution data from historical real-world scenarios where the error between the positioning result and the actual positioning result is within a reasonable range, thus representing the standard form of a normal fixed solution in a multi-dimensional data space. After processing, the anomaly detection model can output a quantitative index, namely an anomaly score. This anomaly score can intuitively represent the degree to which the distribution of feature data in the current epoch of target positioning data deviates from the distribution of feature data in historical normal positioning data. A higher anomaly score indicates that the positioning data corresponding to the current fixed solution deviates from the expected data distribution pattern under a normal fixed solution. Furthermore, to achieve a balance between computational efficiency and recognition accuracy, preset anomaly score filtering conditions can be used to segment the data. When the anomaly score reaches or exceeds the limit of these filtering conditions, it means that the current target positioning data not only deviates from the normal data distribution, but the current fixed solution may also affect the accuracy of the final positioning result, thus requiring further investigation.

[0023] In actual fixed-solution error identification, anomalous data with positioning errors exceeding the preset positioning error threshold often account for a small percentage of the massive overall fixed-solution sample, for example, only about 0.05%. However, there is still a significant imbalance in the distribution of positive and negative samples. If positioning quality features are extracted solely from positioning data, there may be a certain degree of feature overlap between normal fixed solutions and anomalous fixed solutions with large actual errors. Directly using a single supervised classification model to determine whether the fixed solution positioning error exceeds the preset threshold can easily lead to unstable classification boundaries, insufficient generalization ability, and a high misclassification rate. Furthermore, in complex urban canyons and other occlusion environments, some fixed-solution data may exhibit certain anomalies in statistical features due to multipath effects, but their actual physical positioning errors may still be within the acceptable safety range of the downstream system due to the algorithm's internal smoothing effect. If a simple one-size-fits-all strategy is adopted, directly classifying all data with statistically abnormal features as high-error anomalous fixed solutions, it will inevitably trigger safety suppression or downweighting mechanisms frequently and erroneously during real-time dynamic navigation, thereby severely damaging the continuity of terminal positioning results and the stability of the overall system. Based on the above, this application adopts an identification mechanism that combines anomaly detection and supervised classification. The anomaly detection stage identifies the risk of statistical distribution deviation, and the supervised classification stage further makes a substantive judgment on physical errors. This greatly improves the accuracy of error identification while minimizing the risk of misjudgment.

[0024] For location data that meets the anomaly scoring screening criteria, it can be further input into a pre-defined supervised classification model. During the training phase, the supervised classification model can absorb a large number of sample patterns with real physical error labels. After processing, the model outputs an error over-limit confidence score. Unlike anomaly scoring, the error over-limit confidence score has a certain physical orientation, characterizing the probability or severity of the actual positioning error of the fixed solution in real physical space exceeding the system's tolerable safety threshold (pre-defined positioning error threshold) under the current complex interference environment. Based on the comparison results, the system ultimately determines whether the positioning data currently labeled as a fixed solution is in a safe and normal state at the error level, or an anomaly disguised as a fixed solution. This anomaly state can serve as a reference for the final determination of the positioning result, for subsequent navigation result determination or multi-source sensor fusion algorithms.

[0025] The above embodiments overcome the limitations of traditional navigation applications that rely solely on a single fixed solution for positioning result quality control. By using an anomaly detection model to identify abnormal scores where the current fixed solution data deviates from the normal distribution, data affecting the final positioning result is intercepted. Then, a supervised classification model is used to determine the error exceeding the confidence level of the current abnormal score fixed solution. This, combined with a pre-set confidence threshold, determines the potential deviation between the positioning result represented by the current positioning data's fixed solution and the actual positioning result. This identification logic, which progressively advances from data distribution deviation to physical error probability, avoids the enormous computational cost of complex calculations on all data while identifying abnormal fixed solutions that affect the reliability of the positioning result. Through the final output of error anomaly states, navigation drift or path control errors caused by blindly adopting low-confidence fixed solutions are effectively prevented during subsequent positioning result determination, thus improving the positioning reliability and safety of carriers equipped with navigation satellite positioning systems in complex environments.

[0026] In some embodiments of this application, after determining the error anomaly state of the fixed solution of the positioning data based on the error exceedance confidence level, the method further includes: when the error anomaly state indicates that the fixed solution of the positioning data is an abnormal fixed solution, reducing the weight of the fixed solution of the positioning data in positioning fusion based on the Global Navigation Satellite System. Alternatively, maintaining the positioning result of the Global Navigation Satellite System in the previous epoch at the current epoch.

[0027] After outputting the final error anomaly status, this status will be used as a positioning reference in the subsequent determination of the final navigation result. In conventional practice, since the fixed solution itself is already considered a valid positioning result, subsequent integrated navigation algorithms, such as those fusing satellite navigation and inertial navigation units, typically assign a higher observation weight to this fixed solution by default. This means the system can rely on the current positioning result. However, once the navigation satellite positioning result identification method provided in this application issues a warning, explicitly indicating that the error anomaly status of the current positioning data is an anomalous fixed solution, the system can trigger a safety mechanism to reduce the weight of this anomalous fixed solution in the integrated navigation algorithm, prompting the algorithm to rely more on positioning results from other sensors, such as inertial navigation. Furthermore, in some non-integrated navigation pure satellite navigation systems, if an anomalous fixed solution is detected, the system can directly reject the high-risk positioning coordinates generated in the current epoch and instead maintain and use the positioning result of the global navigation satellite system in the previous epoch (i.e., the time when it was previously verified as a normal fixed solution). Therefore, in complex environments such as urban canyons or under overpasses, this application demonstrates that the navigation satellite positioning system no longer passively bears the impact of errors, but rather improves the safety and reliability of positioning in harsh operating environments by actively reducing weights or maintaining epochs, without increasing any additional expensive hardware costs.

[0028] In some embodiments of this application, acquiring global target positioning data includes: acquiring initial positioning data; parsing the initial positioning data and extracting target positioning data related to positioning quality, wherein the target positioning data includes at least one of the following: positioning status flags, the number of satellites participating in positioning in the global navigation satellite system, position accuracy attenuation factor, root mean square error, standard deviation of the major and minor axes of the error ellipsoid, and standard deviation of the position error.

[0029] In the above embodiments, the received positioning data from the Global Navigation Satellite System can be presented as data conforming to the NMEA (National Marine Electronics Association) standard protocol. When parsing the target positioning data, positioning status flags can be extracted to trigger subsequent anomaly identification logic. The number of satellites participating in the positioning can also be extracted to reflect the available observation redundancy at the current epoch; generally, the fewer the number of satellites, the higher the risk of error fixation. Furthermore, a position accuracy attenuation factor characterizing the geometric amplification effect of the current observation satellite constellation distribution topology on positioning accuracy can be extracted. When the satellite distribution is too concentrated, the value of this position accuracy attenuation factor will increase significantly, suggesting a potential risk of multipath effects being amplified. In addition, to further determine the convergence state, the root mean square error can be extracted to assess the overall fluctuation level of the pseudorange residual. Simultaneously, the standard deviations of the major and minor axes of the error ellipsoid and the standard deviations of the position error (such as the standard deviations of latitude and longitude) can be extracted to accurately depict the spatial three-dimensional probability distribution of the current position confidence level by the Kalman filter and other solution algorithms within the GNSS receiver.

[0030] In some embodiments of this application, obtaining positioning data from a global navigation satellite system further includes: preprocessing the target positioning data, wherein the preprocessing includes at least one of the following operations: outlier removal, data smoothing, and missing value completion; inputting the target positioning data into an anomaly detection model includes: inputting the preprocessed target positioning data into the anomaly detection model.

[0031] During the movement of a carrier equipped with a navigation satellite positioning system, the target positioning data extracted directly from the underlying messages inevitably contains sporadic noise or data frame breaks due to the complex external physical environment and the stability of the internal hardware communication links. Directly inputting this positioning data into an anomaly detection model can easily lead to inaccurate anomaly scores, thus affecting the subsequent determination of error anomaly states. Therefore, outlier removal can be performed on the acquired target positioning data to filter out data that clearly violates physical laws due to hardware failures or signal anomalies. Simultaneously, data smoothing can be used to smooth out fluctuations in positioning values ​​across consecutive epochs, such as root mean square error and position accuracy decay factor, filtering out random white noise from the environment and highlighting the true convergence or divergence trend of the positioning data over time. Furthermore, for positioning data with blank fields due to communication packet loss or message output delays, missing value completion processing can be performed to ensure the integrity and temporal continuity of the positioning data fed into the anomaly detection model. The above preprocessing process can be performed by selecting one, two, or three of the operations. By performing outlier removal, data smoothing, or missing value completion in advance, the negative impact of occasional noise and communication anomalies on subsequent model calculations can be reduced directly from the source of the data input.

[0032] In some embodiments of this application, determining that an abnormal score meets the abnormal score screening conditions includes: determining that an abnormal score meets the abnormal score screening conditions when the abnormal score is greater than or equal to a preset abnormal score threshold; and determining the positioning result of the Global Navigation Satellite System based on the positioning data when the abnormal score is less than the preset abnormal score threshold.

[0033] After the anomaly score is determined by the anomaly detection model, if the anomaly score is less than the preset anomaly score threshold, it indicates that the multidimensional features of the current fixed solution are showing a healthy convergence state, and its data distribution is similar to or consistent with the normal fixed solutions in historical safe scenarios. At this time, the positioning data can be determined as a high-confidence normal fixed solution, and there is no need to continue to process it into the supervised classification model. Instead, the positioning data of the Global Navigation Satellite System can be directly trusted and output, and it can be used by business logic, such as route planning or map navigation, thereby ensuring that navigation data for the vast majority of normal driving sections can be processed quickly. Conversely, if the anomaly score is greater than or equal to the preset anomaly score threshold, it indicates that the current feature distribution has substantially deviated from the distribution of the normal fixed solutions in historical safe scenarios. At this time, the positioning data can be determined to meet the anomaly score screening conditions, and the positioning data can be further marked as potentially risky suspected anomaly data. It is necessary to send it into the subsequent supervised classification model for substantive review of whether the physical positioning error exceeds the limit. The preset anomaly scoring threshold can be set based on the distribution of historical location data training data. For example, it can be calibrated statistically, by statistically analyzing the anomaly scoring distribution interval of a large number of normal fixed solution samples during the training or validation phase of the anomaly detection model, and determining the preset anomaly scoring threshold at a specific long-tailed high quantile. Alternatively, the threshold can be calibrated based on the system's false alarm rate and false alarm rate, i.e., by adjusting and determining the preset anomaly scoring threshold on the validation set according to the tolerance threshold for erroneous fixed solutions. If security requirements are high, the preset anomaly scoring threshold can be appropriately lowered to allow more data to enter the subsequent supervised classification network for detection; if more emphasis is placed on saving computational resources and real-time performance, the preset anomaly scoring threshold can be appropriately increased to intercept only severely distorted location data.

[0034] In some embodiments of this application, the error anomaly state of the fixed solution of the positioning data is determined based on the error excess confidence level and a preset confidence threshold, including: when the error excess confidence level is greater than or equal to the preset confidence threshold, the error anomaly state of the fixed solution of the positioning data is determined to be an abnormal fixed solution; when the error excess confidence level is less than the preset confidence threshold, the error anomaly state of the fixed solution of the positioning data is determined to be a normal fixed solution.

[0035] After deep feature profiling by the supervised classification model, an error over-limit confidence score can be obtained to quantify error risk. This score can be represented as a continuous probability value, such as a value distributed between 0 and 1. The closer this value is to the upper limit, the more confident the supervised classification model is that the current positioning error has substantially exceeded the tolerable positioning deviation. When the comparison finds that the error over-limit confidence score is greater than or equal to the preset confidence threshold, it means that the supervised classification model, based on multi-dimensional features, has given a clear warning, confirming that the actual physical positioning error of the current positioning data is highly likely to have exceeded the limit. At this point, the positioning data can no longer be trusted, and its abnormal error state can be identified as an abnormal fixed solution. This part of the data is highly likely to cause potential hazards that could easily lead to vehicle deviation or sudden trajectory changes, and needs to be strictly eliminated or downweighted in downstream operations. Conversely, when the error over-limit confidence score is less than the preset confidence threshold, the abnormal error state of the positioning data can be identified as a normal fixed solution. The above decision logic can play a corrective role in complex urban occlusion environments. Positioning data identified as normal fixed solutions at this stage, due to interference from multipath effects during acquisition, exhibited some degree of statistical distortion in the initial screening stage, thus being transmitted to the current stage. However, the supervised classification model, based on prior knowledge of the true error, determines that this distortion has not yet caused the underlying actual physical positioning error to exceed an acceptable range, and therefore can be ultimately confirmed as a normal fixed solution. By comparing the error over-limit confidence level with a pre-set confidence threshold, on the one hand, it can filter out the truly abnormal fixed solutions that would cause navigation failure, safeguarding the system's safety baseline; on the other hand, it can salvage positioning data with flawed statistical characteristics but still meeting physical accuracy standards, avoiding the indiscriminate classification of all statistically abnormal samples as high-error fixed solutions, preventing frequent and unnecessary triggering of the system's safety suppression mechanism during navigation, and ensuring positioning safety while maintaining the continuity of satellite positioning output and the stability of carrier positioning.

[0036] Figure 2 The diagram illustrates a flowchart of a training method for an anomaly detection model provided in some embodiments of this application. The training method includes: S210: Obtain sample positioning data and corresponding actual positioning error data; S220: Mark the positioning error of the fixed solution in the sample positioning data based on the real positioning error data, and determine the normal fixed solution data; S230: Train and establish an anomaly detection model based on the feature data distribution of normal fixed solution samples; Figure 3 The diagram illustrates a flowchart of a training method for a supervised classification model provided in some embodiments of this application. The training method includes: S310: Input the fixed solution in the sample location data into the anomaly detection model to determine the score anomaly data that meets the anomaly score screening conditions; S320: Based on the rating anomaly data and the localization error label of the rating anomaly data, train and establish a supervised classification model.

[0037] In the training and preparation phase, sample positioning data for model learning can be constructed first. This could be historical positioning data collected by a receiver in various complex real-world scenarios such as urban canyons and tree-lined roads. Simultaneously with the sample positioning data collection, corresponding real positioning error data can be obtained. After determining the sample positioning data, the physical deviation distance between the fixed solution sample and the real positioning error data can be calculated. Based on this real physical deviation and a tolerable physical deviation threshold, each fixed solution is labeled with a precise error category. For example, fixed solutions with a deviation within 1 meter are identified as normal fixed solutions, while those with a deviation greater than or equal to 1 meter are identified as abnormal fixed solutions. Through labeling, normal fixed solution data with small physical errors that meet safety requirements can be extracted from the sample positioning data, thus initiating the training process of the anomaly detection model. Since abnormal fixed solutions not only account for a very small proportion in actual engineering but also have an exhaustive range of distortion patterns caused by different occlusion environments, this application can employ single-class learning or unsupervised / semi-supervised learning approaches to encourage the anomaly detection model to focus on learning the inherent distribution characteristics and boundaries of absolutely healthy normal fixed solution samples in the multi-dimensional feature space. The training objective of this anomaly detection model is not to memorize the appearance of specific errors, but rather to firmly remember the patterns that normal fixed solution data should exhibit. This enables the trained anomaly detection model to score any location data that deviates from the distribution pattern of normal fixed solutions. After the anomaly detection model has been initially trained and the distribution boundaries of normal fixed solution data have been established, the original sample location data can be re-input into the anomaly detection model to calculate and output the corresponding anomaly score for each fixed solution sample. Then, through preset anomaly score screening conditions, those score-high anomaly data that exhibit distortion and deviation in statistical characteristics are selected. These constitute the location data in the entire sample location data that, although they are fixed solutions, contain location data that deviate significantly from the true location results.

[0038] Furthermore, the aforementioned rating anomaly data, along with the corresponding localization error labels determined in step S220, can be input into the supervised classification model. Guided by the supervised learning mechanism, the model learns how to map subtle differences in multidimensional features to whether the actual localization error truly exceeds the limit, thereby completing the convergence and solidification of network weight parameters, and ultimately training and establishing a supervised classification model with physical discrimination capabilities. The above anomaly detection model can be implemented using unsupervised or semi-supervised learning algorithms such as Isolation Forest, One-Class SVM, or Autoencoder. The above supervised classification model is mainly based on deep training of the selected rating anomaly data and its corresponding localization error labels, and can be implemented using Gradient Boosting Decision Tree, Neural Networks, or other related supervised learning methods. By employing the specific algorithm models described above, the two-stage model can improve the discrimination accuracy for high-error fixed solutions while ensuring computational efficiency.

[0039] In some embodiments of this application, determining normal fixed solution data may include: comparing the actual positioning error data corresponding to the fixed solution in the sample positioning data with a preset positioning error threshold; when the actual positioning error data is less than or equal to the preset positioning error threshold, marking the fixed solution of the sample positioning data as a normal fixed solution; when the actual positioning error data is greater than the preset positioning error threshold, marking the fixed solution of the sample positioning data as an abnormal fixed solution.

[0040] This application utilizes extracted normal fixed solution samples to train an anomaly detection model, fitting the distribution boundary of normal data, thereby enabling the detection of anomalies in the distribution of fixed solutions caused by various complex environments compared to historical normal fixed solutions. Furthermore, the pre-trained anomaly detection model filters sample positioning data, using the extracted scoring anomaly data as the training set for a supervised classification model. This supervised classification model is then trained using real positioning error labels, guiding it to establish the ability to identify the error between the physical positioning result represented by the anomaly fixed solution and the actual physical positioning. Based on the training method of this application, the anomaly detection model and supervised classification model can improve the resolution capability of global navigation satellite systems in complex environments when identifying positioning results.

[0041] In some embodiments of this application, the fixed solution in the sample positioning data is marked with positioning error based on the real positioning error data, including: extracting the first time synchronization information of the sample positioning dataset and the second time synchronization information of the real positioning error data; and aligning the fixed solution in the sample positioning dataset with the corresponding real positioning error data in time based on the first time synchronization information and the second time synchronization information to determine the fixed solution corresponding to the real positioning error data.

[0042] Figure 4 The diagram illustrates a schematic of a navigation satellite positioning result identification device provided in some embodiments of this application. The identification device 400 includes: an acquisition unit 410 for acquiring positioning data from a global navigation satellite system; an anomaly scoring determination unit 420 for determining an anomaly score of the positioning data based on the positioning data and a preset anomaly detection model when the positioning status flag in the positioning data is a fixed solution, wherein the anomaly score characterizes the degree of deviation of the feature data distribution of the fixed solution of the current positioning data from the feature data distribution of the fixed solution of normal positioning data; an error exceeding confidence level determination unit 430 for determining an error exceeding confidence level based on the positioning data and a preset supervised classification model when the anomaly score is determined to meet the anomaly score screening conditions, wherein the error exceeding confidence level characterizes the degree to which the positioning error of the fixed solution of the positioning data deviates from a preset positioning error threshold; and an anomaly state determination unit 440 for determining the error anomaly state of the fixed solution of the positioning data based on the error exceeding confidence level and a preset confidence threshold.

[0043] Based on the same technical concept, this application also provides a computer device, including a memory and a processor, wherein when the memory is read by the processor, the method for identifying navigation satellite positioning results provided in the above embodiments is executed.

[0044] Based on the same technical concept, this application also provides a computer-readable storage medium including instructions stored thereon, wherein, when the instructions are executed by a processor, the navigation satellite positioning result identification method provided in the above embodiments is executed.

[0045] 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 method for identifying navigation satellite positioning results, characterized in that, include: Acquire target positioning data, wherein the target positioning data includes positioning status flags; When the positioning status flag is a fixed solution, the target positioning data is input into the anomaly detection model to obtain an anomaly score for the target positioning data. The anomaly score is used to characterize the degree of deviation of the feature data distribution of the fixed solution of the target positioning data from the feature data distribution of the fixed solution of the reference positioning data. When it is determined that the abnormal score meets the abnormal score screening conditions, the error over-limit confidence level of the target positioning data is determined based on the supervised classification model, wherein the error over-limit confidence level is used to characterize the degree to which the positioning error of the fixed solution of the target positioning data deviates from the preset positioning error threshold; The error anomaly state of the fixed solution of the target positioning data is determined based on the error over-limit confidence level.

2. The method for identifying navigation satellite positioning results according to claim 1, characterized in that, After determining the error anomaly state of the fixed solution of the positioning data based on the error excess confidence level, the method further includes: When the error anomaly state is abnormal, the weight of the fixed solution of the positioning data in positioning fusion based on the Global Navigation Satellite System is reduced.

3. The method for identifying navigation satellite positioning results according to claim 2, characterized in that, Also includes: When the error anomaly state of the current epoch is abnormal, the positioning result of the global navigation satellite system in the previous epoch is maintained in the current epoch.

4. The method for identifying navigation satellite positioning results according to any one of claims 1 to 3, characterized in that, The anomaly detection model is constructed based on the following steps: Obtain sample positioning data and corresponding actual positioning error data; The fixed solutions in the sample positioning data are marked with positioning errors based on the actual positioning error data to determine the normal fixed solution data; The anomaly detection model is trained and established based on the feature data distribution of the normal fixed solution samples. The supervised classification model is constructed based on the following steps: The fixed solution in the sample localization data is input into the anomaly detection model to determine the score anomaly data that meets the anomaly score screening conditions. The supervised classification model is trained and established based on the rating anomaly data and the location error label of the rating anomaly data.

5. The method for identifying navigation satellite positioning results according to claim 4, characterized in that, The step of marking the fixed solution in the sample positioning data with positioning error based on the actual positioning error data includes: Extract the first time synchronization information of the sample positioning data and the second time synchronization information of the actual positioning error data; Based on the first time synchronization information and the second time synchronization information, the sample positioning data and the real positioning error data are time-aligned to determine the fixed solution corresponding to the real positioning error data. Based on the time-aligned true positioning error data, the fixed solution is marked with the positioning error.

6. The method for identifying navigation satellite positioning results according to claim 1, characterized in that, The acquisition of target location data includes: Obtain initial location data; The initial positioning data is parsed, and target positioning data related to positioning quality is extracted. The target positioning data includes at least one of the following: the positioning status flag, the number of satellites participating in the positioning, the position accuracy attenuation factor, the root mean square error, the standard deviation of the major and minor axes of the error ellipsoid, and the standard deviation of the position error.

7. The method for identifying navigation satellite positioning results according to claim 6, characterized in that, After extracting the target positioning data related to positioning quality, the process also includes: The target location data is preprocessed, wherein the preprocessing includes at least one of the following operations: outlier removal, data smoothing, and missing value completion. The step of inputting the target location data into the anomaly detection model includes: The preprocessed target location data is input into the anomaly detection model.

8. A device for identifying navigation satellite positioning results, characterized in that, include: An acquisition unit is used to acquire target positioning data, wherein the target positioning data includes a positioning status flag; An anomaly scoring determination unit is used to input the target positioning data into an anomaly detection model when the positioning status flag is a fixed solution, and obtain an anomaly score for the target positioning data. The anomaly score is used to characterize the degree of deviation of the feature data distribution of the fixed solution of the target positioning data from the feature data distribution of the fixed solution of the reference positioning data. An error exceedance confidence determination unit is used to determine the error exceedance confidence of the target positioning data based on a supervised classification model when it is determined that the abnormal score meets the abnormal score screening conditions. The error exceedance confidence is used to characterize the degree to which the positioning error of the fixed solution of the target positioning data deviates from a preset positioning error threshold. An abnormal state determination unit is used to determine the error abnormal state of the fixed solution of the target positioning data based on the error exceedance confidence level.

9. A computer device, characterized in that, It includes a memory and a processor, wherein when the memory is read by the processor, the method for identifying navigation satellite positioning results as described in any one of claims 1-7 is executed.

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

Images