A satellite clock error prediction method and system

Abstract

The present invention relates to the field of satellite navigation and time synchronization technology, and in particular to a satellite clock error prediction method and system. The satellite clock error prediction model constructed by the present invention extracts multi-scale features from input data and performs sequence decomposition. The decomposed seasonal items are subjected to autocorrelation processing by an encoder and a decoder. The encoder and the decoder both perform sequence decomposition on the data. The seasonal items output by the decoder are dimensionally superimposed with the trend items after each sequence decomposition and converted into clock error prediction data. The present invention includes multiple sequence decompositions, adopts a progressive decomposition architecture to gradually remove noise interference, purifies the trend component in the seasonal item through a time convolution operation, and utilizes a residual feedback mechanism to optimize the trend accumulation process through the superposition of trend items, thereby significantly improving the purity of trend extraction, while maintaining the phase integrity of the seasonal component, so that the model can accurately model the multi-scale features of clock error data.

Description

Technical Field

[0001] The present invention relates to the technical field of satellite navigation and time synchronization, and in particular to a satellite clock error prediction method and system. Background Art

[0002] Satellite clock error prediction is a key technology in satellite navigation systems, directly impacting positioning accuracy. Traditional satellite clock error prediction methods rely primarily on offline learning, using historical data for model training. These methods are unable to adapt to the changing nature of real-time data streams. Existing methods suffer from significant deficiencies in prediction accuracy and real-time performance, making them inadequate for the positioning control needs of heavy-load train groups.

[0003] Traditional statistical models, such as ARIMA and Kalman filtering, are widely used for time series forecasting. ARIMA models can capture linear trends and seasonal variations, while Kalman filtering is suitable for state estimation of dynamic systems and can update model parameters in real time. However, these methods assume linear relationships in the data, making them difficult to handle complex nonlinear time series. Furthermore, traditional statistical models place high demands on the stationarity of the data distribution, resulting in poor forecasting performance for nonstationary data and inability to adapt to changes in real-time data streams.

[0004] Machine learning models, such as support vector machines (SVMs) and random forests, process nonlinear data using kernel functions and ensemble learning methods, making them suitable for small and medium-sized datasets. However, these methods suffer from high computational complexity when processing large amounts of data, making them difficult to meet real-time prediction requirements. Furthermore, machine learning models typically only capture local temporal dependencies and struggle to handle long-range dependencies. Furthermore, model performance relies on manually designed features, making feature engineering a complex and time-consuming process.

[0005] Deep learning models, such as LSTM and Transformer, capture long-range dependencies through gating and self-attention mechanisms, making them suitable for large-scale data. LSTM networks capture long-term dependencies in time series through gating mechanisms, while Transformer models capture global dependencies through self-attention. However, these models have high computational complexity, making them difficult to meet the needs of real-time prediction. Furthermore, existing deep learning models typically use single-scale feature extraction, making it difficult to capture both short-term fluctuations and long-term trends. Furthermore, model parameters are typically fixed and cannot be dynamically adjusted to accommodate changes in data distribution.

[0006] In summary, existing methods have significant deficiencies in prediction accuracy, computational complexity, and real-time performance, making them difficult to meet the demands of high-precision satellite positioning, such as in heavy-load train group positioning control. Therefore, a satellite clock error prediction method that can adapt to dynamic data changes is urgently needed. Summary of the Invention

[0007] In order to overcome the defects of the existing satellite clock error prediction algorithms in the above-mentioned prior art, such as high computational complexity, insufficient prediction accuracy and difficulty in adapting to real-time data streams, the present invention proposes a satellite clock error prediction method, which combines data preprocessing and prediction model optimization to efficiently and accurately predict satellite clock errors, thereby improving the accuracy and real-time performance of satellite positioning.

[0008] The present invention proposes a satellite clock error prediction method. First, a satellite clock error prediction model is constructed. Its input is satellite clock error window data that has undergone differential processing and amplification processing. Its output is clock error prediction data for the time period after the input data. The satellite clock error prediction model extracts multi-scale features from the input data and performs sequence decomposition. The decomposed seasonal terms are subjected to autocorrelation processing by an encoder and a decoder. The encoder and decoder both perform sequence decomposition on the data. The seasonal terms output by the decoder are dimensionally superimposed with the trend terms after each sequence decomposition and converted into clock error prediction data.

[0009] Then, the real-time satellite clock error data is collected and processed in a windowed manner. Differential processing and amplification processing are performed on each window data. The processed window data is input into the satellite clock error prediction model to obtain the clock error prediction data.

[0010] Preferably, the satellite clock error prediction model includes a sequentially connected data embedding module, a multi-scale feature extraction module, a sequence decomposition module, an encoder, a decoder, a dimensionality superposition unit and an output module; the multi-scale feature extraction module uses convolution kernels of different sizes to perform convolution processing on the input data and then splices them to generate multi-scale features; the multi-scale features are decomposed into seasonal terms and trend terms by the sequence decomposition module; the seasonal terms are encoded by the encoder and then decomposed again into seasonal terms and trend terms, and the seasonal terms output by the encoder are processed by the decoder and then decomposed again into seasonal terms and trend terms; the dimensionality superposition unit performs dimensionality superposition on the seasonal terms output by the decoder and the trend terms after each sequence decomposition, and then inputs and outputs the data into the output module for processing as clock error prediction data.

[0011] Preferably, the data embedding module adopts a time coding embedding method.

[0012] Preferably, the decoder includes a temporal convolutional network, a first sequence decomposition layer, an autocorrelation network, a second sequence decomposition layer, a feedforward network layer and a third sequence decomposition layer connected in sequence; each sequence decomposition layer performs sequence decomposition on the input, and the decomposed seasonal terms are propagated backward, and the seasonal terms output by the third sequence decomposition layer serve as the seasonal terms output by the decoder; the trend terms of each sequence decomposition are dimensionally superimposed to form the trend terms output by the decoder.

[0013] Preferably, the encoder comprises an autocorrelation layer, a feedforward network layer and a sequence decomposition network connected in sequence.

[0014] Preferably, the output module adopts a fully connected layer.

[0015] Preferably, the training process of the satellite clock error prediction model includes the following steps:

[0016] Collect historical clock error data and perform window processing to obtain 2m consecutive window data X1, X2, ..., Xj, ..., Xm, Y1, Y2, ..., Yj, ..., Ym; construct the data sample {Xj, Yj|1≤j≤m}; the number of steps of X and Y is equal;

[0017] Perform differential processing and amplification on Xj in the data sample {Xj, Yj|1≤j≤m}, convert Xj into preprocessed data Zj, and obtain the training sample {Zj, Yj|1≤j≤m};

[0018] Extract data Z1, Z2, ..., Zj, ..., Zm from multiple training samples and input them into the satellite clock error model to obtain predicted data Y'1, Y'2, ..., Y'j, ..., Y'm; compare the data Y1, Y2, ..., Yj, ..., Ym with the predicted data Y'1, Y'2, ..., Y'j, ..., Y'm, calculate the model loss, and use the gradient descent optimization method to update the model parameters; repeat this step until the model reaches convergence conditions.

[0019] Preferably, the model convergence condition is: the number of model iterations reaches a set value, or the absolute value of the model loss range of at least the last three rounds is less than a set threshold.

[0020] The present invention proposes a satellite clock error prediction system, which includes a data preprocessing module, a memory and a processor. The memory stores a computer program and a satellite clock error prediction model. The data preprocessing module is used to obtain satellite clock error data collected in real time and perform differential processing and amplification processing. The processor is connected to the memory and the data preprocessing module, and the processor is used to execute the computer program to implement the satellite clock error prediction method.

[0021] The present invention proposes a storage medium storing a computer program, which is used to implement the satellite clock error prediction method when executed.

[0022] The advantages of the present invention are:

[0023] (1) The present invention involves multiple sequence decompositions, adopts a progressive decomposition architecture to gradually remove noise interference, purifies the trend component in the seasonal term through a time convolution operation, and optimizes the trend accumulation process by superimposing trend terms and utilizing a residual feedback mechanism, thereby significantly improving the purity of trend extraction while maintaining the phase integrity of the seasonal component, enabling the model to accurately model the multi-scale characteristics of clock error data. Compared with traditional single decomposition methods, the present invention has stronger noise resistance and physical consistency, can adapt to the clock error prediction needs of satellites with different orbital characteristics, and performs well in terms of computational efficiency and long-term prediction stability.

[0024] (2) The present invention performs differential processing on known satellite clock error data and then amplifies it, amplifying the data from extremely small values ​​to a form that can clearly reflect the difference, thereby improving the accuracy of the prediction, avoiding the impact of dimensional differences between different features on model training, and optimizing numerical stability. Experiments have shown that the original clock error differential value is on the order of 1e-7~1e-5 nanoseconds. After amplification by 100,000 times, the effective digital utilization rate is increased from 17% to 100%, and the effective digital value can be amplified by 1e5 times to enter the optimal processing range (0.1~10.0) of a typical deep learning model, while maintaining a strict correspondence with the physical dimension. At the same time, the data after differential and amplification processing ensures high efficiency and low latency. Combined with data windowing processing, it can effectively deal with the concept drift problem in the data stream and ensure the long-term stability of the model.

[0025] (3) The present invention can directly map the real physical environment in which the atomic clock works through time coding. For example, the year and month coding reflects the frequency aging characteristics of the atomic clock of 1e-13 per year; the day and hour coding matches the temperature cycle changes caused by the rotation of the earth; and the minute and second coding captures the high-frequency noise of orbital vibration. Secondly, multi-scale feature decoupling can be performed. For example, the year and month layers model long-term drift, i.e., trend terms; the day and hour layers extract orbital periodic features, i.e., seasonal terms; and the minute and second layers process short-term jitter, i.e., residual terms. Compared with the use of position coding embedding in the original Autoformer model, the present invention uses time coding embedding, which has better physical interpretability in time series data processing, can improve the robustness during on-orbit deployment, and support the maintenance of autonomous time base on board.

[0026] (4) When decomposing the sequence before the output module, the time series convolution and sequence decomposition of the seasonal term are performed to purify the trend term and accumulate it. Finally, the purified trend term and the decomposed seasonal term are dimensionally superimposed and finally enter the output module as input for sequence prediction.

[0027] (5) The multi-scale feature extraction module extracts convolution features through convolution kernels of different sizes to achieve comprehensive capture of local details and global trends; the encoder and decoder work together to implement a lightweight Autoformer module, which uses the autocorrelation mechanism to capture long-distance dependencies, avoiding the limitations of single-scale feature extraction in existing technologies and traditional models that are difficult to handle nonlinear and non-stationary data. It can make full use of multi-scale features and long-distance dependency information, effectively improving prediction accuracy and real-time performance while maintaining low computational complexity.

[0028] (6) The satellite clock error prediction method and system proposed in this invention significantly enhances the significance of time series features through data preprocessing. It comprehensively captures the clock error variation pattern by combining local details with global trends through multi-scale modeling. By separating and purifying seasonal and trend terms, the model's ability to analyze periodic patterns is enhanced, improving the model's physical interpretability. Furthermore, the lightweight network design and dynamic window partitioning in this invention meet the needs of real-time prediction. This invention has important application value in fields such as satellite navigation and time synchronization, and particularly outperforms traditional methods in complex dynamic environments. BRIEF DESCRIPTION OF THE DRAWINGS

[0029] Figure 1 This is a module diagram of the satellite clock error prediction model proposed in the present invention;

[0030] Figure 2 for Figure 1 Diagram of the encoder architecture;

[0031] Figure 3 for Figure 1 Figure 2. Decoder architecture diagram.

[0032] Figure 4 It is the Autoformer network architecture;

[0033] Figure 5 This is a flow chart of the satellite clock error prediction method proposed in the present invention;

[0034] Figure 6 It is a line graph of the raw data of a certain satellite in the embodiment data set;

[0035] Figure 7 for Figure 6 The original data in the dataset show quasi-linear trend characteristics after outlier removal, missing value interpolation, difference and amplification;

[0036] Figure 8 This is a comparison chart of the loss function results of the three models in the embodiment;

[0037] Figure 9 This is a line chart comparing the model predicted value and the true value in the embodiment. DETAILED DESCRIPTION

[0038] The following will provide a clear and complete description of the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention without making any creative efforts shall fall within the scope of protection of the present invention.

[0039] like Figure 1 As shown, this embodiment proposes a satellite clock error prediction model, including a sequentially connected data embedding module, a multi-scale feature extraction module, a sequence decomposition module, an encoder, a decoder, a dimension superposition unit and an output module.

[0040] The model takes pre-processed satellite clock error data as input and outputs clock error prediction data. Data pre-processing includes differential processing and amplification to generate pre-processed data with significant features.

[0041] In specific implementation, the collected window data can be set as X={x1,x2,...,x i ,...,x N}, the future clock error data corresponding to X, i.e. the predicted data, is recorded as Y={y1,y2,...,y i ,...,y N}, where x i represents the i-th historical clock error data, y i Represents x i The corresponding future clock error data, 1≤i≤N, N represents the window length; that is, x1, x2, x N Respectively represent the 1st, 2nd, and Nth historical clock error data in X, y1, y2, y N Represent x1, x2, x N The corresponding future clock error data. Assuming the data x i and x i+1 The time interval is t, then y i with x i The time interval is T, T≥Nt.

[0042] The preprocessing process of window data X is as follows:

[0043] First, perform differential processing on the window data X to obtain the corresponding differential data ΔX={Δx1,Δx2,...,Δx i ,...,Δx N}, among which, Δx1, Δx2, Δx i , Δx N x1, x2, x i 、x N The corresponding difference calculation result, Δxi =x i -x i-1 ;x i-1 (i=1) represents the historical clock error data collected at time interval t before x1;

[0044] Then the differential data ΔX is amplified to obtain the amplified data Z={z1,z2,...,z i ,...,z N}, where z i =Δx i ×σ, σ is the magnification factor. Considering that the satellite clock error is in nanosecond units, in order to highlight the data trend in specific implementation, it may be necessary to magnify it to more than 10,000 times, that is, σ≥10 4 , specifically, the value may be in the range of 50,000 to 100,000, and in subsequent embodiments, the specific value is 100,000.

[0045] The satellite clock error prediction model is trained on the dataset {Z, Y}.

[0046] The data embedding module uses time-coded embedding to embed the preprocessed data Z into the set feature space and generates embedded features to facilitate the convolution processing of the multi-scale feature extraction module.

[0047] The multi-scale feature extraction module includes multiple convolutional networks with convolution kernels of different sizes. Each convolutional network performs convolution processing on the embedded features output by the data embedding module to generate convolution data of different sizes. The convolution data is then spliced ​​into multi-scale feature data through a splicing network. For example, in subsequent embodiments, the multi-scale feature extraction module uses convolution networks with three types of convolution kernels: 1×1, 3×3, and 5×5 to perform convolution processing on the features output by the data embedding module. The data embedding module embeds the data Z into the specified data space, and after convolution processing, performs dimensional splicing to form multi-scale features.

[0048] The sequence decomposition module decomposes the multi-scale feature data into seasonal terms and trend terms and outputs them. The seasonal terms are encoded by the encoder and decomposed again into seasonal terms and trend terms. The seasonal terms output by the encoder are processed by the decoder and decomposed again into seasonal terms and trend terms.

[0049] The dimensionality stacking unit performs dimension stacking on the trend term output by the sequence decomposition module, the trend term output by the encoder, the trend term output by the decoder, and the seasonal term output by the decoder. The stacked features are converted into clock error prediction data via the output module. Specifically, the output module can use an activation function or a fully connected layer.

[0050] In this embodiment, the encoder and decoder use an autocorrelation mechanism and a feedforward neural network to capture the long-range dependencies of the time series.

[0051] Reference Figure 2 The encoder includes a sequentially connected autocorrelation layer, a feedforward network layer, and a sequence decomposition network; the autocorrelation layer is used to extract the attention features of the seasonal terms of the input data, i.e., the output of the sequence decomposition module, and the number of attention heads is set to 8; the attention features are processed by the feedforward network and then further separated into seasonal terms and trend terms by the sequence decomposition network.

[0052] Reference Figure 3 The decoder consists of a sequentially connected temporal convolutional network, a first sequence decomposition layer, an autocorrelation network, a second sequence decomposition layer, a feedforward network layer, and a third sequence decomposition layer. The decoder's input data, the seasonal term output by the encoder, is processed by the temporal convolutional network and then further decomposed into a seasonal term and a trend term by the first sequence decomposition layer. The seasonal term is then passed through the autocorrelation layer to extract attention features, which are then further separated into the seasonal term and the trend term by the second sequence decomposition layer. The seasonal term output by the second sequence decomposition layer is then processed by the feedforward network layer and further decomposed into the seasonal term and the trend term by the third sequence decomposition layer.

[0053] In this embodiment, the encoder and decoder cooperate to form a lightweight Autoformer network architecture. Figure 4 The Autoformer network architecture performs multiple sequence decompositions in the decoder to purify seasonal terms and trend terms, and then accumulates the trend terms extracted by the encoder and decoder through the dimensional superposition unit, realizing the purification and superposition of seasonal terms and trend terms, making the data features more prominent and the time features more obvious. Combined with the time encoding method adopted by the data embedding module, it achieves better physical interpretability, thereby improving the accuracy of time series prediction tasks.

[0054] Obviously, in the satellite clock error prediction model, the data preprocessing module is fixed, and the part of the model to be trained includes the data embedding module, the multi-scale feature extraction module, the sequence decomposition module, the encoder, the decoder, the dimension superposition unit and the output module.

[0055] Reference Figure 5 This embodiment also proposes a satellite clock error prediction method, comprising the following steps:

[0056] St1, the real-time satellite clock error data stream is cut into multiple time windows of equal length X1, X2, ..., Xj, ..., Xm, and each time window is subjected to differential processing and amplification processing to obtain preprocessed data Z1, Z2, ..., Zj, ..., Zm; 1≤j≤m; m is the number of windows, and N is the window length;

[0057] Xj={x (j-1)N+1 ,x (j-1)N+2 ,...,x (j-1)N+i ,...,x jN};

[0058] Zj={z (j-1)N+1 ,z (j-1)N+2 ,...,z (j-1)N+i ,...,z jN};

[0059] x (j-1)N+1 、x (j-1)N+2 、x (j-1)N+i and x jN represent the 1st, 2nd, jth and Nth satellite clock error data in Xj, respectively, z (j-1)N+1 、z (j-1)N+2 、z (j-1)N+i and z jN Represents x (j-1)N+1 、x (j-1)N+2 、x (j-1)N+i and x jN The corresponding pre-processed data after differentiation and amplification; 1≤i≤N.

[0060] In this embodiment, a piece of data is pre-processed as follows Figure 6 and Figure 7 As shown, it can be seen that the data trend is more obvious through difference and amplification processing.

[0061] St2, input the pre-processed data Z1, Z2, ..., Zj, ..., Zm into the trained satellite clock error prediction model, and the model outputs the predicted data {y1, y2, ..., y k ,...,y mN}; y1, y2, y k and y mN Represent the satellite clock error data of the 1st, 2nd, kth, and mNth predictions respectively.

[0062] The satellite clock error prediction model is trained and verified in conjunction with specific embodiments below.

[0063] In this embodiment, sampling is performed from a university's data center to obtain a total of approximately 60,000 satellite clock error data, with data lengths of 15 minutes, half an hour, and one hour.

[0064] In this embodiment, a data sample is constructed for each piece of data. The specific method is as follows:

[0065] First, determine the window length N for window processing of the data, divide the data window into 2m consecutive window data X1, X2, ..., Xj, ..., Xm, Y1, Y2, ..., Yj, ..., Ym; construct the data sample {Xj, Yj|1≤j≤m};

[0066] Xj={x (j-1)N+1 ,x (j-1)N+2,...,x (j-1)N+i ,...,x jN};

[0067] Yj={y (j-1)N+1 ,y (j-1)N+2 ,...,y (j-1)N+i ,...,y jN};

[0068] 1≤j≤m.

[0069] Then, the data {Xj, 1≤j≤m} in each data sample is differentiated and amplified to obtain the processing result Zj corresponding to the data Xj, and the amplification factor is selected as 100,000;

[0070] Zj={z (j-1)N+1 ,z (j-1)N+2 ,...,z (j-1)N+i ,...,z jN};

[0071] Construct the data sample {Zj,Yj|1≤j≤m}.

[0072] In this embodiment, the data samples {Zj, Yj|1≤j≤m} are divided into a training set and a test set, and the above-mentioned satellite clock error prediction model and the given comparison models Autoformer and Transformer are trained on the training set.

[0073] The satellite clock error prediction model in this embodiment is as follows Figure 1-Figure 3 shown.

[0074] The model training process is as follows:

[0075] S1, initialize the model;

[0076] S2. Extract learning samples from the training set and input the data of the learning samples Z1, Z2, ..., Zj, ..., Zm into the satellite clock error prediction model to obtain the model prediction results Y'1, Y'2, ..., Y'j, ..., Y'm;

[0077] Y'j={y' (j-1)N+1 ,y' (j-1)N+2 ,...,y' (j-1)N+i ,...,y' jN};

[0078] y' (j-1)N+1 、y' (j-1)N+2 、y' (j-1)N+i 、y' jN They represent the (j-1)N+1, (j-1)N+2, (j-1)N+i, and jN prediction data respectively.

[0079] S3. Calculate the model loss on the learning sample and use the gradient descent optimization method to update the model parameters by minimizing the model loss. The model loss can be specifically calculated using the mean square error (MSE).

[0080] ;

[0081] {Yj={y (j-1)N+1 ,y (j-1)N+2 ,...,y (j-1)N+i ,...,y jN |1≤j≤m};

[0082] Y'j={y' (j-1)N+1 ,y' (j-1)N+2 ,...,y' (j-1)N+i ,...,y' jN |1≤j≤m};

[0083] S4. Repeat the above steps S2-S3 until the model converges.

[0084] In this embodiment, the model is set to converge after 15 iterations; in specific implementation, the model convergence condition can also be set as the convergence of the loss function, and an early stopping mechanism can be introduced.

[0085] In this embodiment, the trained model is tested on the test set to evaluate the MSE index and MAE index respectively. The test results are shown in Table 1 and Figure 8 shown.

[0086] ;

[0087] Table 1 Comparison of loss function results of three models

[0088] ;

[0089] As can be seen from Table 1, the satellite clock error prediction model proposed in this invention (hereinafter referred to as the present model) has an improvement of 4.29% in MAE and 1.69% in MSE compared to the Autoformer; compared to the Transformer, it has an improvement of 1.32% in MAE and 0.26% in MSE.

[0090] Judging from the test results, the improvement of MSE is lower than that of MAE. This may be because the data source has been subjected to outlier removal and missing value interpolation, which improves the smoothness of the data.

[0091] However, the test results have also proved that the satellite clock error prediction model proposed in the present invention is more accurate in satellite clock error prediction than the existing model.

[0092] In this embodiment, a sample is selected from the test set to further compare the prediction results of the model of the present invention with the true value. The results are shown in Table 2 and Figure 9 shown.

[0093] Table 2 List of error values ​​for predicting 16 time steps

[0094] ;

[0095] The average error over 16 time steps is 0.000424975 ns, and the error reaches the order of 10e-5, which proves the accuracy of the prediction of the present invention.

[0096] In this embodiment, an ablation experiment was also conducted. Figure 1 The decoder in the satellite clock error prediction model shown is replaced by Figure 4 In the decoder shown, the dimension superposition unit still performs dimension superposition on the trend item output by the sequence decomposition module, the trend item output by the encoder, the trend item output by the decoder, and the season item output by the decoder to obtain the ablation model.

[0097] The ablation experiment results are shown in Table 3.

[0098] Table 3 Ablation experiment results

[0099] ;

[0100] Through ablation experiments, it can be proved that the operation of multiple decomposition and purification of trend items in the present invention significantly improves the accuracy of prediction results.

[0101] In this example, the data sample {Xj, Yj|1≤j≤m} was divided into a training set and a test set. The proposed model and the ablation model were trained on the training set {Xj, Yj|1≤j≤m}. The model accuracy was then tested on the test set {Xj, Yj|1≤j≤m} to compare the impact of data preprocessing on prediction accuracy. The results are shown in Table 4.

[0102] Table 4 Dataset experimental results

[0103] ;

[0104] It can be seen that the model trained with the dataset {Zj, Yj|1≤j≤m} has smaller MAE and MSE, that is, the model performance is better, which further proves the advantage of the present invention in performing differential and amplification processing on data.

[0105] Of course, it will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, but also encompasses the same or similar structures that can be implemented in other specific forms without departing from the spirit or essential characteristics of the invention. Therefore, the embodiments should be considered in all respects as illustrative and non-restrictive, and the scope of the invention is defined by the appended claims, not the foregoing description, and it is intended that all variations that fall within the meaning and range of equivalents of the claims be encompassed within the present invention. Any reference signs in the claims should not be construed as limiting the claim to which they relate.

[0106] In addition, it should be understood that although this specification is described in terms of implementation methods, not every implementation method contains only one independent technical solution. This narrative method of the specification is only for the sake of clarity. Those skilled in the art should regard the specification as a whole. The technical solutions in each embodiment can also be appropriately combined to form other implementation methods that can be understood by those skilled in the art.

[0107] The technology, shape, and structure not described in detail in the present invention are all well-known technologies.

Claims

1. A satellite clock error prediction method, characterized in that: First, a satellite clock error prediction model is constructed. Its input is the satellite clock error window data after differential processing and amplification processing, and its output is the clock error prediction data for the period after the input data. The satellite clock error prediction model extracts multi-scale features from the input data and performs sequence decomposition. The decomposed seasonal terms are autocorrelated through the encoder and decoder. The encoder and decoder both perform sequence decomposition on the data. The seasonal terms output by the decoder are dimensionally superimposed with the trend terms after each sequence decomposition and converted into clock error prediction data. Then, the real-time satellite clock error data is collected and processed in a windowed manner. Differential processing and amplification processing are performed on each window data. The processed window data is input into the satellite clock error prediction model to obtain the clock error prediction data.

2. The satellite clock error prediction method according to claim 1, wherein: The satellite clock error prediction model consists of a sequentially connected data embedding module, a multi-scale feature extraction module, a sequence decomposition module, an encoder, a decoder, a dimension stacking unit, and an output module. The multi-scale feature extraction module uses convolution kernels of different sizes to convolve the input data and then splice them to generate multi-scale features. The multi-scale features are decomposed into seasonal items and trend items through the sequence decomposition module; the seasonal items are encoded by the encoder and then decomposed into seasonal items and trend items again, and the seasonal items output by the encoder are processed by the decoder and then decomposed into seasonal items and trend items again; the dimension superposition unit performs dimension superposition on the seasonal items output by the decoder and the trend items after each sequence decomposition, and then inputs and outputs them into the output module to process them into clock error prediction data.

3. The satellite clock error prediction method according to claim 2, wherein: The data embedding module adopts time coding embedding method.

4. The satellite clock error prediction method according to claim 2, wherein: The decoder includes a sequentially connected temporal convolutional network, a first sequence decomposition layer, an autocorrelation network, a second sequence decomposition layer, a feedforward network layer, and a third sequence decomposition layer; Each sequence decomposition layer decomposes the input sequence, and the decomposed seasonal terms are propagated backward. The seasonal terms output by the third sequence decomposition layer are used as the seasonal terms output by the decoder. The trend items of each sequence decomposition are dimensionally superimposed to form the trend items output by the decoder.

5. The satellite clock error prediction method according to claim 2, wherein: The encoder consists of sequentially connected autocorrelation layers, feedforward network layers, and sequence decomposition networks.

6. The satellite clock error prediction method according to claim 2, wherein: The output module uses a fully connected layer.

7. The satellite clock error prediction method according to any one of claims 1 to 6, wherein: The training process of the satellite clock error prediction model includes the following steps: Collect historical clock error data and perform window processing to obtain 2m consecutive window data X1, X2, ..., Xj, ..., Xm, Y1, Y2, ..., Yj, ..., Ym; construct the data sample {Xj, Yj|1≤j≤m}; the number of steps of X and Y is equal; Perform differential processing and amplification on Xj in the data sample {Xj, Yj|1≤j≤m}, convert Xj into preprocessed data Zj, and obtain the training sample {Zj, Yj|1≤j≤m}; Extract data Z1, Z2, ..., Zj, ..., Zm from multiple training samples and input them into the satellite clock error model to obtain predicted data Y'1, Y'2, ..., Y'j, ..., Y'm; compare the data Y1, Y2, ..., Yj, ..., Ym with the predicted data Y'1, Y'2, ..., Y'j, ..., Y'm, calculate the model loss, and use the gradient descent optimization method to update the model parameters; repeat this step until the model reaches convergence conditions.

8. The satellite clock error prediction method according to claim 7, wherein: The model convergence condition is: the number of model iterations reaches the set value, or the absolute value of the model loss range of at least the last three rounds is less than the set threshold.

9. A satellite clock error prediction system, characterized in that: The invention comprises a data preprocessing module, a memory and a processor, wherein the memory stores a computer program and a satellite clock error prediction model, the data preprocessing module is used to obtain satellite clock error data collected in real time and perform differential processing and amplification processing, the processor is connected to the memory and the data preprocessing module, and the processor is used to execute the computer program to implement the satellite clock error prediction method according to any one of claims 1 to 7.

10. A storage medium, characterized in that: A computer program is stored, and when the computer program is executed, it is used to implement the satellite clock error prediction method according to any one of claims 1 to 7.

Images