A deep learning-based integrated navigation position variance elasticity adjustment optimization method, system, device and medium

Abstract

The application provides a GNSS / SINS combined navigation position variance elasticity adjustment optimization method, system, device and medium based on deep learning, belonging to the navigation field, comprising a multi-dimensional precision characteristic factor selection step, a convolutional neural network training step, a variance elasticity adjustment and filtering solution optimization step; the multi-dimensional precision characteristic factor selection step determines the parameter index associated with the position error through a nonlinear correlation calculation method, the convolutional neural network training step performs offline training of the position precision prediction model based on the selection result of the parameter index, and the variance elasticity adjustment and optimization step uses the offline trained model to predict the position precision online and elastically adjust the position variance, so that the true error of the position state in the filtering solution and its variance are more matched, thereby making the prior information and the current information obtain a more reasonable weight ratio, and improving the current combined navigation positioning and pose performance in a complex electromagnetic environment.

Description

Technical Field

[0001] This invention relates to the field of navigation technology, specifically to a method, system, device, and medium for elastic adjustment of position variance in GNSS / SINS integrated navigation based on deep learning and its optimization in Kalman filter solution. Background Technology

[0002] GNSS (Global Navigation Satellite System) provides all-weather, high-precision absolute spatiotemporal reference information, but its positioning performance degrades rapidly in complex electromagnetic environments due to the inherent vulnerability of satellite radio signals. SINS (Strapdown Inertial Navigation System), as a relative positioning method, can provide all-weather, high-frequency navigation information including position, velocity, and attitude, and features passive autonomy and high short-term accuracy. However, its dead reckoning error can diverge over time due to errors in inertial measurement unit components. The GNSS / SINS integrated navigation technology, which combines the advantages of both, offers high accuracy, strong reliability, wide adaptability, and good dynamics, thus becoming a crucial technology for obtaining accurate spatiotemporal references.

[0003] With the gradual development of intelligent carriers and devices and the expansion of their application fields, integrated navigation scenarios are extending from open environments to complex urban environments. The performance requirements of integrated navigation are shifting from a single accuracy indicator to high precision, high reliability, and high availability. The academic community has conducted a more in-depth analysis of the mechanisms affecting navigation performance, among which the inconsistency problem of the mismatch between the actual error and variance of navigation parameters has gradually attracted attention. Due to factors such as observation gross errors, unmodeled errors, and nonlinear non-Gaussian errors, the actual errors and estimated accuracy (post-hoc standard deviation) of various navigation parameters in the integrated navigation system are mismatched. This results in an unreasonable weighting ratio between current and historical observation information in the filtering algorithm, easily causing distortion of the statistical characteristics of the state estimate and reducing the accuracy of the integrated navigation filtering results.

[0004] Due to the inherent limitations and vulnerabilities of GNSS system design, satellite signals are easily affected by interference, jamming, and spoofing. These factors not only degrade GNSS solution performance but also easily cause distortion of positioning statistical characteristics, resulting in a general mismatch between the actual error of the GNSS position result and the estimated accuracy (post-hoc standard deviation). Integrated navigation systems use GNSS position results as observations; however, due to inaccurate internal consistency accuracy, it is difficult to determine a reasonable observation variance, severely impacting the performance of integrated navigation systems in complex environments. To address this issue, current methods such as Helmert, BIQUE, and LS-VCE are used to estimate variance components to mitigate the impact of low-quality observation information, thereby constructing a more accurate GNSS stochastic model. Furthermore, the unmodeled errors of the system have been analyzed, and data processing methods considering factors such as the temporal correlation of GNSS satellite signals, interference attenuation and multipath effects in complex environments, and residual errors in the ionosphere and troposphere have been studied. In addition, robust adaptive filtering methods for compensating for stochastic models have been proposed, and algorithms such as unscented Kalman filtering and particle filtering have been used to mitigate the impact of nonlinear non-Gaussian errors. In recent years, heuristic algorithms such as classification regression trees and bootstrap aggregation have also been used to adjust for location variance and system noise.

[0005] However, the error variation in position status has a complex relationship with numerous influencing factors. Due to the complexity of its own system composition, the vulnerability of active radio signals, and the diversity of error sources, GNSS makes it difficult to find suitable evaluation criteria for positioning accuracy. For example, the standard deviation (STD) that directly represents the accuracy within the positioning result is usually an optimistic estimate due to the inconsistency problem of Kalman filtering, making it difficult to accurately measure the accuracy of the result. Single indicators that indirectly reflect positioning accuracy, such as the position dilution of precision (PDOP) or the fixed ambiguity ratio, are not comprehensive enough to reflect the real error in actual dynamic scenarios. Therefore, existing conventional methods have always struggled to accurately model errors in complex environments and cannot accurately describe the position variance. Summary of the Invention

[0006] In GNSS / SINS integrated navigation, where GNSS position results are used as observations, the complexity of the GNSS system itself, the vulnerability of active radio signals, and the diversity of error sources make it difficult to find suitable evaluation criteria for positioning accuracy and determine reasonable observation variance. This seriously affects the accuracy and reliability of positioning and attitude determination in complex environments. This invention proposes a method, system, device, and medium for elastic adjustment and optimization of position variance based on neural networks, which can improve the positioning and attitude determination performance of integrated navigation in complex electromagnetic environments.

[0007] In a first aspect, the present invention provides a method for elastic adjustment and optimization of position variance in integrated navigation based on deep learning, comprising: Based on historical GNSS and / or SINS data, GNSS observation environment parameters and solution quality parameters are extracted. The correlation between each parameter and position error is calculated using the Spearman order correlation coefficient. Parameters with correlation coefficients greater than a set threshold are selected to form a set of multidimensional accuracy characterization factors. A training dataset is constructed based on historical data. The input is a multidimensional accuracy representation factor and the output is the position error. The position error prediction model is established through offline training using a convolutional neural network. During real-time navigation, the multidimensional accuracy representation factor of the current epoch is obtained and input into the trained position error prediction model to obtain the position error prediction value. Based on HDOP, VDOP, and PDOP, the position error prediction value is decomposed into each direction to construct the position prediction variance matrix. The position prediction variance matrix is ​​then used as the observation variance and substituted into the Kalman filter of the GNSS / SINS integrated navigation to achieve dynamic adjustment of variance and optimization of filtering performance.

[0008] In some instances, the GNSS observation environment parameters include: Position Accuracy Factor (PDOP), Carrier-to-Noise Ratio (CNR) eigenvalue, and Multipath Error; the solution quality parameters include: Intra-Position Consistency (STD), Continuous Fixed Satellite Count (NSAT), and Ambiguity Fixed Ratio value.

[0009] In some instances, the correlation between each parameter and the position error is calculated using the Spearman order correlation coefficient, and parameters with correlation coefficients greater than a set threshold are selected to form a multidimensional accuracy characterization factor set, including: Based on GNSS and / or SINS historical data, the same GNSS positioning algorithm as the real-time system is used to perform offline calculations on the historical data to obtain the historical position calculation results for each epoch. High-precision integrated navigation post-processing technology is used to obtain the corresponding position reference values. The position error for each epoch is obtained from the position reference values ​​and the position calculation results obtained by the GNSS positioning algorithm. At the same time, the values ​​of GNSS observation environment related parameters and calculation quality related parameters for each epoch are recorded. The Spearman order nonlinear correlation coefficient calculation method is used to perform correlation analysis on the obtained GNSS observation environment related parameter values ​​and solution quality related parameter values ​​with position error, and to calculate the correlation coefficient between each parameter value and position error. The parameter index with a correlation coefficient greater than a set threshold is selected as the multidimensional accuracy characterization factor.

[0010] In some instances, the convolutional neural network includes at least four convolutional layers, with ReLU as the activation function, tanh as the activation function for fully connected layers, and root mean square error as the loss function.

[0011] In some instances, the construction of a training dataset based on historical data, with multidimensional accuracy representation factors as input and position error as output, involves offline training using a convolutional neural network to establish a position error prediction model, including: Based on GNSS and / or SINS historical data, the same GNSS positioning algorithm as the real-time system is used to perform offline calculations on the historical data to obtain the historical position calculation results for each epoch. High-precision integrated navigation post-processing technology is used to obtain the corresponding position reference value. By comparing the two, the position error of each epoch is calculated and used as the real label for subsequent model training. The training dataset is constructed by using the multidimensional accuracy representation factor and the corresponding position error for each epoch. Based on the inertial navigation hardware information of the collected data, the training dataset is grouped according to different hardware levels and split into multiple sub-datasets according to hardware level. Each sub-dataset corresponds to an inertial navigation hardware with a specific performance level. For each subset of data, a convolutional neural network is trained independently. The mapping relationship between multidimensional accuracy representation factors and position error is established through the convolutional neural network, resulting in a position error prediction model that matches the hardware level.

[0012] In some instances, the method for constructing the location prediction variance matrix is ​​as follows: the predicted location error values ​​are arranged according to... The prediction error is decomposed into components in each direction to obtain position prediction error components in different directions. These components are then used to fill the main diagonal of the prediction variance matrix, which is then transformed into a position prediction variance matrix corresponding to the coordinate system in which the filtering solution is performed. This is the predicted value of the position error. , , These are the positional variance components in the east, north, and sky directions, respectively.

[0013] In some instances, the state vector of the Kalman filter includes: position error, velocity error, misalignment angle, accelerometer bias, and gyroscope bias.

[0014] In some instances, the method is applicable to various GNSS systems, including GPS, BeiDou, GLONASS, and GALILEO, as well as various SINS systems, including consumer-grade and tactical-grade systems.

[0015] Secondly, the present invention provides a deep learning-based integrated navigation position variance elastic adjustment and optimization system, comprising: The multidimensional accuracy characterization factor selection module is used to extract GNSS observation environment parameters and solution quality parameters based on GNSS and / or SINS historical data, calculate the correlation between each parameter and position error through Spearman order correlation coefficient, and select parameters with correlation coefficients greater than a set threshold to form a multidimensional accuracy characterization factor set. The convolutional neural network training module is used to build a training dataset based on historical data. The input is a multidimensional accuracy representation factor and the output is the position error. It is trained offline through a convolutional neural network to establish a position error prediction model. The variance elastic adjustment and filtering optimization module is used to obtain the multi-dimensional accuracy characterization factor of the current epoch during real-time navigation, input it into the trained position error prediction model, obtain the position error prediction value, decompose the position error prediction value into each direction based on HDOP, VDOP, and PDOP, construct the position prediction variance matrix, and substitute the prediction variance matrix as the observation variance into the Kalman filter of GNSS / SINS integrated navigation to realize dynamic adjustment of variance and optimization of filtering performance.

[0016] Thirdly, the present invention provides a computer device including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of any of the methods described above.

[0017] Fourthly, the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the steps of any of the methods described above.

[0018] In summary, compared with the prior art, the above-described technical solutions conceived by this invention can achieve the following beneficial effects: This invention is a method for elastic adjustment and optimization of position variance in GNSS / SINS integrated navigation based on deep learning, which includes three aspects: first, selection of multi-dimensional accuracy representation factors; second, training of convolutional neural networks; and third, elastic adjustment of variance and optimization of filtering solution.

[0019] The selection of multidimensional accuracy characterization factors is based on historical GNSS / SINS data. The position results are calculated and the position error is obtained through real-time GNSS positioning algorithm. A non-rank nonlinear correlation calculation method is used to analyze the correlation between the values ​​of GNSS observation environment and solution quality related parameters and the values ​​of GNSS position error. The parameters with strong correlation are selected to establish a set of multidimensional accuracy characterization factors, which can avoid the problem that the current single parameter index cannot fully characterize the position error.

[0020] Convolutional neural network training is based on a large amount of historical GNSS / SINS data. The position result is calculated and the position error is obtained through real-time GNSS positioning algorithm. The multi-dimensional accuracy representation factors in the calculation process are saved and recorded, thus forming a large amount of labeled data including the set of position error and multi-dimensional accuracy representation factors. Then, the mapping relationship between multi-dimensional accuracy representation factors and position error is established offline through convolutional neural network, so as to obtain a position accuracy prediction model based on deep learning. Relying on the powerful mining and expression capabilities of deep learning for implicit relationships, an accurate position accuracy prediction model can be obtained.

[0021] Variance elastic adjustment and filtering optimization utilize a trained model to calculate position prediction error during real-time GNSS positioning. The position prediction error is decomposed in different directions using the Horizontal Dilution of Precision (HDOP), Vertical Dilution of Precision (VDOP), and Geometric Dilution of Precision (PDOP) factors, thereby constructing a position prediction variance matrix. The original position variance is then elastically adjusted using the GNSS position prediction variance to optimize subsequent GNSS / SINS integrated navigation filtering, effectively improving the integrated navigation positioning and attitude determination performance in complex electromagnetic environments.

[0022] The advantages of this invention specifically include: (1) This invention proposes a method for elastic adjustment and optimization of position variance of GNSS / SINS integrated navigation based on deep learning. This method has a wide range of applications and can be used in any integrated navigation, positioning and attitude determination system composed of all GNSS (including single GPS or single Beidou) and all SINS (including consumer-grade and tactical-grade strapdown inertial navigation). (2) This invention proposes a method for elastic adjustment and optimization of position variance of GNSS / SINS integrated navigation based on deep learning. Through offline modeling and online use, it does not increase the computational burden when used. (3) The present invention proposes a method for selecting multidimensional precision characterization factors using the Spearman correlation coefficient calculation method. It does not require the analysis object to be continuous or linearly related, and based on the ranking, it has a high tolerance for outliers in the analysis object. (4) The present invention proposes to establish a multi-dimensional accuracy characterization factor set to replace the single parameter index for modeling position error. This method can avoid the limitations of the single parameter index and can accurately establish a model for GNSS position error with complex influencing factors. (5) This invention proposes to establish a position accuracy prediction model using a convolutional neural network, which utilizes the powerful mining and expression capabilities of deep learning for complex implicit relationships, and realizes the mapping relationship between multidimensional accuracy representation factors and position error. (6) This invention proposes a method to decompose the absolute value of position prediction error in different directions by using position accuracy factor. This method is scientific in principle and simple to implement. It can be used to obtain position prediction error components in different directions and fill the position variance matrix. Attached Figure Description

[0023] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0024] Figure 1 This is a schematic diagram of a method for elastic adjustment and optimization of position variance in GNSS / SINS integrated navigation based on deep learning, provided by an embodiment of the present invention. Figure 2 This is a flowchart illustrating the implementation of a deep learning-based GNSS / SINS integrated navigation position variance elastic adjustment and optimization method provided in this embodiment of the invention. Figure 3 This is a schematic diagram illustrating the correlation between commonly used parameter indicators and position errors provided in embodiments of the present invention; Figure 4 This is a schematic diagram of the position error prediction neural network structure provided in an embodiment of the present invention; Figure 5 This is a time series comparison chart of reference position error and predicted position error in real-time calculation provided by an embodiment of the present invention; Figure 6 This is a comparison chart of the real-time GNSS / SINS filtering solution time series before and after the application of predicted location variance provided in this embodiment of the invention; Figure 7 This is a schematic diagram of a GNSS / SINS integrated navigation position variance elastic adjustment and optimization system based on deep learning, provided by an embodiment of the present invention; Figure 8 This is a schematic diagram of the structure of a computer device provided in an embodiment of the present invention. Detailed Implementation

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

[0026] In the following description, specific embodiments of the invention will be illustrated with reference to steps and symbols performed by one or more computers, unless otherwise stated. Therefore, these steps and operations will be referred to several times as being performed by a computer, and computer execution as referred to herein includes operations by a computer processing unit representing electronic signals of data in a structured format. This operation transforms the data or maintains it at a location in the computer's memory system, which can be reconfigured or otherwise alter the operation of the computer in a manner well known to those skilled in the art. The data structure maintained by the data is the physical location of the memory, which has specific characteristics defined by the data format. However, the principles of the invention described above are not intended to be limiting, and those skilled in the art will understand that many of the following steps and operations can also be implemented in hardware.

[0027] The terms "module" or "unit" as used herein can be considered as software objects executing on the computing system. Different components, modules, engines, and services described herein can be considered as implementations on the computing system. The apparatus and methods described herein are preferably implemented in software, but can also be implemented in hardware, both of which are within the scope of this invention.

[0028] Those skilled in the art will understand that, unless specifically stated otherwise, the singular forms “a,” “an,” and “the” used herein may also include the plural forms. It should be further understood that the term “comprising” as used in this specification means the presence of features, integers, steps, operations, elements, and / or components, but does not exclude the presence or addition of one or more other features, integers, steps, operations, elements, components, and / or groups thereof. It should be understood that when we say an element is “connected” or “coupled” to another element, it can be directly connected or coupled to the other element, or there may be intermediate elements. Furthermore, “connected” or “coupled” as used herein can include wireless connections or wireless coupling. The term “and / or” as used herein includes all or any units and all combinations of one or more associated listed items.

[0029] In this embodiment of the invention, a deep learning-based GNSS / SINS integrated navigation position variance elastic adjustment optimization method is provided to solve the problem of inconsistent filtering solutions caused by the inability of existing position variance and its adjustment models to accurately reflect the true position error, thereby improving the integrated navigation positioning and attitude determination performance in current complex environments. The method includes the following steps: S1: Selection of multidimensional accuracy characterization factors: Through nonlinear correlation calculation method, the parameter indexes associated with position error are accurately determined. S2: Convolutional neural network training: Based on the selection results of multi-dimensional accuracy representation factors, a training dataset is established to carry out offline training of the position accuracy prediction model; S3: Variance elastic adjustment and filter solution optimization. The prediction model is used to calculate the position accuracy online and the position variance is elastically adjusted so that the actual error of the position state in the filter solution matches its variance, thereby optimizing the performance of the integrated navigation filter solution.

[0030] Furthermore, the specific method of step S1 is as follows: S1.1: Based on historical GNSS / SINS data, calculate the position result and determine the position error through real-time GNSS positioning algorithm, and save the values ​​of GNSS observation environment related parameters (including PDOP, carrier-to-noise ratio characteristic value, multipath error, etc.) and solution quality related parameters (including positional accuracy, number of continuous fixed satellites, ambiguity fixed ratio value, etc.) generated during the process. S1.2: Using the Spearman order nonlinear correlation coefficient calculation method, the correlation analysis of the GNSS observation environment and solution quality related parameter values ​​and position error values ​​obtained in the previous step is performed, and the correlation coefficient between the value of each parameter and the position error is calculated. S1.3: Select the parameter indicators with strong correlation. Strong correlation is defined as a correlation coefficient greater than 0.4. Define the parameter indicators that meet the conditions as multidimensional accuracy characterization factors and establish a set of multidimensional accuracy characterization factors to avoid the problem that a single parameter indicator cannot accurately reflect the position error.

[0031] Furthermore, the specific method of step S2 is as follows: S2.1: Based on more and larger amounts of historical GNSS / SINS data, calculate the position result and determine the position error through a real-time GNSS positioning algorithm, and save the multi-dimensional accuracy characterization factor determined in step S1. S2.2: Store the values ​​of multidimensional precision characterization factors and position error values ​​for each epoch, retain the inertial navigation hardware information of the acquired data (including gyroscope zero bias stability and accelerometer zero bias stability), and establish corresponding training datasets according to different hardware levels; S2.3: Using the established training dataset, a mapping relationship between multidimensional accuracy representation factors and position error is established through a convolutional neural network to obtain a position accuracy prediction model based on deep learning, thus solving the problem of inaccurate modeling of position error.

[0032] Furthermore, the specific method for step S3 is as follows: S3.1: Conduct online real-time GNSS positioning calculation and simultaneously obtain the multi-dimensional accuracy representation factor of the current epoch. Utilize the trained position accuracy prediction model to calculate the prediction error of the position result in real time. S3.2: The position prediction error is decomposed into different directions using the horizontal position precision factor (HDOP), vertical position precision factor (VDOP), and geometric position precision factor (PDOP) to obtain position prediction error components in different directions. These components are then used to fill the main diagonal of the prediction variance matrix, which is then converted into a position prediction variance matrix corresponding to the coordinate system in which the filtering solution is located. S3.3: Using the position prediction variance matrix as the variance of the current GNSS position result, this variance is closer to the actual variance value. After being applied in the Kalman filter solution of GNSS / SINS integrated navigation, it can avoid the inconsistency problem caused by the mismatch between the actual error and its variance, and improve the positioning and attitude determination effect in complex environments.

[0033] In another embodiment of the present invention, a method for elastically adjusting and optimizing the position variance of GNSS / SINS integrated navigation based on deep learning is provided, such as... Figure 1 , Figure 2 As shown, the process includes the construction of a pre-built GNSS / SINS positioning and attitude database, as well as key steps such as the selection of multi-dimensional accuracy representation factors, convolutional neural network training, variance elastic adjustment, and filter optimization. The selection of multi-dimensional accuracy representation factors and the training of the convolutional neural network are pre-prepared offline steps. These steps use nonlinear correlation calculation methods to determine multi-dimensional accuracy representation factors associated with position errors, and establish a training dataset based on the selected factors to conduct offline training of the position accuracy prediction model. The variance elastic adjustment and filter optimization are implemented online in real time. Using the trained model, the GNSS position accuracy is predicted online, and the position variance in the filter calculation is elastically adjusted to better match the actual position error with its variance. This results in improved accuracy during GNSS / SINS integrated navigation filter calculation. Essentially, this application presents an innovative method for extracting and modeling historical observation information to improve the real-time performance of current integrated navigation systems.

[0034] 1. Steps for selecting multidimensional precision characterization factors First, based on historical GNSS / SINS data, the location result is calculated using the same GNSS positioning algorithm as the real-time system. Simultaneously, the numerical values ​​of GNSS observation environment-related parameters (including PDOP, carrier-to-noise ratio C / N0, multipath error MP, etc.) and solution quality-related parameters (including in-place conformity accuracy STD, number of consecutive fixed satellites NSAT, and ambiguity ratio) generated during the calculation process are saved and recorded. These parameters can indirectly reflect the accuracy of the location result to a certain extent. For parameters such as PDOP, in-place conformity accuracy STD, number of consecutive fixed satellites NSAT, and ambiguity ratio, each epoch has a single value, which can correspond one-to-one with the location result of each epoch. However, for parameters such as carrier-to-noise ratio C / N0 and multipath error MP, each epoch has multiple values ​​(related to the number of GNSS satellites observed at that time), it is necessary to extract the characteristics of each epoch to obtain the statistical characteristic value of each epoch. Taking carrier-to-noise ratio C / N0 as an example, this invention uses the eigenvalue method to solve this problem. First, a membership matrix is ​​established. R As shown in Equation 1: (1) In the formula ,when i , j The closer the carrier-to-noise ratios of two satellites are, the greater their membership degree in the overall dataset. This is due to the membership matrix... R It is a non-negative real symmetric matrix with non-negative real eigenvalues, which can be obtained using the power iteration method. R Eigenvectors corresponding to the maximum modulus eigenvalues ​​of a matrix Thus, the weighting coefficient of a satellite's carrier-to-noise ratio relative to the whole is obtained as shown in Equation 2: (2) Finally, the C / N0 eigenvalues ​​are extracted based on the weighting coefficients. The expression is shown in Equation 3: (3) After obtaining the GNSS position results and the values ​​of parameters related to the GNSS observation environment and solution quality, the next step is to calculate the position error. Since obtaining the true position reference value is quite difficult (for example, for measured vehicle-mounted data, very high-precision inertial navigation is generally required as a reference, or a massive number of road control points are needed to obtain the true position reference value at various locations), a feasible method is to obtain the position reference value through bidirectional smoothing combined navigation post-processing. For GNSS positioning results in complex environments, the positioning results obtained through GNSS / SINS post-processing are closer to the true position value and can replace the true position reference value for calculating the position error.

[0035] This invention then employs the Spearman order nonlinear correlation coefficient calculation method to analyze the correlation between various parameters and position errors. Its advantage is that this method does not require the analyzed objects to be continuous or linearly correlated, and based on the ordinal ranking, it has a high tolerance for outliers in the analyzed objects. Correlation analysis is then performed on the GNSS observation environment and solution quality related parameter values ​​obtained in the previous step, along with the position error values, to calculate the correlation coefficients between each parameter value and the position error. Its expression is shown in Equation 4: (4) The formula first analyzes the object. and Sort, let and The position after sorting is and They referred to their ranks and, yes and The rank difference. Select the parameters with strong correlation, where equal correlation is defined as the correlation coefficient. Parameters exceeding a set threshold are defined as multidimensional accuracy representation factors. Each representation factor has its limitations. For example, PDOP only reflects the quality of GNSS satellite configuration, not the quality of received GNSS data; carrier-to-noise ratio (CNR) can only be used to assess the quality of received GNSS data signals, not position accuracy; and in-place coincidence accuracy (STD) is not only overly optimistic due to inconsistencies, but also only reflects the accuracy of GNSS floating-point solutions, not fixed-point solutions; while ambiguity ratio (Ratio) is related to the quality of fixed-point solutions, its value tends to 1 as the number of satellites increases. Therefore, it is necessary to establish an accurate set of multidimensional accuracy representation factors, incorporating as many parameters as possible related to position error, to support the subsequent deep learning model building.

[0036] like Figure 3 The figure shows the correlation sequence of the number of continuous fixed satellites (NSAT), position accuracy factor (PDOP), position conformity accuracy (STD), carrier-to-noise ratio (C / N0), and positioning error. The horizontal axis represents the number of epochs of measured data in a certain urban environment, and the vertical axis represents the parameter index and positioning error. It can be seen intuitively that no single parameter can characterize the position error. By using the Spearman order nonlinear correlation coefficient calculation method, it can be found that these parameter indexes have a strong correlation relationship and can be included in the multidimensional accuracy characterization factor set.

[0037] 2. Convolutional Neural Network Training Steps Based on a large amount of historical GNSS / SINS data, the position result is calculated using the same GNSS positioning algorithm as the real-time system. The multi-dimensional accuracy characterization factors recorded in the calculation process are saved (the set of multi-dimensional accuracy characterization factors is established by selecting the position accuracy factor PDOP, carrier-to-noise ratio characteristic value C / N0, the modulus of the post-hoc standard deviation STD, the number of consecutive fixed satellites NSAT, the fixed ambiguity ratio value, and the modulus of the solution difference SEP, which are related to the positioning error). The reference true value is obtained and the position error is calculated through a bidirectional smoothing combined navigation post-processing algorithm. This is similar to the aforementioned multi-dimensional accuracy characterization factor selection steps.

[0038] like Figure 4 As shown, a deep learning-based positional accuracy prediction model is obtained by establishing a mapping relationship between multidimensional accuracy representation factors and positional errors offline through a convolutional neural network. This invention employs a convolutional neural network because it has a strong ability to uncover complex implicit relationships that are difficult to model. The model uses a four-layer convolutional neural network, progressively increasing the number of nodes in each layer. After convolution, a fully connected layer is used to regress the results. Furthermore, the convolutional layers use the ReLU function as the activation function, and the fully connected layers use the tanh function. The loss function is constructed using the root mean square error between the predicted result and the true value. The input on the left is the established multidimensional accuracy representation factor, which has been normalized. The figure sequentially represents the input, convolutional layer, fully connected layer, and output, indicating the number of nodes in each layer. Finally, the absolute positional error is output.

[0039] 3. Variance elastic adjustment and filter solution optimization steps After establishing the mapping relationship between multidimensional accuracy representation factors and position errors offline beforehand, the position result and the set of multidimensional accuracy representation factors are obtained in real time and online through GNSS positioning calculation, and the current GNSS position error is predicted. This invention innovatively decomposes the position prediction error in different directions at each epoch using the horizontal position accuracy factor (HDOP), vertical position accuracy factor (VDOP), and geometric position accuracy factor (PDOP), obtaining position prediction error components in different directions. These components are then used to fill the main diagonal of the prediction variance matrix, as shown in Equation 5, to calculate the local coordinate system (…). l Location variance under (system) P l pos : (5) exist P l pos Based on this, it can be converted into the position prediction variance matrix corresponding to the coordinate system where the filtering solution is located. PFor example, if the filtering solution uses the ECEF geocentric-geofixed coordinate system, then it can be transformed into Equation 6. P e pos : (6) In Kalman filtering calculations for GNSS / SINS integrated navigation, the position error of mechanical arrangement can generally be selected. Speed ​​error Misalignment angle and accelerometer zero bias Zero bias of the gyroscope The error state vector is constructed using a total of 15 dimensions, as shown in Equation 7: (7) Its error state differential equation can be expressed as: (8) In the formula F Here is the state transition matrix. It is process noise, where the state transition matrix F The expression is: (9) In the formula Indicates e-series relative to i The projection of the angular velocity of the system into the e-frame. This represents the rotation matrix from frame b to frame e. This refers to the force in the e-frame, and this information is provided by the inertial navigation system. Discretizing the state transition process yields: (10) In the formula It is the discretized error state transition matrix. P Let Variance be the variance matrix of the states. It is the noise matrix of the random process, which is the error state equation of the Kalman filter.

[0040] The observation equations are constructed using navigation parameter information obtained separately from GNSS. After considering the lever relationship, the commonly used position and velocity observation equations are as follows: (11) In the formula , and All data is obtained from the current inertial navigation mechanical programming. It is the arm vector of the GNSS center relative to the SINS center, obtained directly from measurements in the b-frame. Depend on Calculations show that The expansion is Then, a Kalman filter is performed to update the gain: (12) In the formula This refers to the observation variance, which is typically constructed from the GNSS internal coincidence accuracy. Because GNSS is susceptible to nonlinear errors, non-Gaussian errors, and observational gross errors, leading to distortions in positioning statistics, the internal coincidence accuracy usually does not match the actual error and tends to be more optimistic. This can easily cause Kalman filters to incorrectly accept anomalous GNSS results, resulting in significant deviations in error state estimates and reducing the accuracy and reliability of positioning and attitude determination. The core of this invention at this stage is to utilize the predicted position variance. P e pos Instead of observation variance Calculate the filter gain, such as Figure 5 As shown, the predicted position variance is very close to the actual position error, exhibiting a flexible adjustment effect. Finally, the estimated values ​​of the error state and its variance are shown in Equations 12 and 13, where... The new information observed is calculated by subtracting the GNSS position from the position of the inertial navigation mechanical arrangement of the corrected lever arm.

[0041] (13) (14) This invention optimizes the filtering process, improving the accuracy and reliability of positioning and orientation determination in complex environments through pre-training offline and real-time online adjustment. The results are as follows: Figure 6 As shown.

[0042] In another embodiment of the present invention, to facilitate better implementation of the method provided in the embodiments of the present invention, the present invention also provides a system based on the above method. The meanings of the terms are the same as in the above method, and specific implementation details can be found in the description of the method embodiments.

[0043] Please see Figure 7 , Figure 7 This is a schematic diagram of the system provided in an embodiment of the present invention. The system may include a multidimensional precision representation factor selection module 701, a convolutional neural network training module 702, and a variance elasticity adjustment and filtering solution optimization module 703, wherein: The multidimensional accuracy characterization factor selection module 701 is used to extract GNSS observation environment parameters and solution quality parameters based on GNSS and / or SINS historical data, calculate the correlation between each parameter and position error through Spearman order correlation coefficient, and screen parameters with correlation coefficients greater than a set threshold to form a multidimensional accuracy characterization factor set. The convolutional neural network training module 702 is used to construct a training dataset based on historical data. The input is a multidimensional precision representation factor and the output is the position error. It is trained offline through a convolutional neural network to establish a position error prediction model. The variance elastic adjustment and filtering optimization module 703 is used to obtain the multi-dimensional accuracy characterization factor of the current epoch during real-time navigation, input it into the trained position error prediction model, obtain the position error prediction value, decompose the position error prediction value into each direction based on HDOP, VDOP, and PDOP, construct the position prediction variance matrix, and substitute the prediction variance matrix as the observation variance into the Kalman filter of GNSS / SINS integrated navigation to realize dynamic adjustment of variance and optimization of filtering performance.

[0044] The specific implementation methods of each module can be referred to the description of the above method embodiments, and the embodiments of the present invention will not be repeated.

[0045] In another embodiment of the present invention, a computer device is also provided, such as... Figure 8 As shown, it illustrates a structural schematic diagram of a computer device involved in an embodiment of the present invention, specifically: The computer device may include components such as a processor 801 with one or more processing cores, a memory 802 with one or more computer-readable storage media, a power supply 803, and an input unit 804. Those skilled in the art will understand that... Figure 8 The computer device structure shown does not constitute a limitation on the computer device and may include more or fewer components than shown, or combine certain components, or have different component arrangements. Wherein: The processor 801 is the control center of the computer device. It connects various parts of the computer device via various interfaces and lines, and performs various functions and processes data by running or executing software programs and / or modules stored in the memory 802, and by calling data stored in the memory 802, thereby providing overall monitoring of the computer device. Optionally, the processor 801 may include one or more processing cores; preferably, the processor 801 may integrate an application processor and a modem processor, wherein the application processor mainly handles the operation of the storage medium, user interface, and application programs, and the modem processor mainly handles wireless communication. It is understood that the modem processor may not be integrated into the processor 801.

[0046] The memory 802 can be used to store software programs and modules. The processor 801 executes various functional applications and data processing by running the software programs and modules stored in the memory 802. The memory 802 may mainly include a program storage area and a data storage area. The program storage area may store application programs required for operating the storage medium and at least one function (such as sound playback function, image playback function, etc.); the data storage area may store data created according to the use of the computer device. In addition, the memory 802 may include high-speed random access memory and may also include non-volatile memory, such as at least one disk storage device, flash memory device, or other volatile solid-state storage device. Accordingly, the memory 802 may also include a controller to provide the processor 801 with access to the memory 802.

[0047] The computer device also includes a power supply 803 that supplies power to various components. Preferably, the power supply 803 can be logically connected to the processor 801 via a power management storage medium, thereby enabling functions such as charging, discharging, and power consumption management through the power management storage medium. The power supply 803 may also include one or more DC or AC power supplies, recharge storage media, power fault detection circuits, power converters or inverters, power status indicators, and other arbitrary components.

[0048] The computer device may also include an input unit 804, which can be used to receive input digital or character information and generate keyboard, mouse, joystick, optical or trackball signal inputs related to user settings and function control.

[0049] Although not shown, the computer device may also include a display unit, etc., which will not be described in detail here. Specifically, in this embodiment, the processor 801 in the computer device loads the executable files corresponding to the processes of one or more applications into the memory 802 according to the following instructions, and the processor 801 runs the applications stored in the memory 802, thereby implementing the steps in the above method embodiment.

[0050] Those skilled in the art will understand that all or part of the steps in the various methods of the above embodiments can be performed by instructions, or by instructions controlling related hardware. These instructions can be stored in a computer-readable storage medium and loaded and executed by a processor.

[0051] Therefore, embodiments of the present invention provide a computer-readable storage medium having a computer program stored thereon, the computer program being loaded by a processor to execute the steps of any method provided in the embodiments of the present invention.

[0052] For details on the implementation of each of the above operations, please refer to the previous examples, which will not be repeated here.

[0053] The computer-readable storage medium may include: read-only memory (ROM), random access memory (RAM), disk or optical disk, etc.

[0054] Since the computer program stored in the computer-readable storage medium can execute the steps of any of the methods provided in the embodiments of the present invention, the beneficial effects that any of the methods provided in the embodiments of the present invention can achieve can be realized, as detailed in the preceding embodiments, and will not be repeated here.

[0055] The foregoing has provided a detailed description of a method, system, device, and medium for elastic adjustment and optimization of position variance based on deep learning in integrated navigation, as provided in the embodiments of the present invention. Specific examples have been used to illustrate the principles and implementation methods of the present invention. The descriptions of the above embodiments are only for the purpose of helping to understand the method and core ideas of the present invention. At the same time, for those skilled in the art, there will be changes in the specific implementation methods and application scope based on the ideas of the present invention. Therefore, the content of this specification should not be construed as a limitation of the present invention.

Claims

1. A method for deep learning-based combined navigation position variance elasticity adjustment optimization, characterized in that, include: Based on historical GNSS and / or SINS data, GNSS observation environment parameters and solution quality parameters are extracted. The correlation between each parameter and position error is calculated using the Spearman order correlation coefficient. Parameters with correlation coefficients greater than a set threshold are selected to form a set of multidimensional accuracy characterization factors. A training dataset is constructed based on historical data. The input is a multidimensional accuracy representation factor and the output is the position error. The position error prediction model is established through offline training using a convolutional neural network. During real-time navigation, the multidimensional accuracy representation factor of the current epoch is obtained and input into the trained position error prediction model to obtain the position error prediction value. Based on HDOP, VDOP, and PDOP, the position error prediction value is decomposed into each direction to construct the position prediction variance matrix. The position prediction variance matrix is ​​then used as the observation variance and substituted into the Kalman filter of the GNSS / SINS integrated navigation to achieve dynamic adjustment of variance and optimization of filtering performance.

2. The method of claim 1, wherein, The GNSS observation environment parameters include: position accuracy factor (PDOP), carrier-to-noise ratio characteristic value, and multipath error; the solution quality parameters include: position conformity accuracy (STD), number of consecutive fixed satellites (NSAT), and ambiguity fixed ratio value.

3. The method of claim 2, wherein, The correlation between each parameter and the position error is calculated using the Spearman order correlation coefficient. Parameters with correlation coefficients greater than a set threshold are selected to form a multidimensional accuracy characterization factor set, including: Based on GNSS and / or SINS historical data, the same GNSS positioning algorithm as the real-time system is used to perform offline calculations on the historical data to obtain the historical position calculation results for each epoch. High-precision integrated navigation post-processing technology is used to obtain the corresponding position reference values. The position error for each epoch is obtained from the position reference values ​​and the position calculation results obtained by the GNSS positioning algorithm. At the same time, the values ​​of GNSS observation environment related parameters and calculation quality related parameters for each epoch are recorded. The Spearman order nonlinear correlation coefficient calculation method is used to perform correlation analysis on the obtained GNSS observation environment related parameter values ​​and solution quality related parameter values ​​with position error, and to calculate the correlation coefficient between each parameter value and position error. The parameter index with a correlation coefficient greater than a set threshold is selected as the multidimensional accuracy characterization factor.

4. The method according to any one of claims 1 to 3, characterized in that, The convolutional neural network includes at least four convolutional layers, with ReLU as the activation function, tanh as the activation function for fully connected layers, and root mean square error as the loss function.

5. The method of claim 4, wherein, The training dataset constructed based on historical data, with multidimensional accuracy representation factors as input and position error as output, is trained offline using a convolutional neural network to establish a position error prediction model, including: Based on GNSS and / or SINS historical data, the same GNSS positioning algorithm as the real-time system is used to perform offline calculations on the historical data to obtain the historical position calculation results for each epoch. High-precision integrated navigation post-processing technology is used to obtain the corresponding position reference value. By comparing the two, the position error of each epoch is calculated and used as the real label for subsequent model training. The training dataset is constructed by using the multidimensional accuracy representation factor and the corresponding position error for each epoch. Based on the inertial navigation hardware information of the collected data, the training dataset is grouped according to different hardware levels and split into multiple sub-datasets according to hardware level. Each sub-dataset corresponds to an inertial navigation hardware with a specific performance level. For each subset of data, a convolutional neural network is trained independently. The mapping relationship between multidimensional accuracy representation factors and position error is established through the convolutional neural network, resulting in a position error prediction model that matches the hardware level.

6. The method of claim 5, wherein, The method for constructing the position prediction variance matrix is: decomposing the position error prediction value into different directions to obtain position prediction error components in different directions, filling the main diagonal line of the prediction variance matrix by the position prediction error components in different directions, and converting into the position prediction variance matrix corresponding to the coordinate system of the filter solution. , , , , are position variance components in east, north and sky directions respectively.

7. The method of claim 6, wherein, The state vector of the Kalman filter includes: position error, velocity error, misalignment angle, accelerometer zero bias, and gyroscope zero bias.

8. A deep learning-based combined navigation position variance elasticity adjustment optimization system, characterized in that, include: The multidimensional accuracy characterization factor selection module is used to extract GNSS observation environment parameters and solution quality parameters based on GNSS and / or SINS historical data, calculate the correlation between each parameter and position error through Spearman order correlation coefficient, and select parameters with correlation coefficients greater than a set threshold to form a multidimensional accuracy characterization factor set. The convolutional neural network training module is used to build a training dataset based on historical data. The input is a multidimensional accuracy representation factor and the output is the position error. It is trained offline through a convolutional neural network to establish a position error prediction model. The variance elastic adjustment and filtering optimization module is used to obtain the multi-dimensional accuracy characterization factor of the current epoch during real-time navigation, input it into the trained position error prediction model, obtain the position error prediction value, decompose the position error prediction value into each direction based on HDOP, VDOP, and PDOP, construct the position prediction variance matrix, and substitute the prediction variance matrix as the observation variance into the Kalman filter of GNSS / SINS integrated navigation to realize dynamic adjustment of variance and optimization of filtering performance.

9. A computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor, when executing the computer program, implements the steps of the method according to any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the method according to any one of claims 1 to 7.

Images