GNSS positioning algorithm screening method and device, equipment and storage medium

Abstract

The invention discloses a GNSS positioning algorithm screening method and device, equipment and a storage medium, and belongs to the field of GNSS positioning, and the method comprises the steps: reading positioning results corresponding to two to-be-compared positioning algorithms, and calculating the basic performance characteristics corresponding to the positioning algorithms according to the positioning results; calculating a difference value between the two basic performance characteristics corresponding to the two positioning algorithms to obtain a performance difference characteristic; inputting all the basic performance features and the performance difference features into a pre-trained deep learning model, and determining an optimal positioning algorithm; wherein the deep learning model is obtained by performing supervised learning training on a positioning result comparison sample under various application scenes based on expert labeling by using basic performance characteristics and performance difference characteristics extracted from the positioning result comparison sample. According to the invention, the problem of low screening accuracy of the GNSS positioning algorithm in different scenes can be solved.

Description

Technical Field

[0001] This application relates to the field of GNSS positioning, and in particular to a GNSS positioning algorithm selection method, apparatus, device and storage medium. Background Technology

[0002] High-precision Global Navigation Satellite System (GNSS) positioning technology has been widely used in many fields such as surveying and mapping, autonomous driving, and smart agriculture. In practical applications, it is often necessary to compare the performance of different positioning algorithms, hardware platforms, or data processing strategies, such as comparing traditional RTK (Real-Time Kinematic) with PPP-RTK (Precise Point Positioning – Real-Time Kinematic), or comparing the performance of receivers from different manufacturers.

[0003] Existing methods heavily rely on skilled technicians manually calculating and comparing performance indicators. This requires interpreting numerous positioning result files, manually calculating various indicators, and making subjective judgments based on experience. However, complex nonlinear relationships exist among these performance indicators. If only subjective, experience-based judgments are used, the introduction of human subjectivity can easily lead to different technicians reaching different conclusions at different times when selecting GNSS positioning algorithms for various complex scenarios. This results in strong subjectivity and poor consistency in the selection process. Using fixed rules for judgment, however, cannot meet the accuracy requirements for selecting GNSS positioning algorithms across different scenarios.

[0004] Therefore, improving the accuracy of GNSS positioning algorithm selection in different scenarios has become a technical problem that needs to be solved. Summary of the Invention

[0005] This application provides a GNSS positioning algorithm selection method, apparatus, device, and storage medium, which can solve the accuracy problem of GNSS positioning algorithm selection in different scenarios in the prior art.

[0006] This application provides a GNSS positioning algorithm filtering method in some embodiments, including: Read the positioning results of the two positioning algorithms to be compared, and calculate the basic performance characteristics of each positioning algorithm based on the positioning results; Calculate the difference between the two basic performance characteristics corresponding to the two positioning algorithms to obtain the performance difference characteristics; All the aforementioned basic performance features and performance difference features are input into a pre-trained deep learning model to determine the optimal localization algorithm; wherein, the deep learning model is obtained by supervised learning training using basic performance features and performance difference features extracted from the localization result comparison samples in various application scenarios based on expert annotations.

[0007] Compared with existing technologies, the above embodiments have the following beneficial effects: By reading the positioning results of two positioning algorithms and calculating their basic performance characteristics, then calculating the differences between each basic performance characteristic and splicing the two types of features to form a fusion feature, it is possible to simultaneously retain the absolute performance of the original performance indicators and the relative change trends between the two algorithms. This allows the subsequent deep learning model to fully understand the nonlinear and complex relationships between different indicators, thereby ensuring the accuracy of positioning algorithm selection. Furthermore, since the deep learning model is trained by comparing positioning results in various application scenarios based on expert annotations, it can provide objective and consistent judgments on the merits of different algorithms when facing different application scenarios, different hardware devices, or different positioning environments. Compared with traditional methods that rely on human experience or fixed thresholds, this application effectively solves the problems of strong subjective judgment, poor consistency of conclusions, and the difficulty of adapting rule-based methods to different scenarios, thereby improving the accuracy of GNSS positioning algorithm selection in different scenarios.

[0008] Further, the deep learning model includes: a first fully connected layer, a second fully connected layer, a third fully connected layer, and an output layer; the step of inputting all the basic performance features and the performance difference features into the pre-trained deep learning model to determine the optimal localization algorithm includes: The first feature is obtained by extracting a linear combination feature between features from all the basic performance features and the performance difference features through the first fully connected layer; The second feature is obtained by extracting all the basic performance features and the nonlinear relationship features between the performance difference features from the first feature through the second fully connected layer; The second feature is compressed using the third fully connected layer to obtain the third feature; The output layer reduces the dimensionality of the third feature to the same dimension as the number of preset comparison results in order to obtain the probability value corresponding to each preset comparison result. The optimal positioning algorithm among the two positioning algorithms is determined based on the preset comparison results corresponding to the maximum probability value.

[0009] Compared with existing technologies, the above embodiments have the following beneficial effects: By using the linear combination relationship between the features of the first fully connected layer steps, such as the feature combination between horizontal accuracy and flypoint rate, the original indicators and difference information are unified and integrated; by using the second fully connected layer for nonlinear mapping, higher-level abstract features are further extracted, and the complex nonlinear relationship of the steps is improved, such as the performance improvement relationship where the difference in the fixed rate is positive and the difference in the convergence time is negative; by using the third fully connected layer to scale the above high-dimensional relationship, the features on which the classification depends are more focused and more robust. Finally, the compressed features are mapped to the probability values ​​corresponding to each preset comparison result through the output layer, so that the judgment result has interpretability and probabilistic credibility, thereby solving the problem that traditional manual comparison cannot handle the complex GNSS algorithm performance indicator relationship in complex GNSS positioning scenarios.

[0010] Furthermore, after obtaining the first feature or the second feature, the first feature or the second feature is processed sequentially through a batch normalization layer, an activation function layer, and a regularization layer.

[0011] Compared to existing technologies, the above embodiments have the following advantages: the batch normalization layer can standardize features, suppress internal covariate bias, improve training speed, and reduce sensitivity to weight initialization; the activation function layer introduces nonlinearity, enabling the model to characterize the complex multidimensional nonlinear coupling relationship between positioning accuracy features and solution performance features; the regularization layer effectively suppresses overfitting by randomly shielding some neurons, allowing the model to maintain stable prediction performance in different scenarios. By combining these three elements, the selection results of the GNSS positioning algorithm are not affected by a single scene, noise, or sample bias, making it more universal and reliable.

[0012] Further, the output layer includes: a fourth fully connected layer and a Softmax layer; the step of reducing the dimensionality of the third feature to the same dimension as the number of preset comparison results through the output layer to obtain the probability value corresponding to each preset comparison result includes: The third feature is mapped to the same dimension as the number of preset comparison results through the fourth fully connected layer to obtain the feature value corresponding to each preset comparison result; The Softmax layer maps the feature value corresponding to each preset comparison result to the corresponding probability value.

[0013] Compared to existing technologies, the above embodiments have the following beneficial effects: The fourth fully connected layer maps the compressed high-density discriminative features to a dimension consistent with the number of preset comparison results, enabling each discriminative category to possess independent response features; the Softmax layer further normalizes these response values ​​into a probability distribution, making each category comparable and mutually exclusive. Through probabilistic output, the model can not only provide the judgment result of the optimal localization algorithm, but also provide the confidence level of the judgment, which helps in credibility management in high-risk or low-quality data scenarios, thereby improving the transparency and interpretability of the algorithm's screening results.

[0014] Further, the positioning result includes: the solution state and positioning coordinates; the basic performance characteristics include: positioning accuracy characteristics and solution performance characteristics; the calculation of the basic performance characteristics corresponding to each positioning algorithm based on the positioning result includes: The positioning coordinates are converted from a spatial rectangular coordinate system or a geodetic coordinate system to a northeast-sky coordinate system, and the positioning accuracy characteristics are calculated based on the converted positioning coordinates. Based on the state of the solution, the solution performance characteristics are calculated.

[0015] Compared with the prior art, the above embodiments have the following beneficial effects: by using the solution state and positioning coordinates in the positioning results as basic inputs, and uniformly converting the positioning coordinates to the northeast-northeast coordinate system, the difference in error definition caused by inconsistent coordinate formats can be eliminated, and the horizontal accuracy, vertical accuracy, vertical peak value and other indicators can be kept comparable between different devices and different output formats; the solution state is used to directly calculate the fixation rate and convergence time, so that the solution performance can be objectively and quantitatively evaluated.

[0016] Furthermore, the positioning accuracy features include: horizontal accuracy, vertical accuracy, vertical peak value, and fly-spot rate.

[0017] Compared to existing technologies, the above embodiments have the following advantages: horizontal and vertical accuracy respectively characterize the average error performance in the planar and height directions; the vertical peak value can identify short-term sudden changes or extreme deviations, avoiding anomalies that the average value cannot capture; the flying point rate reflects the proportion of abnormal outputs of the algorithm in complex environments and is highly correlated with actual positioning stability. Therefore, by unifying horizontal accuracy, vertical accuracy, vertical peak value, and flying point rate as positioning accuracy features, the model can comprehensively evaluate the spatial error performance and stability of the positioning algorithm from multiple complementary dimensions.

[0018] Furthermore, the solution performance characteristics include: convergence time and fixation rate.

[0019] Compared to existing technologies, the above embodiments have the following advantages: convergence time reflects the time efficiency of the algorithm from initialization to obtaining a stable solution, which is extremely critical in dynamic scenarios or applications with frequent startups; the fixation rate reflects the proportion of fixed solutions throughout the entire positioning cycle, measuring the overall stability and reliability of the algorithm. By combining these two features, the model can not only determine whether an algorithm has high accuracy, but also whether it has rapid recovery capabilities and long-term stability, making the final selection results more in line with actual usage requirements.

[0020] Another embodiment of this application also provides a GNSS positioning algorithm screening device, including: a basic performance characteristic calculation module, a performance difference characteristic calculation module, and a positioning result comparison module; The basic performance feature calculation module is used to read the positioning results of the two positioning algorithms to be compared, and calculate the basic performance features corresponding to each positioning algorithm based on the positioning results. The performance difference feature calculation module is used to calculate the difference between the two basic performance features corresponding to the two positioning algorithms to obtain the performance difference feature. The location result comparison module is used to input all the basic performance features and the performance difference features into a pre-trained deep learning model to determine the optimal location algorithm; wherein, the deep learning model is obtained by supervised learning training using the basic performance features and performance difference features extracted from the location result comparison samples in various application scenarios based on expert annotation.

[0021] Another embodiment of this application also provides a terminal device, including: a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor. When the processor executes the computer program, it implements the steps of the GNSS positioning algorithm filtering method of this application.

[0022] Another embodiment of this application also provides a computer-readable storage medium item, including: a stored computer program, which, when the computer program is running, controls the device where the computer-readable storage medium is located to perform the steps of the GNSS positioning algorithm screening method of this application. Attached Figure Description

[0023] To more clearly illustrate the technical solution of this application, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0024] Figure 1This is a flowchart illustrating a GNSS positioning algorithm filtering method provided in some embodiments of this application; Figure 2 This is a schematic diagram of the structure of a GNSS positioning algorithm filtering device provided in some embodiments of this application. Detailed Implementation

[0025] To make the objectives, technical solutions, and advantages of this application clearer, the technical solutions of this application will be clearly and completely described below with reference to the accompanying drawings of the embodiments. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0026] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application pertains; the terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the application; the terms “comprising” and “having”, and any variations thereof, in the specification, claims, and foregoing description of the drawings are intended to cover non-exclusive inclusion.

[0027] In the description of the embodiments of this application, technical terms such as "first" and "second" are used only to distinguish different objects and should not be construed as indicating or implying relative importance or implicitly specifying the number, specific order, or primary and secondary relationship of the indicated technical features. In the description of the embodiments of this application, "multiple" means two or more, unless otherwise explicitly defined.

[0028] In this document, the term "embodiment" means that a particular feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of this application. The appearance of this phrase in various places throughout the specification does not necessarily refer to the same embodiment, nor is it a separate or alternative embodiment mutually exclusive with other embodiments. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described herein can be combined with other embodiments.

[0029] In the description of the embodiments in this application, the term "and / or" is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, and B existing alone. Additionally, the character " / " in this document generally indicates that the preceding and following related objects have an "or" relationship.

[0030] In the description of the embodiments of this application, the term "multiple" refers to two or more (including two), similarly, "multiple sets" refers to two or more (including two sets), and "multiple pieces" refers to two or more (including two pieces).

[0031] In the description of the embodiments of this application, unless otherwise expressly specified and limited, technical terms such as "installation," "connection," "joining," and "fixing" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral part; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; they can refer to the internal communication of two components or the interaction between two components. For those skilled in the art, the specific meaning of the above terms in the embodiments of this application can be understood according to the specific circumstances.

[0032] Existing methods heavily rely on skilled technicians manually calculating and comparing performance indicators. This requires interpreting numerous positioning result files, manually calculating various indicators, and making subjective judgments based on experience. However, complex nonlinear relationships exist among these performance indicators. If only subjective, experience-based judgments are used, the introduction of human subjectivity can easily lead to different technicians reaching different conclusions at different times when selecting GNSS positioning algorithms for various complex scenarios. This results in strong subjectivity and poor consistency in the selection process. Using fixed rules for judgment, however, cannot meet the accuracy requirements for selecting GNSS positioning algorithms across different scenarios.

[0033] Please refer to Figure 1 To address the issue of low accuracy in GNSS positioning algorithm selection under different scenarios in existing technologies, this application provides a GNSS positioning algorithm selection method, comprising steps S101 to S103, specifically: S101: Read the positioning results of the two positioning algorithms to be compared, and calculate the basic performance characteristics of each positioning algorithm based on the positioning results.

[0034] Furthermore, in some embodiments of this application, the positioning result includes: the solution state and positioning coordinates; the basic performance characteristics include: positioning accuracy characteristics and solution performance characteristics; the calculation of the basic performance characteristics corresponding to each positioning algorithm based on the positioning result includes: The positioning coordinates are converted from a spatial rectangular coordinate system or a geodetic coordinate system to a northeast-sky coordinate system, and the positioning accuracy characteristics are calculated based on the converted positioning coordinates. Based on the state of the solution, the solution performance characteristics are calculated.

[0035] Specifically, in some embodiments of this application, the positioning result includes: the positioning coordinates of each epoch and the state of the solution; wherein, the positioning coordinates include: spatial rectangular coordinates or geodetic coordinates; the state of the solution includes: fixed solution, floating-point solution and single-point solution, etc., and the time taken from starting positioning to obtaining a fixed solution that reaches a specific accuracy threshold for the first time (i.e., convergence time).

[0036] By using the solution state and positioning coordinates in the positioning results as basic inputs, and uniformly converting the positioning coordinates to the Northeast-Eastern-Sky coordinate system, the differences in error definitions caused by inconsistent coordinate formats can be eliminated, making indicators such as horizontal accuracy, vertical accuracy, and vertical peak value comparable across different devices and output formats. The solution state is used to directly calculate the fixation rate and convergence time, enabling an objective quantitative evaluation of the solution performance.

[0037] Furthermore, in some embodiments of this application, the positioning accuracy features include: horizontal accuracy, vertical accuracy, vertical peak value, and fly-spot rate.

[0038] Specifically, in some embodiments of this application, the method for calculating the horizontal accuracy is as follows: It should be noted that, in the northeast-northeast coordinate system, the values ​​for the east and north directions represent the errors in the east and north directions, and the errors in the east and north directions are further combined into the horizontal error using the following formula: in, This represents the error in the horizontal direction; This represents the error in the eastward direction; This represents the error in the north direction.

[0039] Assuming there are N epochs, the horizontal precision is calculated using the following formula: in, For horizontal accuracy; This represents the horizontal error in the i-th epoch.

[0040] Specifically, in some embodiments of this application, the method for calculating the vertical accuracy is as follows: It should be noted that in the northeast-northeast coordinate system, the value in the zenith direction represents the error in the vertical direction. Assuming there are N epochs, the formula for calculating the vertical accuracy is: in, For vertical accuracy; This represents the error in the vertical direction of the i-th epoch.

[0041] Specifically, in some embodiments of this application, the method for calculating the vertical peak value is as follows: the maximum absolute value of the zenith direction of all epochs in the northeast-northeast coordinate system is taken as the vertical peak value.

[0042] Specifically, in some embodiments of this application, the method for calculating the flying point rate is as follows: determine whether the horizontal error and the vertical error exceed a preset threshold, determine any epoch with an error exceeding the threshold as a flying point, and further calculate the flying point rate according to the following formula: flying point rate = number of flying point epochs / total number of epochs * 100%.

[0043] Horizontal and vertical accuracy characterize the average error performance in the planar and vertical directions, respectively; the vertical peak value can identify short-term abrupt changes or extreme deviations, avoiding anomalies that the average value cannot capture; the fly-spot rate reflects the proportion of abnormal outputs of the algorithm in complex environments and is highly correlated with actual positioning stability. Therefore, by unifying horizontal accuracy, vertical accuracy, vertical peak value, and fly-spot rate as positioning accuracy features, the model can comprehensively evaluate the spatial error performance and stability of the positioning algorithm from multiple complementary dimensions.

[0044] Furthermore, in some embodiments of this application, the computational performance characteristics include: convergence time and fixation rate.

[0045] Specifically, in some embodiments of this application, the convergence time is the time elapsed from the start of positioning to the first acquisition of a fixed solution that reaches a specific accuracy threshold, such as the time elapsed when the horizontal accuracy is <0.1m. For cases where a fixed solution has been reached but subsequent re-convergence occurs due to loss of lock, the average of multiple convergence times can be recorded.

[0046] Specifically, in some embodiments of this application, the fixation rate is calculated using the following formula: Fixation rate = Number of fixed solution epochs / Total number of epochs * 100%.

[0047] Convergence time reflects the time efficiency of an algorithm from initialization to obtaining a stable solution, which is crucial in dynamic scenarios or applications with frequent startups. The fixation rate reflects the proportion of fixed solutions throughout the entire positioning cycle, providing an overall measure of the algorithm's stability and reliability. By combining these two features, the model can not only determine whether an algorithm has high accuracy but also whether it possesses rapid recovery capabilities and long-term stability, making the final selection results more aligned with practical application needs.

[0048] S102: Calculate the difference between the two basic performance characteristics corresponding to the two positioning algorithms to obtain the performance difference characteristics.

[0049] Step S101 yields the basic performance characteristics of each positioning algorithm. Taking the horizontal accuracy, vertical accuracy, vertical peak value, fly-spot rate, convergence time, and fixation rate mentioned in the above embodiments as examples, two 6-dimensional basic performance characteristics can be obtained. Subtracting the two 6-dimensional basic performance characteristics yields the following 6-dimensional performance difference characteristics, including: the difference in fixation rate, the difference in horizontal accuracy (it should be noted that a decrease in the difference in accuracy values ​​indicates an improvement in the performance of the subtracted algorithm; a negative difference often means that the subtracted algorithm performs better), the difference in vertical accuracy, the difference in vertical peak value, the difference in fly-spot rate, and the difference in convergence time.

[0050] S103: Input all the basic performance features and the performance difference features into the pre-trained deep learning model to determine the optimal localization algorithm; wherein, the deep learning model is obtained by supervised learning training using basic performance features and performance difference features extracted from the localization result comparison samples in various application scenarios based on expert annotations.

[0051] Furthermore, in some embodiments of this application, the deep learning model includes: a first fully connected layer, a second fully connected layer, a third fully connected layer, and an output layer; the step of inputting all the basic performance features and the performance difference features into the pre-trained deep learning model to determine the optimal localization algorithm includes: The first feature is obtained by extracting a linear combination feature between features from all the basic performance features and the performance difference features through the first fully connected layer; The second feature is obtained by extracting all the basic performance features and the nonlinear relationship features between the performance difference features from the first feature through the second fully connected layer; The second feature is compressed using the third fully connected layer to obtain the third feature; The output layer reduces the dimensionality of the third feature to the same dimension as the number of preset comparison results in order to obtain the probability value corresponding to each preset comparison result. The optimal positioning algorithm among the two positioning algorithms is determined based on the preset comparison results corresponding to the maximum probability value.

[0052] Furthermore, in some embodiments of this application, after obtaining the first feature or the second feature, the first feature or the second feature is processed sequentially through a batch normalization layer, an activation function layer, and a regularization layer.

[0053] Specifically, in some embodiments of this application, the deep neural network is a multilayer perceptron, where each layer undertakes specific feature transformation and abstraction functions. Taking the two 6-dimensional basic performance features and one 6-dimensional performance difference feature as examples, the design logic is as follows: The input layer receives an 18-dimensional feature vector (a concatenated feature vector of the 12-dimensional basic performance feature and the 6-dimensional performance difference feature) as the data input interface. The first fully connected layer learns the linear combination relationship between features, increasing the 18-dimensional vector to 128 dimensions. That is, feature fusion is achieved through the weight matrix W1 (18×128) and the bias vector b1, thereby capturing feature combination patterns such as horizontal precision difference and flying point rate difference to obtain the first feature. Furthermore, the first feature is standardized by a batch normalization layer to improve training stability and reduce parameter initialization sensitivity. Then, the ReLU activation function introduces non-linear transformation capability; its functional form is f(x) = max(0,x), enhancing model expressiveness and generating feature sparsity. Finally, the Dropout layer randomly masks some neurons during training to prevent overfitting and improve model robustness; in this embodiment, the dropout rate is set to 0.3. The second fully connected layer extracts high-level abstract features from the first feature, compressing the 128-dimensional first feature into a 64-dimensional second feature, thereby capturing complex non-linear relationships, such as performance non-linear relationships where the difference in recognition rate is positive and the difference in convergence time is negative. Subsequently, it passes through batch normalization, ReLU, and Dropout layers to further complete feature transformation and regularization. The third fully connected layer maps the 64-dimensional feature to a 32-dimensional feature, achieving final feature refinement and preparing for classification decisions. The output layer maps the 32-dimensional feature to a dimensional space with the same number of preset comparison results, obtaining the probability distribution of each preset comparison result.

[0054] Specifically, in some embodiments of this application, the preset comparison results include: better, worse, and similar.

[0055] By leveraging the linear combination relationships between features in the first fully connected layer, such as the feature combination between horizontal accuracy and flypoint rate, the original indicators and difference information are unified and integrated. The second fully connected layer performs nonlinear mapping to further extract high-level abstract features and complex nonlinear relationships between steps, such as the performance improvement relationship where the difference in fixed rate is positive and the difference in convergence time is negative. The third fully connected layer performs scale compression on the aforementioned high-dimensional relationships, making the features relied upon for classification more focused and robust. Finally, the output layer maps the compressed features to the probability values ​​corresponding to each preset comparison result, making the judgment results interpretable and probabilistically reliable. This solves the problem that traditional manual comparison cannot handle complex GNSS algorithm performance indicator relationships in complex GNSS positioning scenarios.

[0056] Further, in some embodiments of this application, the output layer includes: a fourth fully connected layer and a Softmax layer; the step of reducing the dimensionality of the third feature to the same dimension as the number of preset comparison results through the output layer to obtain the probability value corresponding to each preset comparison result includes: The third feature is mapped to the same dimension as the number of preset comparison results through the fourth fully connected layer to obtain the feature value corresponding to each preset comparison result; The Softmax layer maps the feature value corresponding to each preset comparison result to the corresponding probability value.

[0057] Specifically, in some embodiments of this application, the formula for calculating the probability value using the Softmax function is as follows: in, The feature value representing the i-th preset comparison result; This represents the probability of belonging to the i-th preset comparison result; The preset number of comparison results.

[0058] The fourth fully connected layer maps the compressed high-density discriminative features to a dimension consistent with the preset number of comparison results, giving each discriminative category independent response features. The Softmax layer further normalizes these response values ​​into a probability distribution, making the categories comparable and mutually exclusive. Through probabilistic output, the model can not only provide the judgment result of the optimal localization algorithm, but also provide the confidence level of the judgment, which helps in credibility management in high-risk or low-quality data scenarios, thereby improving the transparency and interpretability of the algorithm's screening results.

[0059] Specifically, in some embodiments of this application, the training process of the deep neural network includes: Step 1: The process begins with collecting a large amount of raw GNSS observation data, covering various scenarios, equipment, and environmental conditions. Each expert independently evaluates the raw positioning results of each pair of comparison algorithms, such as Method A and Method B, and provides a labeling conclusion based on a weighted assessment of multiple indicators, stating that "A is better than B," "A is worse than B," or "A and B are similar." All experts' labeling results are integrated through a multi-expert back-to-back calibration process. When experts disagree, a team of senior experts arbitrates the dispute to ensure the accuracy and consistency of the labeling results.

[0060] Step 2: Transform the expert-annotated raw data into feature representations usable by the model (i.e., basic performance features and performance difference features). Model training and optimization employ a supervised learning paradigm, using classification cross-entropy as the loss function. The training data is divided into training, validation, and test sets in a 7:2:1 ratio. Optimization uses the Adam algorithm with an initial learning rate of 0.001 and an exponential decay strategy (decreasing by 0.5 every 20 epochs). Batch processing (batch size 32) is used during training, and performance is evaluated on the validation set after each epoch to monitor accuracy and loss curves.

[0061] Step 3: Training Process Management. Implement an early stopping strategy. Terminate training when the validation set loss no longer decreases for 10 consecutive epochs, and restore the model parameters that best perform on the validation set. After training, comprehensively evaluate model performance on an independent test set, calculating metrics such as accuracy, precision, recall, and F1 score, and analyze classification performance for each category. When saving the final model, record the standardized parameters (mean and variance) to ensure the same preprocessing method is used in subsequent predictions.

[0062] Preferably, in some embodiments of this application, after training the deep neural network, the method further includes: converting the probability values ​​output by the model into a final comparison conclusion, selecting the preset comparison result with the highest probability value as the output result, and providing a prediction confidence level. The confidence level is calculated based on the probability distribution of the Softmax output, taking the maximum probability value as the prediction confidence level. This confidence level reflects the model's certainty about the prediction result, providing users with a quantitative reference for the reliability of their decisions. For example, when the output result is [0.92, 0.05, 0.03], the prediction result is "A is better than B", with a confidence level of 92%, indicating that the model has a high degree of confidence in this prediction. The system can generate a structured report, which details the comparison of various feature indicators and their contribution to the final conclusion, providing users with transparent and interpretable decision support.

[0063] In summary, the GNSS positioning algorithm selection method provided in this application has the following advantages compared to existing technologies: By reading the positioning results of two positioning algorithms and calculating their basic performance characteristics, then calculating the differences between each basic performance characteristic and concatenating the two types of features to form a fusion feature, it can simultaneously retain the absolute performance of the original performance indicators and the relative change trends between the two algorithms. This allows the subsequent deep learning model to fully understand the nonlinear and complex relationships between different indicators, thereby ensuring the accuracy of positioning algorithm selection. Furthermore, since the deep learning model is trained by comparing positioning results in various application scenarios based on expert annotations, it can provide objective and consistent judgments on the merits of different algorithms in different application scenarios, hardware devices, or positioning environments. Compared to traditional methods that rely on human experience or fixed thresholds, this application effectively solves the problems of strong subjectivity in judgment, poor consistency of conclusions, and the difficulty of adapting rule-based methods to different scenarios, thereby improving the accuracy of GNSS positioning algorithm selection in different scenarios.

[0064] like Figure 2 As shown, based on the above-described method embodiments, this application provides a GNSS positioning algorithm screening device, including: a basic performance feature calculation module 201, a performance difference feature calculation module 202, and a positioning result comparison module 203; wherein, the basic performance feature calculation module 201 is used to read the positioning results corresponding to the two positioning algorithms to be compared, and calculate the basic performance features corresponding to each positioning algorithm based on the positioning results; the performance difference feature calculation module 202 is used to calculate the difference between the two basic performance features corresponding to the two positioning algorithms to obtain the performance difference features; the positioning result comparison module 203 is used to input all the basic performance features and the performance difference features into a pre-trained deep learning model to determine the optimal positioning algorithm; wherein, the deep learning model is obtained by supervised learning training using basic performance features and performance difference features extracted from the positioning result comparison samples based on expert annotations in various application scenarios.

[0065] Further, in some embodiments of this application, the deep learning model includes: a first fully connected layer, a second fully connected layer, a third fully connected layer, and an output layer; the localization result comparison module 203 includes: a first feature extraction unit, a second feature extraction unit, a third feature extraction unit, a probability output unit, and an optimal determination unit; the localization result comparison module 203 is used to input all the basic performance features and the performance difference features into the pre-trained deep learning model to determine the optimal localization algorithm, including: the first feature extraction unit, used to extract linear groups between features from all the basic performance features and the performance difference features through the first fully connected layer. The system combines features to obtain a first feature; a second feature extraction unit is used to extract all the basic performance features and the nonlinear relationship features between the performance difference features from the first feature through the second fully connected layer to obtain a second feature; a third feature extraction unit is used to perform feature scale compression on the second feature through the third fully connected layer to obtain a third feature; a probability output unit is used to reduce the dimensionality of the third feature to the same dimension as the number of preset comparison results through the output layer to obtain the probability value corresponding to each preset comparison result; and an optimal determination unit is used to determine the optimal positioning algorithm among the two positioning algorithms based on the preset comparison result corresponding to the maximum probability value.

[0066] Furthermore, in some embodiments of this application, after obtaining the first feature or the second feature, the first feature or the second feature is processed sequentially through a batch normalization layer, an activation function layer, and a regularization layer.

[0067] Furthermore, in some embodiments of this application, the output layer includes: a fourth fully connected layer and a Softmax layer; the probability output unit is used to reduce the dimensionality of the third feature to the same dimension as the number of preset comparison results through the output layer to obtain the probability value corresponding to each preset comparison result, including: mapping the third feature to the same dimension as the number of preset comparison results through the fourth fully connected layer to obtain the feature value corresponding to each preset comparison result; and mapping the feature value corresponding to each preset comparison result to the corresponding probability value through the Softmax layer.

[0068] Further, in some embodiments of this application, the positioning result includes: the state of the solution and the positioning coordinates; the basic performance characteristics include: positioning accuracy characteristics and solution performance characteristics; the basic performance characteristic calculation module 201 includes: a positioning accuracy characteristic calculation unit and a solution performance characteristic calculation unit; the basic performance characteristic calculation module 201 is used to calculate the basic performance characteristics corresponding to each of the positioning algorithms based on the positioning result, including: the positioning accuracy characteristic calculation unit is used to convert the positioning coordinates from a spatial rectangular coordinate system or a geodetic coordinate system to a northeast-sky coordinate system, and calculate the positioning accuracy characteristics based on the converted positioning coordinates; the solution performance characteristic calculation unit is used to calculate the solution performance characteristics based on the state of the solution.

[0069] Furthermore, in some embodiments of this application, the positioning accuracy features include: horizontal accuracy, vertical accuracy, vertical peak value, and fly-spot rate.

[0070] Furthermore, in some embodiments of this application, the computational performance characteristics include: convergence time and fixation rate.

[0071] It is understood that the above-described device embodiments correspond to the method embodiments of this application, and can implement the GNSS positioning algorithm filtering method provided by any of the above-described method embodiments of this application.

[0072] In summary, the GNSS positioning algorithm screening device provided in this application has the following advantages compared to the prior art: By reading the positioning results of two positioning algorithms and calculating their basic performance characteristics, then calculating the difference between each basic performance characteristic and splicing the two types of features to form a fusion feature, it can simultaneously retain the absolute performance of the original performance indicators and the relative change trend between the two algorithms. This allows the subsequent deep learning model to fully understand the nonlinear and complex relationship between different indicators, thereby ensuring the accuracy of positioning algorithm screening. Furthermore, since the deep learning model is trained by comparing positioning results in various application scenarios based on expert annotations, it can provide objective and consistent judgments on the merits of different algorithms when facing different application scenarios, different hardware devices, or different positioning environments. Compared to traditional methods that rely on human experience or fixed thresholds, this application effectively solves the problems of strong subjectivity in judgment, poor consistency of conclusions, and the difficulty of adapting rule-based methods to different scenarios, thereby improving the accuracy of GNSS positioning algorithm screening in different scenarios.

[0073] It should be noted that the device embodiments described above are merely illustrative, and some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Furthermore, in the accompanying drawings of the device embodiments provided in this application, the connection relationships between modules indicate that they have communication connections, which can specifically be implemented as one or more communication buses or signal lines. Those skilled in the art can understand and implement this without any creative effort.

[0074] Based on the above embodiments of the GNSS positioning algorithm filtering method, another embodiment of this application provides a terminal device, which includes a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor. When the processor executes the computer program, it implements the GNSS positioning algorithm filtering method of any embodiment of this application.

[0075] For example, in this embodiment, the computer program can be divided into one or more modules, which are stored in the memory and executed by the processor to complete this application. The one or more module units may be a series of computer program instruction segments capable of performing a specific function, which describe the execution process of the computer program in the terminal device.

[0076] The terminal device may be a desktop computer, laptop, handheld computer, or cloud server, etc. The terminal device may include, but is not limited to, a processor and a memory.

[0077] The processor can be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor can be a microprocessor or any conventional processor. The processor is the control center of the terminal device, connecting all parts of the terminal device via various interfaces and lines.

[0078] Based on the above-described method embodiments, another embodiment of this application provides a computer-readable storage medium including a stored computer program, wherein, when the computer program is executed, it controls the device where the computer-readable storage medium is located to execute the GNSS positioning algorithm screening method described in any of the above-described method embodiments of this application.

[0079] The modules / units integrated in the device / terminal equipment, if implemented as software functional units and sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, all or part of the processes in the methods of the above embodiments can also be implemented by a computer program instructing related hardware. The computer program can be stored in a computer-readable storage medium, and when executed by a processor, it can implement the steps of the various method embodiments described above. The computer program includes computer program code, which can be in the form of source code, object code, executable files, or certain intermediate forms. The computer-readable medium can include: any entity or device capable of carrying the computer program code, recording media, USB flash drives, portable hard drives, magnetic disks, optical disks, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signals, telecommunication signals, and software distribution media, etc.

Claims

1. A GNSS positioning algorithm screening method, characterized in that, The method comprises the following steps: reading positioning results corresponding to two positioning algorithms to be compared, and calculating basic performance characteristics corresponding to each of the two positioning algorithms according to the positioning results; calculating the difference between the two basic performance characteristics corresponding to the two positioning algorithms to obtain performance difference characteristics; inputting all the basic performance characteristics and the performance difference characteristics into a pre-trained deep learning model to determine the optimal positioning algorithm; wherein the deep learning model is obtained by supervised learning and training of basic performance characteristics and performance difference characteristics extracted from positioning result comparison samples in various application scenarios based on expert annotation.

2. The GNSS positioning algorithm screening method of claim 1, wherein, The deep learning model comprises a first fully connected layer, a second fully connected layer, a third fully connected layer, and an output layer. The method of inputting all the basic performance characteristics and the performance difference characteristics into the pre-trained deep learning model to determine the optimal positioning algorithm comprises: extracting linear combination characteristics between the features from all the basic performance characteristics and the performance difference characteristics through the first fully connected layer to obtain first features; extracting non-linear relationship characteristics between all the basic performance characteristics and the performance difference characteristics from the first features through the second fully connected layer to obtain second features; performing feature scale compression on the second features through the third fully connected layer to obtain third features; dimensionally reducing the third features to the same dimension as the number of preset comparison results through the output layer to obtain probability values corresponding to each preset comparison result; determining the optimal positioning algorithm from the two positioning algorithms according to the preset comparison result corresponding to the maximum probability value.

3. The GNSS positioning algorithm screening method of claim 2, wherein, After obtaining the first features or the second features, the first features or the second features are sequentially processed through a batch normalization layer, an activation function layer, and a regularization layer.

4. The GNSS positioning algorithm screening method of claim 2, wherein, The output layer comprises a fourth fully connected layer and a Softmax layer. The method of dimensionally reducing the third features to the same dimension as the number of preset comparison results through the output layer to obtain probability values corresponding to each preset comparison result comprises: mapping the third features to the same dimension as the number of preset comparison results through the fourth fully connected layer to obtain feature values corresponding to each preset comparison result; mapping the feature values corresponding to each preset comparison result to the corresponding probability values through the Softmax layer.

5. The GNSS positioning algorithm screening method of claim 1, wherein, The positioning results comprise a solution state and positioning coordinates. The basic performance characteristics comprise positioning accuracy characteristics and solution performance characteristics. The method of calculating the basic performance characteristics corresponding to each of the positioning algorithms according to the positioning results comprises: converting the positioning coordinates from a spatial rectangular coordinate system or a geodetic coordinate system to an east-north-up coordinate system, and calculating the positioning accuracy characteristics according to the converted positioning coordinates; calculating the solution performance characteristics according to the solution state.

6. The GNSS positioning algorithm screening method of claim 5, wherein, The positioning accuracy characteristics comprise horizontal accuracy, vertical accuracy, vertical direction peak value, and fly point rate.

7. The GNSS positioning algorithm screening method of claim 5, wherein, The solution performance characteristics comprise convergence time and fixation rate.

8. A GNSS positioning algorithm screening device, characterized in that ​ The basic performance feature calculation module, the performance difference feature calculation module, and the positioning result comparison module; The basic performance feature calculation module is configured to read positioning results corresponding to two positioning algorithms to be compared, and calculate basic performance features corresponding to the two positioning algorithms according to the positioning results. The performance difference feature calculation module is configured to calculate a difference between two basic performance features corresponding to the two positioning algorithms, and obtain a performance difference feature. The positioning result comparison module is configured to input all the basic performance features and the performance difference feature into a pre-trained deep learning model, and determine an optimal positioning algorithm. The deep learning model is obtained by supervised learning and training of basic performance features and performance difference features extracted from positioning result comparison samples in various application scenarios based on expert annotation.

9. A terminal device, comprising: The computer readable storage medium comprises a stored computer program, wherein the computer program, when executed, controls a device in which the computer readable storage medium is located to perform the GNSS positioning algorithm screening method according to any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer readable storage medium comprises a stored computer program, wherein the computer program, when executed, controls a device in which the computer readable storage medium is located to perform the GNSS positioning algorithm screening method according to any one of claims 1 to 7.

Images