Real-time satellite clock error estimation reference jump compensation method based on deep learning

Through the deep learning model, the value after the change of the satellite clock difference reference is solved, and the accurate estimation problem when the satellite clock difference reference is changed is achieved, and the accurate compensation of the satellite clock difference is achieved, ensuring the stability of navigation positioning and timing services.

CN118425997BActive Publication Date: 2024-11-08WUHAN UNIV
View PDF 6 Cites 0 Cited by

Patent Information

Application Number
CN202410524689.1
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2024-04-29
Publication Date
2024-11-08
Estimated Expiration
2044-04-29

AI Technical Summary

Technical Problem

The prior art is difficult to provide accurate real-time satellite clock difference estimates when the satellite clock difference reference changes, and there is a clock difference jump problem, which affects the continuity and stability of navigation positioning and timing services.

Method used

Using the real-time satellite clock difference estimation benchmark jump compensation method based on deep learning, the trained satellite clock difference prediction model predicts the change of the clock difference reference value, and calculates the compensation value using the clock difference estimation and prediction value to realize the satellite Accurate compensation for clock failure.

Benefits of technology

When the satellite clock difference reference jumps, accurate satellite clock difference estimation can be achieved, avoiding the impact of reducing clock difference estimation accuracy and ensuring high-precision navigation positioning and timing services.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118425997B_ABST
    Figure CN118425997B_ABST
Patent Text Reader

Abstract

The embodiment of the present application discloses a real-time satellite clock error estimation benchmark jump compensation method based on deep learning, which relates to the field of GNSS navigation positioning and timing technology. The method includes: real-time acquisition of clock error estimation values ​​of all observed satellites; if it is determined that the clock error benchmark of the satellite clock has changed according to the clock error estimation values ​​of all satellites, the clock error estimation values ​​of multiple epochs before the clock error benchmark changes are input into the satellite clock error prediction model, and the clock error prediction values ​​of each epoch are obtained through model prediction, and then the clock error compensation value of the corresponding epoch is calculated, and the satellite to be compensated is compensated according to the clock error compensation value. The present application calculates the clock error compensation value using the clock error prediction value and the clock error estimation value, and then can compensate for the satellite clock clock after the benchmark is changed. Even if the clock error benchmark jumps, accurate satellite clock estimation can be achieved to avoid the impact of reduced clock error estimation accuracy caused by the satellite clock benchmark jump, and provide technical support for high-precision navigation and positioning.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present application relates to the field of GNSS navigation, positioning and timing technology, and in particular to a real-time satellite clock error estimation reference jump compensation method based on deep learning. Background Art

[0002] Precise Point Positioning (PPP) technology is widely used in various fields, such as traffic navigation, precision agriculture, GNSS meteorology, aviation and navigation, map making, emergency rescue, military applications, personal positioning services, timing services and scientific research.

[0003] Precise point positioning does not rely on base stations. It is based only on the post-precision orbit and clock products provided by the International GPS service for Geodynamics (IGS) or the Multi-GNSS Experiment (MGEX). It can provide post-centimeter-level or even millimeter-level position determination and timing services better than 1ns on a global scale without the need for other infrastructure support. With the continuous advancement of satellite navigation technology, the improvement of the performance of GNSS receivers and computing equipment, and the increasing demand for high-precision positioning and real-time timing in various fields, real-time precise point positioning technology has developed rapidly and can provide real-time decimeter-level navigation positioning and nanosecond-level timing services.

[0004] Continuous and stable real-time satellite clock error estimation is a prerequisite for ensuring real-time precise single-point positioning to provide high-precision navigation positioning and timing services. Since the satellite atomic clock is affected by various factors such as the space environment during its in-orbit operation, the satellite clock error sequence is complex and exhibits nonlinear changes. Therefore, it is difficult to use the historical information of satellite clock errors to accurately fit and predict satellite clock errors in the long term.

[0005] At present, in order to obtain high-precision and high-reliability real-time satellite clock errors, it is still necessary to rely on ground station observations for real-time estimation. However, real-time satellite clock errors still have defects such as unstable benchmarks, low algorithm efficiency, difficult quality control, and poor solution accuracy. The receiver clock can be used as a benchmark, but this method is greatly affected by the surrounding environment. For example, when the receiver encounters problems such as network communication delay, the satellite clock error estimation will lose the benchmark, resulting in unstable benchmarks. At this time, the receiver needs to be replaced as a new benchmark, and the satellite clock error will jump significantly; if the satellite clock is used as a benchmark, this method requires the benchmark star to be able to observe for a long time, otherwise the benchmark star will need to be replaced frequently, and the solved clock error will also jump, which will affect the continuity and stability of positioning and timing services; the satellite zero mean can also be used as the clock error benchmark, but when a new satellite is observed and loses lock with the old satellite, the benchmark will change, and the real-time satellite clock error will also face the problem of jumping.

[0006] In summary, the relevant technology cannot provide accurate real-time satellite clock error when the reference changes, and cannot solve the problem of clock error jump. Summary of the invention

[0007] The embodiment of the present application provides a real-time satellite clock error estimation reference jump compensation based on deep learning, which is used to solve the defects of the related art that it cannot provide accurate real-time satellite clock error when the reference changes and cannot solve the problem of clock error jump. The technical solution is as follows:

[0008] In a first aspect, an embodiment of the present application provides a real-time satellite clock error estimation reference jump compensation method based on deep learning, comprising:

[0009] Collect clock error estimates of all observed satellites in real time;

[0010] If it is determined that the clock error reference of the satellite clock has changed according to the clock error estimation values ​​of all the satellites, the clock error estimation values ​​of a plurality of epochs before the clock error reference has changed are input into the trained satellite clock error prediction model;

[0011] Predicting the clock error prediction value of each epoch after the clock error reference changes by the trained satellite clock error prediction model;

[0012] The clock error compensation value of the satellite to be compensated in each epoch is calculated according to the clock error estimation value of the satellite to be compensated in each epoch and the clock error prediction value of the corresponding epoch;

[0013] The satellite to be compensated is compensated according to the clock error compensation value of the satellite to be compensated at each epoch.

[0014] In an optional solution of the first aspect, inputting the clock error estimation values ​​of a plurality of epochs before the clock error reference changes into a trained satellite clock error prediction model, and predicting the clock error prediction value of each epoch after the clock error reference changes by the trained satellite clock error prediction model, comprises:

[0015] Taking the estimated clock error values ​​of all satellites in the plurality of epochs as input values, normalizing the input values;

[0016] Inputting the normalized input value into the convolution layer of the trained satellite clock error prediction model, and increasing the variable dimension through the convolution layer;

[0017] Performing a dimension exchange operation to exchange the variable dimension and the time dimension of the input value after the variable dimension is increased, and obtaining a feature vector by feature extraction;

[0018] Inputting the feature vector into the attention mechanism layer of the trained satellite clock error prediction model, and extracting the clock error time series features through the attention mechanism layer;

[0019] Input the clock error time series features into the first linear layer of the trained satellite clock error prediction model, predict the change trend of the clock error prediction value based on the clock error time series features, and output the prediction result;

[0020] Input the prediction result into the second linear layer of the trained satellite clock error prediction model, and after exchanging the time dimension and the variable dimension of the prediction result through the second linear layer, input the prediction result into the denormalization layer of the satellite clock error prediction model;

[0021] The corresponding data is restored through the denormalization layer, and the clock error prediction values ​​of multiple epochs after the clock error reference changes are output.

[0022] In an optional solution of the first aspect, the clock error estimation values ​​of all satellites in the multiple epochs are used as input values, and after normalizing the input values, the process further includes:

[0023] If the number of variables in the normalized input value is less than a preset threshold, data supplementation is performed on the normalized input value;

[0024] The input value after data supplementation is input into the convolution layer of the satellite clock error prediction model, and the variable dimension is increased through the convolution layer.

[0025] In an optional solution of the first aspect, the trained satellite clock error prediction model is trained based on the following steps, including:

[0026] A training set and a validation set are established based on the clock error estimates of multiple satellites with the same clock error reference in the same time span;

[0027] Inputting the training set into the satellite clock error prediction model for training;

[0028] Inputting the verification set into the satellite clock error prediction model, and calculating the loss function of multiple epochs according to the output result of the satellite clock error prediction model;

[0029] If the value of the loss function increases for a preset number of consecutive times, the training is completed and the trained satellite clock error prediction model is output.

[0030] In an optional solution of the first aspect, a test set is established based on clock error estimates of multiple satellites having the same clock error reference in the same time span;

[0031] After outputting the trained satellite clock error prediction model, the method further includes:

[0032] The test set is input into the trained satellite clock error prediction model, and the prediction accuracy of the trained satellite clock error prediction model for the clock error prediction value of each epoch is calculated.

[0033] In an optional solution of the first aspect, the clock error compensation value of the satellite to be compensated at each epoch is calculated based on the clock error estimation value of the satellite to be compensated at each epoch and the clock error prediction value of the corresponding epoch, including:

[0034] For the first epoch after the clock error reference changes, the clock error compensation value of the first epoch is calculated based on the difference between the clock error estimate value of the satellite to be compensated in the first epoch and the clock error prediction value of the corresponding epoch;

[0035] For the nth epoch after the clock error reference changes, calculate the difference between the clock error estimate of the satellite to be compensated at the nth epoch and the clock error prediction value of the corresponding epoch, use the prediction accuracy of the trained satellite clock error prediction model at the nth epoch as the weight of the difference between the clock error estimate and the clock error prediction value, and obtain the clock error compensation value of the nth epoch by weighted calculation;

[0036] Here, n is an integer greater than 1.

[0037] In an optional solution of the first aspect, the clock error compensation value of the satellite to be compensated at each epoch is calculated based on the clock error estimation value of the satellite to be compensated at each epoch and the clock error prediction value of the corresponding epoch, including:

[0038] Calculate and obtain the clock error compensation value of the satellite to be compensated at the nth epoch, and use the clock error compensation value of the nth epoch as the clock error compensation value of each epoch after the nth epoch;

[0039] Here, n is an integer greater than or equal to 4.

[0040] In a second aspect, an embodiment of the present application further provides a real-time satellite clock error estimation reference jump compensation device based on deep learning, comprising:

[0041] The observation module is used to collect the clock error estimates of all observed satellites in real time;

[0042] A clock error prediction module, used for inputting the clock error estimation values ​​of a plurality of epochs before the clock error estimation values ​​of all satellites change into a trained satellite clock error prediction model when the observation module determines that the clock error reference of the satellite clock changes according to the clock error estimation values ​​of all satellites, and predicting the clock error prediction values ​​of a plurality of epochs after the clock error reference changes by the trained satellite clock error prediction model;

[0043] A clock error compensation module is used to calculate the clock error compensation value of the satellite to be compensated in each epoch according to the clock error estimation value of the satellite to be compensated in each epoch and the clock error prediction value of the corresponding epoch;

[0044] The clock error compensation module is also used to compensate the satellite to be compensated according to the clock error compensation value of the satellite to be compensated in each epoch.

[0045] In a third aspect, an embodiment of the present application further provides an electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, the method provided in the first aspect of the embodiment of the present application or any one of the implementation methods of the first aspect is implemented.

[0046] In a fourth aspect, the present application also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the method provided by the first aspect of the embodiment of the present application or any one of the implementations of the first aspect.

[0047] The beneficial effects brought about by the technical solutions provided by some embodiments of the present application include at least:

[0048] The embodiment of the present application provides a real-time satellite clock error estimation benchmark jump compensation method based on deep learning. When the satellite clock error benchmark changes, the clock error of the satellite clock can be predicted by a trained satellite clock error prediction model. Since all satellites are under the same benchmark, the trained satellite clock error prediction model can capture the original benchmark information after the clock error benchmark changes, thereby achieving accurate prediction. The clock error compensation value is then calculated using the clock error prediction value and the clock error estimation value, thereby compensating for the satellite clock clock after the benchmark is changed. Even if the clock error benchmark jumps, accurate satellite clock estimation can be achieved, thereby avoiding the impact of reduced clock error estimation accuracy caused by satellite clock benchmark jumps, and providing technical support for high-precision navigation and positioning. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] In order to more clearly illustrate the technical solutions in the present application or related technologies, the drawings required for use in the embodiments or related technical descriptions are briefly introduced below. Obviously, the drawings described below are some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0050] Figure 1 A schematic diagram of the architecture of a real-time satellite clock error estimation reference jump compensation system based on deep learning in an embodiment of the present application;

[0051] Figure 2 It is a flowchart of a method for real-time satellite clock error estimation reference jump compensation based on deep learning in an embodiment of the present application;

[0052] Figure 3 is a schematic diagram of a satellite clock error prediction model provided in an embodiment of the present application;

[0053] Figure 4 It is a structural schematic diagram of a real-time satellite clock error estimation reference jump compensation device based on deep learning provided in an embodiment of the present application;

[0054] Figure 5 It is a schematic diagram of the structure of an electronic device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0055] In order to make the purpose, technical solutions and advantages of this application clearer, the technical solutions in this application will be clearly and completely described below in conjunction with the drawings in this application. Obviously, the described embodiments are part of the embodiments of this application, not all of them. Based on the embodiments in this application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0056] The terms "including" and "having" and any variations thereof in the specification and claims of the present application and the above drawings are intended to cover non-exclusive inclusions. For example, a process, method, system, product or device including a series of steps or modules is not limited to the listed steps or modules, but may optionally include steps or modules that are not listed, or may optionally include other steps or modules that are inherent to these processes, methods, products or devices.

[0057] It should be noted that the terms "first\second" involved in the present application are only used to distinguish similar objects, and do not represent a specific order for the objects. It is understandable that "first\second" can be interchanged with a specific order or sequence where permitted. It should be understood that the objects distinguished by "first\second" can be interchanged where appropriate, so that the embodiments of the present application described herein can be implemented in an order other than those described or illustrated herein.

[0058] Please refer to the following Figure 1 , which is a schematic diagram of the architecture of a real-time satellite clock error estimation reference jump compensation system based on deep learning provided by an exemplary embodiment of the present application, such as Figure 1 As shown, the system includes an information transceiver unit 110 and a calculation unit 120 .

[0059] Among them, the information collection and transceiver unit 110 can be a device for observing satellites, and can exchange information with the observed satellites. For example, it can obtain the clock error estimates of all satellites in real time, and determine whether the clock error reference of the satellite clock has changed based on the clock error estimates; the information collection and transceiver unit 110 can send the observation data of multiple epochs before the change to the calculation unit 120 when the satellite clock error reference changes.

[0060] The calculation unit 120 may be a terminal, server, etc. that provides a variety of real-time satellite clock error estimation reference jump compensation calculations based on deep learning. The calculation unit 120 can obtain a real-time clock error estimation value through the information transceiver unit 110, and when the clock error reference of the satellite clock changes, the calculation unit 120 can use the trained satellite clock error prediction model on board to calculate the clock error prediction value of the subsequent multiple epochs based on the clock error estimation value of the previous multiple epochs, and calculate the clock error compensation value of the satellite to be compensated in each epoch according to the clock error estimation value of the satellite to be compensated in each epoch and the clock error prediction value of the corresponding epoch, and compensate the satellite to be compensated according to the clock error compensation value of the satellite to be compensated in each epoch, thereby solving the jump problem of the satellite clock error.

[0061] Understandably, Figure 1The number of information transceiver units and computing units in the real-time satellite clock error estimation reference jump compensation system shown is only an example. In a specific implementation, the real-time satellite clock error estimation reference jump compensation system may include any number of information transceiver units and computing units. The embodiment of the present application does not specifically limit this. For example, but not limited to, the information transceiver unit may be an information transceiver unit composed of multiple information sending terminals and information receiving terminals, and the computing unit may be a cluster of terminal devices composed of multiple terminal devices.

[0062] The present application is described in detail below with reference to specific embodiments.

[0063] Next, combine Figure 1 , introduces the real-time satellite clock error estimation reference jump compensation method based on deep learning provided by the embodiment of the present application. For details, please refer to Figure 2 ,like Figure 2 As shown, the real-time satellite clock error estimation reference jump compensation method based on deep learning includes the following steps:

[0064] S201, collecting clock error estimates of all observed satellites in real time.

[0065] It is understandable that it is not difficult to obtain the real-time clock difference estimates of all satellites through observation equipment, and the embodiment of the present application does not limit the specific method for collecting the clock difference estimates.

[0066] Furthermore, it is possible to determine whether there is a change in the clock error benchmark based on the clock error estimates of all satellites, for example, whether the receiver serving as the benchmark is interfered with or the satellite serving as the benchmark is lost, or whether the satellite zero mean is used as the benchmark, or whether a new satellite is observed and the original satellite is lost. The embodiments of the present application are not limited to this.

[0067] Further, if it is determined that the clock error reference of the satellite clock has changed, the following steps are performed:

[0068] S202, inputting the clock error estimation values ​​of multiple epochs before the clock error reference changes into the trained satellite clock error prediction model.

[0069] Specifically, the clock error estimates of multiple epochs before the clock error reference changes can be understood as taking the moment when the clock error reference changes as the dividing point, extracting the clock error estimates within a time period along the time axis to the left from the dividing point as the endpoint. This time period can be divided into multiple epochs, and the clock error estimates of each satellite in each epoch within this time period can be obtained.

[0070] It can be understood that each clock error estimate input into the trained satellite clock error prediction model is independent of each other, that is, each satellite corresponds to different clock error estimate data in multiple epochs, and multiple satellites correspond to different clock error estimate data in the same epoch, but should have the same reference.

[0071] Preferably, data from 12 epochs before the clock error reference changes are selected as input values ​​for the satellite clock error prediction model.

[0072] S203, predicting the clock error prediction value of each epoch after the clock error reference changes through the trained satellite clock error prediction model.

[0073] It should be noted that before training the satellite clock error prediction model, a training set can be created based on the satellite clock error data of the same benchmark. Each training set can have a different number of features but the same time span. For example, each training set can use a fixed 12 epoch data as input to predict the next 4 epoch data, containing a total of 16 epochs of data. During the training process, the satellite clock error prediction model learns the common benchmark information between different satellite clock error variables based on the input training set.

[0074] Specifically, during the training process, the parameters of the satellite clock error prediction model can be updated through the back-propagation algorithm and optimizer, so that the satellite clock error prediction model can gradually learn the characteristics and rules of the input data. Some important parameters can also be set, such as learning rate, batch size, and number of training cycles. By adjusting these parameters, the satellite clock error prediction model can converge better and obtain better prediction performance.

[0075] Optionally, data augmentation, regularization, and early stopping can be used to avoid model overfitting and underfitting problems.

[0076] In some embodiments, when establishing a training set, a portion of the data can be selected to generate a validation set. The validation set can be input into a satellite clock error prediction model, and the loss function of multiple epochs can be calculated based on the output results of the satellite clock error prediction model. The satellite clock error prediction model can be judged whether its training is complete based on the results of the loss function.

[0077] Specifically, the loss function of each epoch can be obtained in turn to determine the changing trend of the value of the loss function. If the values ​​of the loss function for a preset number of consecutive times are all increasing, it indicates that the satellite clock error prediction model has converged, and the trained satellite clock error prediction model is output.

[0078] Preferably, the preset number can be selected as 3, that is, when the loss of the validation set increases for three consecutive epochs, it indicates that the training is completed.

[0079] In some embodiments, when establishing a training set, a portion of the data can be selected to generate a test set, and the prediction accuracy of the clock error prediction values ​​of different epochs predicted by the satellite clock error prediction model obtained through preliminary training can be tested through the test set, and a weight can be added to the difference between the clock error estimate and the clock error prediction value of the satellite to be compensated in the corresponding epoch according to the prediction accuracy.

[0080] Optionally, the accuracy of the satellite clock error prediction model obtained through preliminary training can be further used to determine whether it meets the accuracy requirements. If the accuracy does not meet the required accuracy threshold, retraining can be performed by adjusting parameters, etc. This embodiment of the present application is not limited to this.

[0081] Based on this, the trained satellite clock error prediction model can determine the common benchmark information between different satellite clock error variables based on the input clock error estimate.

[0082] S204, calculating the clock error compensation value of the satellite to be compensated in each epoch according to the clock error estimation value of the satellite to be compensated in each epoch and the clock error prediction value of the corresponding epoch.

[0083] Specifically, for the first epoch after the clock error reference changes, the clock error compensation value of the first epoch is calculated based on the difference between the clock error estimate value of the satellite to be compensated in the first epoch and the clock error prediction value of the corresponding epoch;

[0084] Specifically, for multiple epochs after the first epoch, for example, the nth epoch after the clock error reference changes, the difference between the clock error estimate of the satellite to be compensated at the nth epoch and the clock error prediction value of the corresponding epoch can be calculated. The prediction accuracy of the trained satellite clock error prediction model at the nth epoch is taken as the weight of the difference between the clock error estimate and the clock error prediction value, and the clock error compensation value of the nth epoch is obtained by weighted calculation.

[0085] Here, n is an integer greater than 1.

[0086] It can be understood that the weight of the difference between the estimated clock error value and the predicted clock error value at each epoch is determined by the prediction accuracy of the satellite clock error prediction model at the corresponding epoch.

[0087] For example, for example, the clock error prediction values ​​of four epochs are predicted, and the prediction accuracy of the satellite clock error prediction model in predicting the clock error prediction value of each epoch is known. The process of calculating the clock error compensation value includes:

[0088] For the first epoch, apply the formula:

[0089]

[0090] Calculate the clock error compensation value of the first epoch

[0091] For the second epoch, this includes:

[0092] Calculate the difference between the estimated clock error of the first epoch and the predicted clock error of the epoch

[0093] The difference between the estimated clock error of the second epoch and the predicted clock error of this epoch is

[0094] The difference between the estimated clock error and the predicted clock error is obtained based on the prediction accuracy of the satellite clock error prediction model at the first epoch. The weighted RMSE1 is:

[0095] P1=1 / RMSE1;

[0096] The difference between the estimated clock error and the predicted clock error is obtained based on the prediction accuracy of the satellite clock error prediction model in the second epoch. The weighted RMSE2 is:

[0097] P2=1 / RMSE2;

[0098] Calculate the clock compensation value of the second epoch, perform weighted calculation based on the weight, and apply the formula:

[0099]

[0100] Calculate the clock error compensation value of the first epoch

[0101] It should be noted that the longer the prediction time is, the worse the prediction accuracy is. Therefore, starting from the second epoch, for the second epoch and subsequent epochs, the weighted average of the difference between the clock error prediction value and the clock error estimate value at different epochs is used as the clock error compensation value.

[0102] Specifically, the formula for calculating the clock error compensation value of the nth epoch can be inferred from this, where n is an integer greater than 1:

[0103] P n =1 / RMSE n ;

[0104]

[0105]

[0106] Among them, the variable is the clock error compensation value of the nth epoch, is the estimated clock error at the nth epoch, is the predicted value of the clock error at the nth epoch, is the difference between the estimated clock error and the predicted clock error at the nth epoch, P n is the weight of the difference between the estimated clock error and the predicted clock error at the nth epoch, RMSE n is the accuracy of the clock error prediction value at the nth epoch.

[0107] Exemplarily, the clock error compensation value of the 4th epoch is calculated, and the formula is applied:

[0108]

[0109] Calculate the clock compensation value of the 4th epoch

[0110] In some embodiments, since the longer the prediction time is, the lower the prediction accuracy is, the clock error compensation value of a preset epoch can be selected as the clock error compensation value of the epoch and subsequent epochs based on the prediction accuracy.

[0111] Exemplarily, the clock error compensation value of the satellite to be compensated at the nth epoch can be calculated, and the clock error compensation value of the nth epoch is used as the clock error compensation value of each epoch after the nth epoch.

[0112] Preferably, n is an integer greater than or equal to 4.

[0113] For example, for the 4th epoch and subsequent epochs, the clock error compensation value of the 4th epoch is is the clock error compensation value for all epochs after the 4th epoch.

[0114] S205, compensating the satellite to be compensated according to the clock error compensation value of the satellite to be compensated in each epoch.

[0115] Specifically, after calculating the compensation value, the clock error compensation value of the corresponding epoch can be subtracted from the clock error estimate of each epoch, which can solve the jump problem caused by the reference change of the satellite to be compensated and make the calculated clock error product continuous.

[0116] Exemplarily, for a single satellite to be compensated, the clock error compensation value of each epoch within the predicted four epochs is obtained, and the clock error estimation values ​​of the corresponding four epochs can be obtained by real-time observation. For each epoch, the clock error estimation value is subtracted from the clock error compensation value, thereby completing the compensation of the satellite to be compensated in the corresponding epoch.

[0117] Specifically, compensation calculations may be performed on multiple satellites to be compensated, which is not limited in the embodiments of the present application.

[0118] Next, combine with Figure 3The trained satellite clock error prediction model provided in the embodiment of the present application is explained. The satellite clock error prediction model provided in the embodiment of the present application includes a normalization layer, a data filling and masking layer, a convolutional layer and a dimension exchange layer, an attention mechanism layer, a time prediction linear layer, a dimension exchange and mapping layer, and a denormalization layer in sequence.

[0119] Specifically, the satellite clock error prediction model in the embodiment of the present application adopts a linear structure, that is, an encoder structure, which is different from the encoder + decoder structure of the Transformer model in the related art. This structure directly maps the original input to the prediction result, omitting the intermediate segmented mapping step. This design makes the model more concise and efficient, effectively reduces the loss of information and the accumulation of errors, and reduces the complexity of training and reasoning.

[0120] It should be noted that the Transformer model of the related technology is difficult to meet the real-time and high-precision requirements of the satellite clock error prediction task in practical applications. Compared with the Transformer model, the satellite clock error prediction model provided in the embodiment of the present application effectively reduces the memory and time overhead through this structure.

[0121] Specifically, in order to improve the prediction ability of non-stationary data, the model double normalizes the data in the time dimension and variable dimension at the time of input, and denormalizes it at the time of output. At the same time, the model also exchanges the time and variable dimensions of the input data, so that the deep learning network first learns the relationship between variables and then predicts the time dimension, thereby significantly improving the ability of multivariate prediction.

[0122] Specifically, the model uses two linear layers, namely the time prediction linear layer and the mapping layer. These two linear layers replace the feedforward neural network in the traditional Transformer model, which helps to greatly reduce the complexity of the model.

[0123] Specifically, the satellite clock error prediction model provided in the embodiments of the present application can effectively overcome the problem of poor prediction accuracy of the Transformer model in the related art for non-stationary time series data.

[0124] In some embodiments, the process of obtaining a clock error prediction value by predicting a satellite clock error prediction model may include the following steps:

[0125] S301, taking the estimated clock error values ​​of all satellites in multiple epochs as input values, the input values ​​are normalized through a normalization layer.

[0126] Specifically, first, when processing the variable dimension, a normalization operation is implemented to ensure that the input data has a consistent scale and range, so that the model can better understand and process the relationship between different variables; secondly, normalization is also performed on the time dimension, so that the model can better adapt to the changing characteristics of data in different time periods, thereby enhancing the model's predictive ability for non-stationary time series data.

[0127] It should be noted that this double normalization method can effectively reduce the complexity of the data, making it easier and more accurate for the satellite clock error prediction model to capture the patterns and trends in the data. That is, the common clock error benchmark of multiple satellites before the clock error benchmark changes can respond more robustly to data changes.

[0128] It should be noted that this innovative data processing method provides strong support for improving the performance of the model, enabling the satellite clock error prediction model to achieve better results when processing non-stationary time series data, and can improve the accuracy and reliability of the prediction.

[0129] S302, input the normalized input value into the convolution layer of the trained satellite clock error prediction model, and increase the variable dimension through the convolution layer; perform a dimension exchange operation to exchange the variable dimension and time dimension of the input value after the variable dimension is increased, and obtain a feature vector through feature extraction.

[0130] Specifically, by further increasing the variable dimension of the data and extracting richer local features of the variables, more information can be provided for the satellite clock error prediction model.

[0131] Specifically, in order to make more effective use of satellite clock error data, especially to make it easier for the satellite clock error prediction model to obtain the common benchmark information between different satellite clock error variables, by transforming the data dimension and exchanging variables and time dimensions, it is beneficial to improve the satellite clock error prediction model's ability to grasp the relationship between variables in the data, and to improve the model's ability to capture and understand data features, so that the satellite clock error prediction model first predicts the relationship between different variables, that is, the common benchmark information between different satellite clock error variables, and then focuses on the characteristics of the time dimension.

[0132] In some embodiments, before S302, the input variables may be adjusted to adapt to different actual conditions of the number of satellites observed in different time periods.

[0133] Specifically, a fixed length of the data can be set, for example, to 30. When the number of variables in the input data is less than 30, the samples with less than 30 variables can be supplemented to 30 data, and a mask can be added to the additional data so that the supplemented data does not affect the subsequent training.

[0134] Optionally, the maximum value of the number of variables in the input data is generally taken as the corresponding fixed length, and the length of all input data is adjusted to the fixed length.

[0135] Specifically, the mask operation can be used to control the model so that the model determines whether to process a specific part of the input sequence or ignore the specific part of the input sequence according to the mask.

[0136] Optionally, the mask operation uses a mask matrix whose dimension is the same as the dimension of the input sequence, and whose elements are used to indicate whether the input at the corresponding position should be operated on.

[0137] Further, the convolution layer may increase the variable dimension based on the adjusted variable in S302.

[0138] S303, input the feature vector into the attention mechanism layer of the trained satellite clock error prediction model, and extract the clock error time series features through the attention mechanism layer.

[0139] Specifically, the main operation of the attention mechanism layer is to copy the input feature variable X three times to form a query vector Q, a key vector K, and a numerical vector V.

[0140] Further, Q, K, and V are input into the self-attention mechanism layer

[0141] Further, the attention result Xattn is obtained, and Softmax is the activation function. Xattn continues to enter the next layer of network for training.

[0142] Furthermore, the clock error time series features are extracted through the attention mechanism layer.

[0143] It should be noted that due to the short length of the input data and the data to be predicted, it is difficult for the model in the related art to predict data in a short period of time based on the shorter input data. The satellite clock error prediction model provided in the embodiment of the present application focuses on the model's learning ability for trend items, and uses two linear layers to replace the feedforward neural network of the Transformer model in the related art.

[0144] S304, input the clock error time series features into the first linear layer of the trained satellite clock error prediction model, predict the change trend of the clock error prediction value based on the clock error time series features, and output the prediction result.

[0145] Specifically, the first linear layer, namely the time prediction linear layer, is used to predict the time dimension. This linear layer can predict future data by learning the time trend in the data.

[0146] S305, input the prediction result into the second linear layer of the trained satellite clock error prediction model, exchange the time dimension and variable dimension of the prediction result through the second linear layer, input it into the denormalization layer of the satellite clock error prediction model, restore the corresponding data through the denormalization layer, and output the clock error prediction value of multiple epochs after the clock error benchmark changes.

[0147] Specifically, the second linear layer is a mapping layer, through which the variable dimension can be restored, and the data can be remapped back to the variable dimension before being input into the convolutional layer in S302.

[0148] It can be understood that by using the first linear layer and the second linear layer (i.e., the time prediction linear layer and the mapping layer), it is not only possible to better adapt to the situation of shorter data length, but also possible to more accurately capture the trend information in the data and obtain the common benchmark information between different satellite clock error data.

[0149] Finally, the denormalization layer can restore the data to the same range and distribution as before normalization in S301, and then output the clock error prediction value with the same range and distribution.

[0150] The following are device embodiments of the present application, which can be used to execute the method embodiments of the present application. For details not disclosed in the device embodiments of the present application, please refer to the method embodiments of the present application.

[0151] See next Figure 4 , is a structural diagram of a real-time satellite clock error estimation reference jump compensation device based on deep learning provided by an exemplary embodiment of the present application. The device can be implemented as all or part of a terminal through software, hardware, or a combination of both, and can also be integrated on a server as an independent module. The real-time satellite clock error estimation reference jump compensation device based on deep learning in the embodiment of the present application can be applied to a terminal or a cloud. The device 40 includes an observation module 410, a clock error prediction module 420, and a clock error compensation module 430, wherein:

[0152] The observation module 410 is used to collect clock error estimates of all observed satellites in real time;

[0153] The clock error prediction module 420 is used to input the clock error estimation values ​​of multiple epochs before the clock error estimation values ​​of all satellites change into the trained satellite clock error prediction model when the observation module 410 determines that the clock error reference of the satellite clock changes according to the clock error estimation values ​​of all satellites, and predict the clock error prediction values ​​of multiple epochs after the clock error reference changes by the trained satellite clock error prediction model;

[0154] The clock error compensation module 430 is used to calculate the clock error compensation value of the satellite to be compensated in each epoch according to the clock error estimation value of the satellite to be compensated in each epoch and the clock error prediction value of the corresponding epoch;

[0155] The clock error compensation module 430 is further used to compensate the satellite to be compensated according to the clock error compensation value of the satellite to be compensated in each epoch.

[0156] It should be noted that when the device 40 provided in the above embodiment executes the real-time satellite clock error estimation reference jump compensation method based on deep learning, only the division of the above-mentioned functional modules is used as an example. In practical applications, the above-mentioned functions can be assigned to different functional modules as needed, that is, the internal structure of the device is divided into different functional modules to complete all or part of the functions described above. In addition, the device provided in the above embodiment and the embodiment of the real-time satellite clock error estimation reference jump compensation method based on deep learning belong to the same concept. The implementation process is detailed in the method embodiment, which will not be repeated here.

[0157] An embodiment of the present application also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the steps of any of the above-mentioned method embodiments when executing the program.

[0158] See also Figure 5 , which is a structural block diagram of an electronic device provided in an embodiment of the present application.

[0159] like Figure 5 As shown, the electronic device 500 includes: a processor 501 and a memory 502 .

[0160] In the embodiment of the present application, the processor 501 is the control center of the computer system, which can be a processor of a physical machine or a processor of a virtual machine. The processor 501 may include one or more processing cores, such as a 4-core processor, an 8-core processor, etc. The processor 501 may be implemented in at least one hardware form of DSP (Digital Signal Processing), FPGA (Field-Programmable Gate Array), and PLA (Programmable Logic Array).

[0161] The processor 501 may also include a main processor and a coprocessor. The main processor is a processor for processing data in the awake state, also called a CPU (Central Processing Unit); the coprocessor is a low-power processor for processing data in the standby state.

[0162] The memory 502 may include one or more computer-readable storage media, which may be non-transitory. The memory 502 may also include a high-speed random access memory, and a non-volatile memory, such as one or more disk storage devices, flash memory storage devices. In some embodiments of the present application, the non-transitory computer-readable storage medium in the memory 502 is used to store at least one instruction, which is used to be executed by the processor 501 to implement the method in the embodiment of the present application.

[0163] In some embodiments, the electronic device 500 further includes: a peripheral device interface 503 and at least one peripheral device. The processor 501, the memory 502 and the peripheral device interface 503 can be connected via a bus or a signal line. Each peripheral device can be connected to the peripheral device interface 503 via a bus, a signal line or a circuit board. Specifically, the peripheral devices include: a display screen 504, a camera 505 and an audio circuit 506. The peripheral device interface 503 can be used to connect at least one peripheral device related to I / O (Input / Output) to the processor 501 and the memory 502.

[0164] In some embodiments of the present application, the processor 501, the memory 502, and the peripheral device interface 503 are integrated on the same chip or circuit board; in some other embodiments of the present application, any one or two of the processor 501, the memory 502, and the peripheral device interface 503 can be implemented on a separate chip or circuit board. This embodiment of the present application does not specifically limit this.

[0165] The display screen 504 is used to display a UI. The UI may include graphics, text, icons, videos, and any combination thereof. When the display screen 504 is a touch display screen, the display screen 504 also has the ability to collect touch signals on the surface or above the surface of the display screen 504. The touch signal may be input as a control signal to the processor 501 for processing. At this time, the display screen 504 may also be used to provide virtual buttons and / or virtual keyboards, also known as soft buttons and / or soft keyboards.

[0166] In some embodiments of the present application, the display screen 504 may be one, which is arranged on the front panel of the electronic device 500; in other embodiments of the present application, the display screen 504 may be at least two, which are arranged on different surfaces of the electronic device 500 or are folded; in still other embodiments of the present application, the display screen 504 may be a flexible display screen, which is arranged on the curved surface or folded surface of the electronic device 500. Even more, the display screen 504 may be arranged in a non-rectangular irregular shape, that is, a special-shaped screen. The display screen 504 may be made of materials such as LCD (Liquid Crystal Display), OLED (Organic Light-Emitting Diode) and the like.

[0167] The camera 505 is used to capture images or videos. Optionally, the camera 505 includes a front camera and a rear camera. Usually, the front camera is arranged on the front panel of the electronic device, and the rear camera is arranged on the back of the electronic device. In some embodiments, there are at least two rear cameras, which are any one of a main camera, a depth of field camera, a wide-angle camera, and a telephoto camera, so as to realize the fusion of the main camera and the depth of field camera to realize the background blur function, the fusion of the main camera and the wide-angle camera to realize the panoramic shooting and VR (Virtual Reality) shooting function or other fusion shooting functions. In some embodiments of the present application, the camera 505 may also include a flash. The flash can be a monochrome temperature flash or a dual-color temperature flash. The dual-color temperature flash refers to a combination of a warm light flash and a cold light flash, which can be used for light compensation at different color temperatures.

[0168] The audio circuit 506 may include a microphone and a speaker. The microphone is used to collect sound waves from the user and the environment, and convert the sound waves into electrical signals and input them into the processor 501 for processing. For the purpose of stereo sound collection or reduction, there may be multiple microphones, which are respectively arranged at different parts of the electronic device 500. The microphone may also be an array microphone or an omnidirectional collection microphone.

[0169] The power supply 507 is used to power various components in the electronic device 500. The power supply 507 can be an alternating current, a direct current, a disposable battery, or a rechargeable battery. When the power supply 507 includes a rechargeable battery, the rechargeable battery can be a wired rechargeable battery or a wireless rechargeable battery. A wired rechargeable battery is a battery charged through a wired line, and a wireless rechargeable battery is a battery charged through a wireless coil. The rechargeable battery can also be used to support fast charging technology.

[0170] The electronic device structure block diagram shown in the embodiment of the present application does not constitute a limitation on the electronic device 500. The electronic device 500 may include more or fewer components than shown in the figure, or combine certain components, or adopt a different component arrangement.

[0171] The embodiment of the present application also provides a computer-readable storage medium, on which a computer program is stored, and when the program is executed by a processor, the steps of the method of any of the above embodiments are implemented. The computer-readable storage medium may include, but is not limited to, any type of disk, including a floppy disk, an optical disk, a DVD, a CD-ROM, a micro drive, and a magneto-optical disk, a ROM, a RAM, an EPROM, an EEPROM, a DRAM, a VRAM, a flash memory device, a magnetic card or an optical card, a nanosystem (including a molecular memory IC), or any type of medium or device suitable for storing instructions and / or data.

[0172] Through the description of the above implementation methods, those skilled in the art can clearly understand that each implementation method can be implemented by means of software plus a necessary general hardware platform, and of course, by hardware. Based on this understanding, the above technical solution can be essentially or in other words, the part that contributes to the relevant technology can be embodied in the form of a software product, which can be stored in a computer-readable storage medium, such as ROM / RAM, a disk, an optical disk, etc., including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or some parts of the embodiment.

[0173] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit it. Although the present application has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. A real-time satellite clock error estimation reference jump compensation method based on deep learning, characterized in that: include: Collect clock error estimates of all observed satellites in real time; If it is determined that the clock error reference of the satellite clock has changed according to the clock error estimation values ​​of all the satellites, the clock error estimation values ​​of a plurality of epochs before the clock error reference has changed are input into the trained satellite clock error prediction model; Predicting the clock error prediction value of each epoch after the clock error reference changes by the trained satellite clock error prediction model; The clock error compensation value of the satellite to be compensated in each epoch is calculated according to the clock error estimation value of the satellite to be compensated in each epoch and the clock error prediction value of the corresponding epoch; Compensating the satellite to be compensated according to the clock error compensation value of the satellite to be compensated at each epoch; The step of inputting the clock error estimation values ​​of a plurality of epochs before the clock error reference changes into a trained satellite clock error prediction model, and predicting the clock error prediction value of each epoch after the clock error reference changes by the trained satellite clock error prediction model comprises: Taking the estimated clock error values ​​of all satellites in the plurality of epochs as input values, normalizing the input values; Inputting the normalized input value into the convolution layer of the trained satellite clock error prediction model, and increasing the variable dimension through the convolution layer; Performing a dimension exchange operation to exchange the variable dimension and the time dimension of the input value after the variable dimension is increased, and obtaining a feature vector by feature extraction; Inputting the feature vector into the attention mechanism layer of the trained satellite clock error prediction model, and extracting the clock error time series features through the attention mechanism layer; Input the clock error time series features into the first linear layer of the trained satellite clock error prediction model, predict the change trend of the clock error prediction value based on the clock error time series features, and output the prediction result; Input the prediction result into the second linear layer of the trained satellite clock error prediction model, and after exchanging the time dimension and the variable dimension of the prediction result through the second linear layer, input the prediction result into the denormalization layer of the satellite clock error prediction model; Recovering the corresponding data through the denormalization layer, and outputting the clock error prediction values ​​of multiple epochs after the clock error reference changes; The method further comprises: taking the estimated clock error values ​​of all satellites in the plurality of epochs as input values, and performing normalization processing on the input values; If the number of variables in the normalized input value is less than a preset threshold, data supplementation is performed on the normalized input value; The input value after data supplementation is input into the convolution layer of the satellite clock error prediction model, and the variable dimension is increased through the convolution layer.

2. The method according to claim 1, characterized in that The trained satellite clock error prediction model is trained based on the following steps, including: A training set and a validation set are established based on the clock error estimates of multiple satellites with the same clock error reference in the same time span; Inputting the training set into the satellite clock error prediction model for training; Inputting the verification set into the satellite clock error prediction model, and calculating the loss function of multiple epochs according to the output result of the satellite clock error prediction model; If the value of the loss function increases for a preset number of consecutive times, the training is completed and the trained satellite clock error prediction model is output.

3. The method according to claim 2, characterized in that A test set is established based on the clock error estimates of multiple satellites with the same clock error reference in the same time span; After outputting the trained satellite clock error prediction model, the method further includes: The test set is input into the trained satellite clock error prediction model, and the prediction accuracy of the trained satellite clock error prediction model for the clock error prediction value of each epoch is calculated.

4. The method according to claim 1 or 3, characterized in that: The step of calculating the clock error compensation value of the satellite to be compensated at each epoch according to the clock error estimation value of the satellite to be compensated at each epoch and the clock error prediction value of the corresponding epoch respectively comprises: For the first epoch after the clock error reference changes, the clock error compensation value of the first epoch is calculated based on the difference between the clock error estimate value of the satellite to be compensated in the first epoch and the clock error prediction value of the corresponding epoch; For the nth epoch after the clock error reference changes, calculate the difference between the clock error estimate of the satellite to be compensated at the nth epoch and the clock error prediction value of the corresponding epoch, use the prediction accuracy of the trained satellite clock error prediction model at the nth epoch as the weight of the difference between the clock error estimate and the clock error prediction value, and obtain the clock error compensation value of the nth epoch by weighted calculation; Here, n is an integer greater than 1.

5. The method according to claim 4, characterized in that The step of calculating the clock error compensation value of the satellite to be compensated at each epoch according to the clock error estimation value of the satellite to be compensated at each epoch and the clock error prediction value of the corresponding epoch respectively comprises: Calculate and obtain the clock error compensation value of the satellite to be compensated at the nth epoch, and use the clock error compensation value of the nth epoch as the clock error compensation value of each epoch after the nth epoch; Here, n is an integer greater than or equal to 4.

6. A real-time satellite clock error estimation reference jump compensation device based on deep learning, characterized in that: include: The observation module is used to collect the clock error estimates of all observed satellites in real time; The clock error prediction module is used to input the clock error estimation values ​​of multiple epochs before the clock error benchmark changes into the trained satellite clock error prediction model when the observation module determines that the clock error benchmark of the satellite clock changes according to the clock error estimation values ​​of all satellites, and predict the clock error prediction values ​​of multiple epochs after the clock error benchmark changes by the trained satellite clock error prediction model, specifically including: The clock error prediction module is further used to use the clock error estimation values ​​of all satellites in the multiple epochs as input values ​​and perform normalization processing on the input values; Inputting the normalized input value into the convolution layer of the trained satellite clock error prediction model, and increasing the variable dimension through the convolution layer; Performing a dimension exchange operation to exchange the variable dimension and the time dimension of the input value after the variable dimension is increased, and obtaining a feature vector by feature extraction; Inputting the feature vector into the attention mechanism layer of the trained satellite clock error prediction model, and extracting the clock error time series features through the attention mechanism layer; Input the clock error time series features into the first linear layer of the trained satellite clock error prediction model, predict the change trend of the clock error prediction value based on the clock error time series features, and output the prediction result; Input the prediction result into the second linear layer of the trained satellite clock error prediction model, and after exchanging the time dimension and the variable dimension of the prediction result through the second linear layer, input the prediction result into the denormalization layer of the satellite clock error prediction model; Recovering the corresponding data through the denormalization layer, and outputting the clock error prediction values ​​of multiple epochs after the clock error reference changes; The clock error prediction module is further configured to use the clock error estimation values ​​of all satellites in the plurality of epochs as input values, and after normalizing the input values, further comprising: If the number of variables in the normalized input value is less than a preset threshold, data supplementation is performed on the normalized input value; Inputting the data-supplemented input value into the convolution layer of the satellite clock error prediction model, and increasing the variable dimension through the convolution layer; A clock error compensation module is used to calculate the clock error compensation value of the satellite to be compensated in each epoch according to the clock error estimation value of the satellite to be compensated in each epoch and the clock error prediction value of the corresponding epoch; The clock error compensation module is also used to compensate the satellite to be compensated according to the clock error compensation value of the satellite to be compensated in each epoch.

7. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the program, the steps of the method according to any one of claims 1 to 5 are implemented.

8. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 5 are implemented.

Citation Information

Patent Citations

  • Satellite clock error reference smoothing method and system

    CN112987043A

  • Multi-GNSS satellite clock error modeling forecasting method based on LSTM and quadratic polynomial combination model

    CN117406253A

  • Satellite clock error compensation method and system

    CN117452450A

  • Method and apparatus in positioning without broadcast ephemeris

    US20080270026A1

  • Method and apparatus of correcting clock drift error

    US20120313815A1