Satellite clock error prediction method and device based on deep learning, equipment and medium

Abstract

The embodiment of the invention discloses a satellite clock error prediction method and device based on deep learning, equipment and a medium, and relates to the technical field of satellite clock error prediction.The method comprises the steps that satellite clock error data before SSR correction information interruption is called, the satellite clock error data is preprocessed and then input into a trained satellite clock error long-time prediction model, and a satellite clock error prediction result is obtained; the model predicts to obtain a first predicted clock error feature; when the IGU product is available, obtaining a second predicted clock error feature based on the IGU product; and carrying out weighted calculation on the first and second predicted clock error features based on an adaptive weight training layer, outputting a third predicted clock error feature, and processing to obtain a clock error prediction result. And if the IGU product is unavailable, obtaining a clock error prediction result based on the first prediction clock error feature. According to the invention, clock correction compensation can be provided in real time through time sequence feature extraction and prediction of the model when the SSR correction information is interrupted, so that a GNSS system is ensured to still have high time synchronization precision in a communication interruption environment.

Description

Technical Field

[0001] The present application relates to the technical field of satellite clock error prediction, and in particular to a satellite clock error prediction method, device, equipment and medium based on deep learning. Background Art

[0002] The accuracy of the Global Navigation Satellite System (GNSS) is easily affected by factors such as satellite clock error, orbit error, and atmospheric delay. In order to reduce the impact of these error sources on positioning accuracy, the error sources of satellite signals can be corrected in real time through the state space representation (SSR) correction technology to achieve the high-precision positioning requirements required for precise point positioning (PPP). However, in the case of unstable communication links, such as in high-dynamic environments such as low-orbit satellites moving at high speed, or in remote areas with underdeveloped communication facilities, SSR correction information may be interrupted during some periods of time, resulting in users being unable to obtain continuous error correction information, thereby reducing the accuracy of time synchronization.

[0003] At present, when SSR correction information is interrupted, the ultra-fast precise ephemeris (IGU) can generally be used to quickly solve and predict clock error data. However, IGU products rely on real-time satellite observation data. In the case of communication link obstruction, data update delay or interruption, the lack of high-precision data provided by SSR correction information will greatly affect the positioning accuracy and stability of IGU products, and it is not suitable for long-term clock error prediction.

[0004] Therefore, there is currently a lack of a method that can predict the clock error and provide clock error compensation when the SSR correction information is interrupted, so as to avoid the defect of reduced time synchronization accuracy in this case. Summary of the invention

[0005] The embodiments of the present application provide a satellite clock error prediction method, device, equipment and medium based on deep learning to solve the defects of the above-mentioned related technologies. The technical solution is as follows:

[0006] In a first aspect, an embodiment of the present application provides a satellite clock error prediction method based on deep learning, comprising:

[0007] When the SSR correction information is interrupted, calling the satellite clock difference data before the SSR correction information is interrupted;

[0008] Preprocessing the satellite clock error sequence data;

[0009] Inputting the preprocessed satellite clock difference sequence data into a trained satellite clock difference long-term prediction model, so that the satellite clock difference long-term prediction model predicts a first predicted clock difference feature according to the input data;

[0010] When the IGU product is available, obtaining a second predicted clock error feature based on the predicted clock error of the IGU product;

[0011] Based on the adaptive weight training layer of the satellite clock error long-term prediction model, weighted calculation is performed on the first predicted clock error feature and the second predicted clock error feature, and a third predicted clock error feature is output;

[0012] The clock error prediction result after the SSR correction information is interrupted is obtained based on the third predicted clock error feature processing.

[0013] In an optional solution of the first aspect, the preprocessing of the satellite clock error sequence data includes:

[0014] Performing wavelet transform processing on the satellite clock difference sequence data, decomposing the time series corresponding to the satellite clock difference sequence data into a plurality of components with different frequencies, and obtaining a matrix including time domain features and frequency domain features; wherein the rows of the matrix represent different time steps, and the columns of the matrix represent different frequency components;

[0015] The step of inputting the preprocessed satellite clock error sequence data into the trained satellite clock error long-term prediction model comprises:

[0016] The matrix containing time domain features and frequency domain features obtained by the preprocessing is input into the trained satellite clock error long-term prediction model.

[0017] In an optional solution of the first aspect, the satellite clock error long-term prediction model predicts a first predicted clock error feature based on input data, including:

[0018] Normalizing the input data through the normalization layer of the satellite clock error long-term prediction model;

[0019] Using the normalized data as the input of the multi-head attention layer of the satellite clock error long-term prediction model, extracting features from the normalized data based on the multi-head attention mechanism, and outputting self-attention features;

[0020] The normalized data and the self-attention features are connected through residual connection, and the features after residual connection are used as input of the feedforward neural network layer of the satellite clock error long-term prediction model;

[0021] Obtain a first predicted clock error feature output by the feedforward neural network layer.

[0022] In an optional solution of the first aspect, obtaining a second predicted clock error feature based on the predicted clock error of the IGU product includes:

[0023] Obtain the predicted clock difference generated by the IGU product, extract features of the predicted clock difference of the IGU product through a normalization layer and a linear layer in sequence, and output the second predicted clock difference feature;

[0024] The clock error prediction result after the SSR correction information is interrupted obtained based on the third predicted clock error feature processing includes:

[0025] Using the third predicted clock error feature as the input of the mapping layer of the satellite clock error long-term prediction model, mapping the data in the high-dimensional feature space corresponding to the clock error feature back to the time series of the clock error;

[0026] The time series of the clock difference is input into the denormalization layer of the satellite clock difference long-term prediction model, and the denormalized clock difference prediction result is output.

[0027] In an optional solution of the first aspect, after the satellite clock difference long-term prediction model predicts the first predicted clock difference feature according to the input data, it further includes:

[0028] In the case where the IGU product is unavailable, the clock error prediction result after the SSR correction information is interrupted is obtained based on the first predicted clock error feature processing.

[0029] In an optional solution of the first aspect, the training process of the satellite clock error long-term prediction model includes:

[0030] Inputting the preprocessed training set into the satellite clock error long-term prediction model to be trained, so as to train the satellite clock error long-term prediction model to be trained based on the clock error features in the training set; wherein part of the training samples in the training set are composed of satellite clock error data and prediction data of IGU products;

[0031] Determine the initial weight corresponding to each feature in the satellite clock difference long-term prediction model to be trained, perform forward propagation based on the initial weight and the training set, and obtain the prediction output of the satellite clock difference long-term prediction model to be trained;

[0032] Calculating the difference between the predicted output and the true value corresponding to the training set, and based on the difference, updating the initial weights by a back propagation algorithm to obtain updated weights;

[0033] The steps of forward propagation and backward propagation are repeated based on the updated weights until the satellite clock difference long-term prediction model to be trained meets the preset convergence conditions, and the trained satellite clock difference long-term prediction model is output.

[0034] In an optional solution of the first aspect, the method further includes:

[0035] When the continuous duration of the SSR correction information interruption exceeds the preset duration, the clock error prediction result output in the previous cycle is used as the input data of the next cycle, and the process proceeds to the step of preprocessing the satellite clock error sequence data to obtain the clock error prediction result of the next cycle until the SSR correction information is restored.

[0036] In a second aspect, an embodiment of the present application further provides a satellite clock error prediction device based on deep learning, comprising:

[0037] A data acquisition unit, used for calling the satellite clock difference data before the SSR correction information is interrupted when the SSR correction information is interrupted;

[0038] A data preprocessing unit, used for preprocessing the satellite clock error sequence data;

[0039] A clock error prediction unit, used for inputting the satellite clock error sequence data preprocessed by the data preprocessing unit into a trained satellite clock error long-term prediction model, so that the satellite clock error long-term prediction model predicts a first predicted clock error feature according to the input data;

[0040] When the IGU product is available, the clock error prediction unit is further used to obtain a second predicted clock error feature based on the predicted clock error of the IGU product;

[0041] The clock error prediction unit is further used to perform weighted calculation on the first predicted clock error feature and the second predicted clock error feature based on the adaptive weight training layer of the satellite clock error long-term prediction model, and output a third predicted clock error feature;

[0042] The clock error prediction unit is further used to obtain a clock error prediction result after the SSR correction information is interrupted based on the third predicted clock error feature processing.

[0043] 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.

[0044] 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.

[0045] The beneficial effects brought about by the technical solutions provided by some embodiments of the present application include at least:

[0046] The embodiments of the present application provide a deep learning-based satellite clock error prediction method, device, equipment and medium, which can provide clock error compensation in real time through time series feature extraction and prediction of the model when SSR correction information is interrupted, thereby ensuring that the GNSS system still has high time synchronization accuracy in harsh or communication interrupted environments. The embodiments of the present application not only provide a new fault-tolerant solution for high-precision navigation systems, but also further improve the reliability and continuity of GNSS services in harsh or communication interrupted environments, which can effectively improve the user experience. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] 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.

[0048] Figure 1 It is a flowchart of a satellite clock error prediction method based on deep learning provided in an embodiment of the present application;

[0049] Figure 2 It is a flowchart of a satellite clock error prediction method based on deep learning provided in an embodiment of the present application;

[0050] Figure 3 It is a structural schematic diagram of a satellite clock error prediction device based on deep learning provided in an embodiment of the present application;

[0051] Figure 4 It is a schematic diagram of the structure of an electronic device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0052] 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.

[0053] 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.

[0054] 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.

[0055] It should be noted that the state space representation (SSR) correction technology can provide information such as the satellite's orbit error, clock correction, and signal delay correction, so that the user equipment can perform real-time correction on the error source of the satellite signal, thereby achieving the high-precision positioning requirements required for precise single-point positioning. In the SSR technology framework, the ground control center receives and processes GNSS satellite signal data through a series of base stations, calculates the satellite status and error parameters (such as satellite clock error, orbit deviation, signal delay, etc.), and transmits these correction information to the user terminal through a communication link. The user terminal uses these correction parameters to correct the errors of the received satellite signals to improve positioning accuracy.

[0056] It should be noted that the IGU product is part of the ultra-rapid precise ephemeris (IGS Ultra-Rapid) released by the International GNSS Service (IGS), which is mainly used to provide satellite clock error forecast data. The update frequency of ultra-rapid precise ephemeris is very high, usually updated several times a day, or even once an hour. This high-frequency update enables users to obtain the latest satellite orbit information in a very short time, so that they can calculate their precise position more quickly. Due to its rapid update characteristics, ultra-rapid precise ephemeris plays a significant role in improving the first positioning time and improving positioning accuracy. However, it should be noted that although the ultra-rapid precise ephemeris is updated quickly, its accuracy may not be as good as the standard precise ephemeris (usually called precise ephemeris or final ephemeris) updated daily, which is obtained after a long period of processing based on more observation data.

[0057] Therefore, when SSR correction information is missing, the positioning accuracy will decrease when the relevant technology only makes predictions based on the clock error provided by the IGU product.

[0058] The present application is described in detail below with reference to specific embodiments.

[0059] Next, combine Figure 1 , introduces a satellite clock error prediction method based on deep learning provided by the embodiment of the present application. For details, please refer to Figure 1 , Figure 1 FIG. 1 is a flow chart of a satellite clock error prediction method based on deep learning provided in an embodiment of the present application. Figure 1 As shown, the method comprises the following steps:

[0060] S101, when the SSR correction information is interrupted, calling the satellite clock difference data before the SSR correction information is interrupted;

[0061] S102, preprocessing the satellite clock error sequence data;

[0062] S103, inputting the preprocessed satellite clock difference sequence data into a trained satellite clock difference long-term prediction model, so that the satellite clock difference long-term prediction model predicts a first predicted clock difference feature according to the input data;

[0063] If the IGU product is available, execute S104:

[0064] S104, obtaining a second predicted clock difference feature based on the predicted clock difference of the IGU product;

[0065] S105, performing weighted calculation on the first predicted clock error feature and the second predicted clock error feature based on the adaptive weight training layer of the satellite clock error long-term prediction model, and outputting a third predicted clock error feature;

[0066] S106, obtaining a clock error prediction result after the SSR correction information is interrupted based on the third predicted clock error feature processing.

[0067] If the IGU product is unavailable, execute S107:

[0068] S107, obtaining a clock error prediction result after the SSR correction information is interrupted based on the first predicted clock error feature processing.

[0069] Specifically, in S101, when an interruption in the SSR correction information is detected, a response mechanism for this situation can be triggered, and the historical clock error data before the interruption of the SSR correction information can be called. The historical clock error data stored locally or on a server or other device can be called. Generally, the clock error data within a 24-hour period before the moment when the SSR correction information is interrupted can be selected as the input of the model in S103. The specific input data can be processed into a time series of satellite clock errors, that is, the above-mentioned satellite clock error data.

[0070] Specifically, in S102, since satellite clock errors usually contain non-stationary and complex periodic components, directly using the original data for modeling will be disturbed and it is difficult to capture the law of clock error changes over time. Therefore, the satellite clock error data can be preprocessed, including:

[0071] The satellite clock difference sequence data is subjected to wavelet transform processing to decompose the time series corresponding to the satellite clock difference sequence data into a plurality of components with different frequencies.

[0072] Among them, wavelet transform is a time-frequency analysis method that decomposes the time series into multiple components of different frequencies to reveal the short-term and long-term characteristics in the data.

[0073] After wavelet transform processing, a matrix containing time domain features and frequency domain features is obtained; wherein the rows of the matrix represent different time steps, and the columns of the matrix represent different frequency components;

[0074] The step of inputting the preprocessed satellite clock error sequence data into the trained satellite clock error long-term prediction model comprises:

[0075] The matrix containing time domain features and frequency domain features obtained by the preprocessing is input into the trained satellite clock error long-term prediction model.

[0076] Through such decomposition, the model can better capture the multi-level characteristics in the clock error data, especially those changes related to periodicity and frequency.

[0077] Specifically, in S103, the matrix obtained after preprocessing is input into the trained satellite clock difference long-term prediction model. The trained satellite clock difference long-term prediction model provided in the embodiment of the present application includes a normalization layer, a self-attention layer, a feedforward neural network layer (FNN), an adaptive weight training layer, a mapping layer, a denormalization layer and a linear layer.

[0078] Specifically, please refer to Figure 2 , Figure 2A flowchart of a satellite clock error prediction method based on deep learning provided in an embodiment of the present application specifically illustrates the processing flow of the trained satellite clock error long-term prediction model on input data, specifically including:

[0079] The data preprocessed by S102, that is, the matrix processed by wavelet transform, is input into the normalization layer of the trained satellite clock error long-term prediction model, and the input data is normalized by the normalization layer.

[0080] Furthermore, the normalized data is used as the input of the multi-head attention layer, and features are extracted from the normalized data based on the multi-head attention mechanism to output self-attention features. Specifically, it includes:

[0081] The Attention mechanism generates a query vector (Q), a key vector (K), and a value vector (V) by applying the formula:

[0082]

[0083] The above formula can be used to dynamically adjust the weights of different time steps or frequency components, so that the model can automatically focus on the more important parts of clock error prediction. Compared with traditional convolutional layers or recursive layers, the self-attention mechanism is more flexible and can effectively model long-term dependencies in sequences. It is particularly suitable for processing possible periodic changes in satellite clock errors. Through the Attention layer, the model can identify key time points or frequency features that affect clock error fluctuations, thereby improving the accuracy and stability of the prediction.

[0084] Furthermore, the normalized data and the self-attention features are connected through residual connection, and the features after residual connection are input into the feedforward neural network layer. The function of this layer is to combine and map the features extracted by the self-attention mechanism to generate a high-level feature representation to capture more complex clock error variation patterns, and obtain the first predicted clock error feature output by the feedforward neural network layer. This feature can be recorded as X 1 .

[0085] It should be noted that the embodiment of the present application combines the self-attention mechanism through wavelet decomposition: the model introduces wavelet transform in the preprocessing stage, decomposes the time series of satellite clock error data into components of different frequencies, and can effectively extract multi-level frequency domain features in the clock error data, so that the model can better capture periodic and non-stationary changes. At the same time, the multi-frequency subsequence after wavelet decomposition is combined with the self-attention mechanism, so that the model can selectively focus on the key features of different frequencies and time points, thereby greatly improving the modeling ability of complex clock error changes and improving the computational efficiency of the model.

[0086] Next, in order to improve the accuracy of the prediction, the availability of the IGU product can be monitored in real time. If the IGU product is available, the satellite clock error long-term prediction model provided in the embodiment of the present application will input the prediction data of the IGU product as an additional constraint condition, and further optimize the prediction accuracy by combining the latest data of the IGU product; the prediction data of the IGU product can effectively compensate for the deviation that may be caused by historical data, and improve the sensitivity and accuracy of the model to future clock error changes.

[0087] If the IGU product is available, S104 is executed to extract features from the prediction data output by the IGU product, such as Figure 1-2 As shown, the process specifically includes:

[0088] The predicted clock error generated by the IGU product is obtained, and the predicted clock error of the IGU product is subjected to feature extraction through the normalization layer and the linear layer in sequence, and the second predicted clock error feature is output. The second predicted clock error feature can be recorded as X 2 .

[0089] Further, the step S105 is executed to synchronously input the first predicted clock difference feature and the second predicted clock difference feature into the adaptive weight training layer of the satellite clock difference long-term prediction model, and weights are respectively assigned to the first predicted clock difference feature and the second predicted clock difference feature through the adaptive weight training layer, and weighted calculation is performed, and the formula is applied:

[0090] X=α*X 1 +β*X 2 ;

[0091] After weighting, the third predicted clock difference feature is output, which can be recorded as X.

[0092] Among them, α and β are back-propagation matrices, which can be used to adjust the weights of the first predicted clock error feature and the second predicted clock error feature.

[0093] Among them, the prediction clock error of the IGU product is equivalent to a constraint on the prediction results output by the model, which can be used to improve the reliability and accuracy of the prediction.

[0094] Further, the step S106 is performed, such as Figure 2 As shown, the third predicted clock error feature is used as the input of the mapping layer of the satellite clock error long-term prediction model, and the data in the high-dimensional feature space corresponding to the clock error feature is mapped back to the time series of the clock error;

[0095] The time series of the clock difference is input into the denormalization layer of the satellite clock difference long-term prediction model, and the denormalized clock difference prediction result is output to ensure the consistency of data between input and output, so that the output clock difference prediction data is interpretable and easy to use.

[0096] In the event that IGU products are not available, such as Figure 1 As shown, execute S107, process only the satellite clock error data obtained in S101, and perform independent prediction of the clock error according to the first predicted clock error feature extracted based on the satellite clock error data before the SSR correction information is interrupted in S103, thereby obtaining a prediction result.

[0097] The execution process of S107 is similar to that of S106. Figure 2 As shown, the first predicted clock error feature is used as the input of the mapping layer of the satellite clock error long-term prediction model, and is further input into the denormalization layer to obtain the clock error prediction result.

[0098] It should be noted that the unavailability of the IGU product can be understood as the data output by the IGU product cannot be obtained or the IGU product has not updated the data for the time period that needs to be predicted. For example, the server of the IGU product cannot be connected, or the signal is poor, and the predicted data of the IGU product cannot be obtained.

[0099] The embodiment of the present application uses a multi-input mode design and combines the IGU product for dynamic adjustment. The model adjusts the prediction strategy according to the availability of the IGU product (the ultra-fast product of the international GNSS service) during the prediction stage. If the IGU product is available, the model will integrate the IGU data into the prediction process. The IGU product data provides additional constraints and adjustments for the prediction of satellite clock errors, which can further improve the prediction performance of the model, especially in the absence of SSR correction information, to ensure the reliability and stability of the prediction results; if the IGU product is not available, the model uses historical data and the Attention mechanism to predict. This dynamic strategy ensures that the model can still stably and accurately output clock error prediction results when data is missing or conditions change.

[0100] In some embodiments, the clock error prediction results generated by the model can cover a preset duration after the SSR correction information is interrupted, for example, it can be set to 24 hours, and the output clock error prediction results can be a time series of clock error data, which can be set to clock error information at intervals of every 5 minutes.

[0101] In some embodiments, when the continuous duration of the SSR correction information interruption exceeds a preset duration and the SSR correction information is still not restored, the clock error prediction result output in the previous cycle can be used as the input data for the next cycle, and the process proceeds to the step of preprocessing the satellite clock error sequence data as described in S102 to obtain the clock error prediction result for the next cycle, that is, steps S102-S106 or steps S102, S103, and S107 are repeated until the SSR correction information is restored.

[0102] In some embodiments, before training the satellite clock error long-term prediction model, the steps include:

[0103] The clock error data of different satellites in different time periods and IGU products are collected to form a training set.

[0104] The acquired data is organized into input-output pairs. The input data can be a time series of clock error data for the past period (such as the satellite clock error for the previous 24 hours), and the output is the clock error for the next 24 hours. During the training process, the model makes predictions based on the input data, and the model is trained by comparing the predicted values ​​with the output values ​​(true values) in the training set.

[0105] When generating a training set, it is necessary to remove outliers or missing values ​​to ensure the continuity and accuracy of the time series.

[0106] Furthermore, all acquired data can be divided into training set, validation set and test set to ensure the generalization performance of the model. At the same time, several sets can be randomly selected from the training set, validation set and test set to add the prediction data of IGU products. This ensures the extensiveness and universality of model training.

[0107] In some embodiments, the training process of the satellite clock error long-term prediction model includes:

[0108] Inputting the preprocessed training set into the satellite clock error long-term prediction model to be trained, so as to train the satellite clock error long-term prediction model to be trained based on the clock error features in the training set; wherein part of the training samples in the training set are composed of satellite clock error data and prediction data of IGU products;

[0109] Determine the initial weight corresponding to each feature in the satellite clock difference long-term prediction model to be trained, perform forward propagation based on the initial weight and the training set, and obtain the prediction output of the satellite clock difference long-term prediction model to be trained;

[0110] The difference between the predicted output and the true value corresponding to the training set is calculated, and based on the difference, the initial weight is updated by a back propagation algorithm to obtain an updated weight.

[0111] In this process, the model can calculate the loss through the back-propagation algorithm and use the Adam optimizer to update the model parameters. Specifically, the optimizer Adam can be used to update the parameters. The Adam optimization algorithm combines momentum and adaptive learning rate, which can effectively speed up the convergence speed and reduce the possibility of falling into the local optimum. Through Adam's gradient descent and back-propagation algorithm, the model parameters are gradually optimized to approach the global optimum.

[0112] The steps of forward propagation and backward propagation are repeated based on the updated weights until the satellite clock difference long-term prediction model to be trained meets the preset convergence conditions, and the trained satellite clock difference long-term prediction model is output.

[0113] In the training process, mean square error (MSE) can be used as a loss function to measure the error between the predicted value and the actual satellite clock difference. MSE can effectively reflect the degree of deviation between the predicted value and the actual value and is an important indicator for optimizing model parameters.

[0114] In some embodiments, hyperparameters such as learning rate, batch size, and number of training cycles can be set during the model training process to achieve the best convergence effect. On this basis, data enhancement and regularization methods (Dropout) are combined to prevent model overfitting; and an early stopping mechanism is used to stop training in advance when the validation set loss no longer decreases, thereby avoiding underfitting or overfitting and improving the generalization performance of the model.

[0115] In some embodiments, during the training process of the model, the validation set can be used to monitor the model performance to determine whether the training can be completed. If the loss of the validation set increases for 5 consecutive epochs, it is considered that the training of the model has reached convergence and overfitting has been avoided. The final performance of the model can also be evaluated through the test set. By recording the prediction accuracy of different epochs and analyzing the error change trend of the model, after comparing the test results with the actual clock error data, additional weights can be given to the difference between the estimated and predicted values ​​of the satellite clock error calculated next, so as to further optimize the reliability of the clock error prediction.

[0116] 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.

[0117] See next Figure 3 , is a schematic diagram of the structure of a satellite clock error prediction 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. A satellite clock error prediction device based on deep learning in an embodiment of the present application can be applied to a terminal or a cloud. The device 30 includes a data acquisition unit 301, a data preprocessing unit 302, and a clock error prediction unit 303, wherein:

[0118] The data acquisition unit 301 is used to call the satellite clock difference data before the SSR correction information is interrupted when the SSR correction information is interrupted;

[0119] The data preprocessing unit 302 is used to preprocess the satellite clock error sequence data;

[0120] The clock error prediction unit 303 is used to input the satellite clock error sequence data preprocessed by the data preprocessing unit into the trained satellite clock error long-term prediction model, so that the satellite clock error long-term prediction model predicts the first predicted clock error feature according to the input data;

[0121] In the case where the IGU product is available, the clock error prediction unit 303 is further configured to obtain a second predicted clock error feature based on the predicted clock error of the IGU product;

[0122] The clock error prediction unit 303 is further used to perform weighted calculation on the first predicted clock error feature and the second predicted clock error feature based on the adaptive weight training layer of the satellite clock error long-term prediction model, and output a third predicted clock error feature;

[0123] The clock error prediction unit 303 is further used to obtain the clock error prediction result after the SSR correction information is interrupted based on the third predicted clock error feature processing.

[0124] It should be noted that when the device 30 provided in the above embodiment executes a satellite clock error prediction method based on deep learning, only the division of the above functional modules is used as an example. In practical applications, the above 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 an embodiment of a satellite clock error prediction method based on deep learning belong to the same concept. The implementation process is detailed in the method embodiment and will not be repeated here.

[0125] 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.

[0126] See also Figure 4 , which is a structural block diagram of an electronic device provided in an embodiment of the present application.

[0127] like Figure 4 As shown, the electronic device 400 includes: a processor 401 and a memory 402 .

[0128] In the embodiment of the present application, the processor 401 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 401 may include one or more processing cores, such as a 4-core processor, an 8-core processor, etc. The processor 401 may be implemented in at least one hardware form of DSP (Digital Signal Processing), FPGA (Field-Programmable Gate Array), and PLA (Programmable Logic Array).

[0129] The processor 401 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.

[0130] The memory 402 may include one or more computer-readable storage media, which may be non-transitory. The memory 402 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 402 is used to store at least one instruction, which is used to be executed by the processor 401 to implement the method in the embodiment of the present application.

[0131] In some embodiments, the electronic device 400 further includes: a peripheral device interface 403 and at least one peripheral device 404. The processor 401, the memory 402 and the peripheral device interface 403 can be connected via a bus or a signal line. Each peripheral device 404 can be connected to the peripheral device interface 403 via a bus, a signal line or a circuit board. Specifically, the peripheral device 404 includes: a display screen, a camera and an audio circuit. The peripheral device interface 403 can be used to connect at least one peripheral device related to I / O (Input / Output) to the processor 401 and the memory 402.

[0132] In some embodiments of the present application, the processor 401, the memory 402, and the peripheral device interface 403 are integrated on the same chip or circuit board; in some other embodiments of the present application, any one or two of the processor 401, the memory 402, and the peripheral device interface 403 can be implemented on a separate chip or circuit board. This embodiment of the present application does not specifically limit this.

[0133] The electronic device structure block diagram shown in the embodiment of the present application does not constitute a limitation on the electronic device 400. The electronic device 400 may include more or fewer components than shown in the figure, or combine certain components, or adopt a different component arrangement.

[0134] 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.

[0135] 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.

[0136] 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 satellite clock error prediction method based on deep learning, characterized in that: include: When the SSR correction information is interrupted, calling the satellite clock difference data before the SSR correction information is interrupted; Preprocessing the satellite clock error sequence data; Inputting the preprocessed satellite clock difference sequence data into a trained satellite clock difference long-term prediction model, so that the satellite clock difference long-term prediction model predicts a first predicted clock difference feature according to the input data; When the IGU product is available, obtaining a second predicted clock error feature based on the predicted clock error of the IGU product; Based on the adaptive weight training layer of the satellite clock error long-term prediction model, weighted calculation is performed on the first predicted clock error feature and the second predicted clock error feature, and a third predicted clock error feature is output; The clock error prediction result after the SSR correction information is interrupted is obtained based on the third predicted clock error feature processing.

2. The satellite clock error prediction method based on deep learning according to claim 1, characterized in that: The preprocessing of the satellite clock error sequence data comprises: Performing wavelet transform processing on the satellite clock difference sequence data, decomposing the time series corresponding to the satellite clock difference sequence data into a plurality of components with different frequencies, and obtaining a matrix including time domain features and frequency domain features; wherein the rows of the matrix represent different time steps, and the columns of the matrix represent different frequency components; The step of inputting the preprocessed satellite clock error sequence data into the trained satellite clock error long-term prediction model comprises: The matrix containing time domain features and frequency domain features obtained by the preprocessing is input into the trained satellite clock error long-term prediction model.

3. The satellite clock error prediction method based on deep learning according to claim 1, characterized in that: The satellite clock error long-term prediction model predicts a first predicted clock error feature based on input data, including: Normalizing the input data through the normalization layer of the satellite clock difference long-term prediction model; Using the normalized data as the input of the multi-head attention layer of the satellite clock error long-term prediction model, extracting features from the normalized data based on the multi-head attention mechanism, and outputting self-attention features; The normalized data and the self-attention features are connected through residual connection, and the features after residual connection are used as input of the feedforward neural network layer of the satellite clock error long-term prediction model; Obtain a first predicted clock error feature output by the feedforward neural network layer.

4. The satellite clock error prediction method based on deep learning according to claim 3, characterized in that: The second predicted clock error feature is obtained based on the predicted clock error of the IGU product, including: Obtain the predicted clock difference generated by the IGU product, extract features of the predicted clock difference of the IGU product through a normalization layer and a linear layer in sequence, and output the second predicted clock difference feature; The clock error prediction result after the SSR correction information is interrupted obtained based on the third predicted clock error feature processing includes: Using the third predicted clock error feature as the input of the mapping layer of the satellite clock error long-term prediction model, mapping the data in the high-dimensional feature space corresponding to the clock error feature back to the time series of the clock error; The time series of the clock difference is input into the denormalization layer of the satellite clock difference long-term prediction model, and the denormalized clock difference prediction result is output.

5. The satellite clock error prediction method based on deep learning according to claim 4, characterized in that: After the satellite clock error long-term prediction model predicts the first predicted clock error feature according to the input data, it also includes: In the case where the IGU product is unavailable, the clock error prediction result after the SSR correction information is interrupted is obtained based on the first predicted clock error feature processing.

6. The satellite clock error prediction method based on deep learning according to claim 1, characterized in that: The training process of the satellite clock error long-term prediction model includes: Inputting the preprocessed training set into the satellite clock error long-term prediction model to be trained, so as to train the satellite clock error long-term prediction model to be trained based on the clock error features in the training set; wherein part of the training samples in the training set are composed of satellite clock error data and prediction data of IGU products; Determine the initial weight corresponding to each feature in the satellite clock difference long-term prediction model to be trained, perform forward propagation based on the initial weight and the training set, and obtain the prediction output of the satellite clock difference long-term prediction model to be trained; Calculating the difference between the predicted output and the true value corresponding to the training set, and based on the difference, updating the initial weights by a back propagation algorithm to obtain updated weights; The steps of forward propagation and backward propagation are repeated based on the updated weights until the satellite clock difference long-term prediction model to be trained meets the preset convergence conditions, and the trained satellite clock difference long-term prediction model is output.

7. A satellite clock error prediction method based on deep learning according to any one of claims 1 to 6, characterized in that: The method further comprises: When the continuous duration of the SSR correction information interruption exceeds the preset duration, the clock error prediction result output in the previous cycle is used as the input data of the next cycle, and the process proceeds to the step of preprocessing the satellite clock error sequence data to obtain the clock error prediction result of the next cycle until the SSR correction information is restored.

8. A satellite clock error prediction device based on deep learning, characterized in that: include: A data acquisition unit, used for calling the satellite clock difference data before the SSR correction information is interrupted when the SSR correction information is interrupted; A data preprocessing unit, used for preprocessing the satellite clock error sequence data; A clock error prediction unit, used for inputting the satellite clock error sequence data preprocessed by the data preprocessing unit into a trained satellite clock error long-term prediction model, so that the satellite clock error long-term prediction model predicts a first predicted clock error feature according to the input data; When the IGU product is available, the clock error prediction unit is further used to obtain a second predicted clock error feature based on the predicted clock error of the IGU product; The clock error prediction unit is further used to perform weighted calculation on the first predicted clock error feature and the second predicted clock error feature based on the adaptive weight training layer of the satellite clock error long-term prediction model, and output a third predicted clock error feature; The clock error prediction unit is further used to obtain a clock error prediction result after the SSR correction information is interrupted based on the third predicted clock error feature processing.

9. 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 steps of the method according to any one of claims 1 to 7 are implemented.

10. 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 7 are implemented.

Images