Clock error prediction method and device of two-channel neural network

Abstract

The invention relates to a clock error prediction method and device of a two-channel neural network. The method comprises the following steps: carrying out differential processing on satellite clock error data, and calculating a gross error generation moment according to a median robust estimation method and a primary differential result; after the gross error is positioned by utilizing the generation moment of the gross error, removing the gross error at the corresponding position, complementing missing data by utilizing a cubic spline interpolation method, sampling complemented clock error data by adopting a preset sampling interval and sequence length, and segmenting a sequence by utilizing the sampled data; carrying out normalization on the segmented feature sequence; a dual-channel neural network model is constructed, and a preset training set is used for training; inputting the normalized feature sequence into the trained dual-channel neural network model, and outputting a predicted value; carrying out reverse normalization on the predicted value; and recovering the clock error according to the reverse normalization predicted value to obtain a clock error predicted value. By adopting the method, error accumulation can be reduced, and the prediction precision can be improved.

Description

Technical Field

[0001] The present application relates to the field of satellite navigation technology, and in particular to a clock error prediction method and device using a dual-channel neural network. Background Art

[0002] Atomic clocks are important payloads of satellite navigation systems and are used to maintain the system time of satellite navigation systems. In order to achieve high-precision positioning and timing, it is necessary to provide users with accurate clock error prediction models, correct satellite clock errors, and obtain system time benchmarks.

[0003] At present, through the extraction of satellite clock error characteristics and noise modeling, relevant scholars have established a variety of efficient clock error prediction models. Commonly used models include linear models (LP), quadratic polynomial models (QP), gray models (GM), etc., which can achieve a certain prediction accuracy. However, the above models all model the linear trend of the clock error series, while the satellite-borne atomic clock also exhibits nonlinear fluctuations due to environmental influences, and this feature cannot be accurately described. In order to further improve the prediction accuracy, a series of clock error prediction models based on deep learning have been proposed, including extreme learning machines, RBF neural networks, BP neural networks, LSTM neural networks and other methods, which have achieved performance improvements to a certain extent, but there will be error accumulation in the long-term prediction of clock errors, resulting in performance degradation. In order to further improve the network prediction performance and reduce the error accumulation of multi-step predictions, it is necessary to propose a new clock error prediction network architecture and method. Summary of the invention

[0004] Based on this, it is necessary to provide a clock error prediction method and device of a dual-channel neural network that can reduce error accumulation and improve prediction accuracy in response to the above technical problems.

[0005] A dual-channel neural network clock error prediction method, the method comprising: Obtain satellite clock error data; perform differential processing on the satellite clock error data to obtain a primary differential result; The time of occurrence of the gross error is calculated based on the median robust estimation method and the first difference result; the gross error is located at the time of occurrence of the gross error, and the gross error at the corresponding position is removed, and the missing data is supplemented by the cubic spline interpolation method to obtain the supplemented clock error data; The completed clock error data is sampled using a preset sampling interval and sequence length, and the sequence is segmented using the sampled data; the segmented feature sequence is normalized to obtain a normalized feature sequence; A dual-channel neural network model is constructed and trained using a pre-set training set; the dual-channel neural network model includes two channels, one channel is composed of LSTM units, which is used to extract the temporal features of the sequence, and the other channel is composed of a one-dimensional convolutional layer, which is used to extract the spatial features of the sequence. The outputs of the two channels are added and sent to a three-layer fully connected layer to fuse the temporal and spatial features; The normalized feature sequence is input into the trained dual-channel neural network model to output the predicted value; the predicted value is denormalized to obtain the denormalized predicted value; the clock error is restored according to the denormalized predicted value to obtain the clock error predicted value.

[0006] A dual-channel neural network clock error prediction device, characterized in that the device comprises: The clock error completion module is used to obtain satellite clock error data; perform differential processing on the satellite clock error data to obtain a first-order differential result; calculate the time when the gross error occurs based on the median robust estimation method and the first-order differential result; locate the gross error at the time when the gross error occurs and then remove the gross error at the corresponding position, and use the cubic spline interpolation method to complete the missing data to obtain the completed clock error data; The sampling and sequence segmentation module is used to sample the completed clock error data using a preset sampling interval and sequence length, and use the sampled data to sample the sequence; normalize the segmented feature sequence to obtain a normalized feature sequence; Construct a dual-channel neural network model module, which is used to construct a dual-channel neural network model and train it using a pre-set training set; the dual-channel neural network model includes two channels, one channel is composed of LSTM units, which is used to extract the temporal features of the sequence, and the other channel is composed of a one-dimensional convolutional layer, which is used to extract the spatial features of the sequence. The outputs of the two channels are added and sent to a three-layer fully connected layer to fuse the temporal and spatial features; The clock error prediction module is used to input the normalized feature sequence into the trained dual-channel neural network model and output the predicted value; perform denormalization on the predicted value to obtain the denormalized predicted value; restore the clock error according to the denormalized predicted value to obtain the clock error predicted value.

[0007] The above-mentioned dual-channel neural network clock error prediction method and device first obtains and differentially processes the clock error data to remove the trend component; uses the median robust estimation method to detect and remove gross errors, and fills in missing data through the cubic spline interpolation method to ensure the integrity and continuity of the data. Then, the data is sampled and divided into small sequences using a preset sampling interval and sequence length, and normalized to eliminate the non-stationarity and dimensional differences of the data. Subsequently, a dual-channel neural network is constructed, in which the LSTM channel extracts time features and the convolution channel extracts spatial features, and the information of the two is fused through a fully connected layer, ultimately achieving high-precision clock error prediction, effectively reducing the noise accumulation problem, and fully exploring the nonlinear laws in the clock error sequence, thereby significantly improving the prediction accuracy. This application eliminates extreme outliers in the data by removing gross errors, reduces interference with the subsequent modeling process, uses interpolation to ensure the continuity and smoothness of the data, avoids the negative impact of missing data on model training, improves the quality of the data, and makes subsequent model training more stable, thereby reducing error accumulation and improving the accuracy of clock error prediction. At the same time, data sampling, segmentation and normalization improve the efficiency of training and the generalization ability of the model. Through local training, the model can better adapt to data from different satellites or different time periods, thereby improving the overall prediction accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0008] Figure 1 A schematic diagram of a flow chart of a clock error prediction method using a dual-channel neural network in one embodiment; Figure 2 A schematic diagram of the network structure of a dual-channel neural network model in one embodiment; Figure 3 A basic schematic diagram of a data sampling process in one embodiment; Figure 4 It is a schematic diagram of clock error prediction residuals of the present application on the G02 satellite in another embodiment, and is compared with the residuals of models such as the linear polynomial (LP), the quadratic polynomial (QP), the grey model (GM), the autoregressive moving average (ARMA), the random forest (RF), and the long short-term memory network (LSTM); Figure 5 It is a diagram showing the clock error prediction residual of the present application on the G07 satellite in an embodiment, and comparing it with the residuals of the first-order polynomial (LP), second-order polynomial (QP), gray model (GM), autoregressive moving average (ARMA), random forest (RF), long short-term memory network (LSTM) and other models; Figure 6 Schematic diagram of a clock error prediction device of a dual-channel neural network in one embodiment. DETAILED DESCRIPTION

[0009] In order to make the purpose, technical solution and advantages of the present application more clearly understood, the present application is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0010] In one embodiment, Figure 1 As shown, a clock error prediction method using a dual-channel neural network is provided, comprising the following steps: Step 102, obtain satellite clock error data; perform differential processing on the satellite clock error data to obtain a first-order differential result; calculate the time when the gross error occurs based on the median robust estimation method and the first-order differential result; locate the gross error using the time when the gross error occurs, remove the gross error at the corresponding position, and use the cubic spline interpolation method to fill in the missing data to obtain the filled clock error data.

[0011] Collect satellite clock error data to provide basic data for subsequent modeling. Perform differential processing on the satellite clock error data. Through a single differential, the linear component is weakened and the nonlinear characteristics are exposed, which is conducive to the subsequent gross error removal. The satellite clock error data is differentially processed and the single differential result is:

[0012] in, is the first difference result at time t, is the clock difference at time t.

[0013] The time when the gross error occurs is calculated based on the median robust estimation method and the result of the first difference. The median robust estimation method is used to remove the gross error, which can ensure that there is no abnormal data in the data set that affects the overall trend. This is crucial for the stability and consistency of the data, especially when processing time series data, where outliers often cause prediction errors and model instability. By locating the time when the gross error occurs and removing these outliers, the quality of the data is significantly improved. If the missing or outliers in the data are not removed, it may lead to the accumulation of errors when training the model, thereby affecting the accuracy of the prediction. Using the cubic spline interpolation method to fill in the missing data can smoothly restore the data, avoiding the large fluctuations or mutations that may be caused by simple interpolation methods. The value process not only maintains the continuity of the data, but also avoids the errors or non-smoothness that may occur during the interpolation process. The processed data is more in line with the actual change law, which is conducive to improving the prediction accuracy of the model. At the same time, in the multi-step prediction process of the time series, any small prediction error may gradually increase with the increase in the number of steps, resulting in performance degradation (i.e., the error accumulation problem). Gross error removal and interpolation completion methods can effectively ensure the data quality at each moment, thereby reducing the propagation and accumulation of initial errors. By ensuring that the input data at each step is strictly preprocessed (such as gross error removal, difference, interpolation repair), the error in the model training process is controlled at a low level, and the prediction error will not increase over time due to defects in the data itself. This is especially important for long-term predictions, because error accumulation will significantly affect the accuracy of the prediction results.

[0014] Step 104, sampling the completed clock error data using a preset sampling interval and sequence length, and using the sampled data sampling sequence; normalizing the segmented feature sequence to obtain a normalized feature sequence.

[0015] The clock error data is sampled regularly through the preset sampling interval and sequence length to ensure that each sampling point contains enough information within the time period. In this way, each sampling point represents the key features in the clock error sequence as much as possible without being affected by noise or extreme fluctuations in the data. The setting of the sampling interval and sequence length can effectively control the resolution of the data. Appropriate sampling intervals can avoid redundant information caused by oversampling, while ensuring that key information in the data is retained and reducing the impact of noise. When the data sequence is too long, the model may encounter training difficulties or inaccurate predictions due to long-term dependence or error accumulation. By dividing the data into multiple subsequences of fixed length, short time series are more likely to capture local regularities and reduce the gradual accumulation of errors.

[0016] At the same time, data sampling and segmentation effectively reduce the size of input data, allowing the model to be trained more efficiently. Compared with processing overly long original sequences, the segmented subsequences are easier to process in batches, reducing the computational overhead during training.

[0017] Atomic clocks have frequency drift, and data after a single difference often exhibits certain non-stationarity. In order to improve the generalization ability of the model and its ability to learn sequence features, normalizing the feature sequence helps reduce the impact of these non-stationary features on the model, allowing the model to process all features equally. When dealing with tasks that contain multiple features, normalization can ensure that each feature contributes evenly to the prediction results, which helps improve the generalization ability of the model, especially when dealing with data across satellites or time periods.

[0018] Step 106, construct a dual-channel neural network model and train it using a preset training set; the dual-channel neural network model includes two channels, one channel is composed of LSTM units, which is used to extract the temporal features of the sequence, and the other channel is composed of a one-dimensional convolutional layer, which is used to extract the spatial features of the sequence. The outputs of the two channels are added and sent to a three-layer fully connected layer for fusing the temporal and spatial features.

[0019] like Figure 2 The figure shows the network structure of the dual-channel neural network model. One channel is composed of LSTM units, which is used to extract the time features of the sequence. The other channel is composed of a one-dimensional convolutional layer, which is used to extract the spatial features of the sequence. The outputs of the two channels are added and sent to the three-layer fully connected layer to fuse the time and space features, so as to more accurately predict the sequence. The segmented training set data is used for training to obtain the optimal parameters of the network. Modeling based on time or space features alone may not fully characterize the complexity of the clock error series. The dual-channel network combines the advantages of time series modeling and local feature extraction, and can more fully extract the nonlinear laws in the clock error data and improve the prediction accuracy.

[0020] Step 108, input the normalized feature sequence into the trained dual-channel neural network model and output the predicted value; perform denormalization on the predicted value to obtain the denormalized predicted value; restore the clock error according to the denormalized predicted value to obtain the clock error predicted value.

[0021] The above-mentioned dual-channel neural network clock error prediction method first obtains and differentially processes the clock error data to remove the trend component; uses the median robust estimation method to detect and remove gross errors, and uses the cubic spline interpolation method to fill in the missing data to ensure the integrity and continuity of the data. Then, the data is sampled and divided into small sequences using the preset sampling interval and sequence length, and normalized to eliminate the non-stationarity and dimensional differences of the data. Subsequently, a dual-channel neural network is constructed, in which the LSTM channel extracts time features and the convolution channel extracts spatial features, and the information of the two is fused through the fully connected layer, ultimately achieving high-precision clock error prediction, effectively reducing the noise accumulation problem, and fully exploring the nonlinear laws in the clock error sequence, thereby significantly improving the prediction accuracy. This application eliminates extreme outliers in the data by removing gross errors, reduces interference with the subsequent modeling process, uses interpolation to ensure the continuity and smoothness of the data, avoids the negative impact of missing data on model training, improves the quality of the data, and makes subsequent model training more stable, thereby reducing error accumulation and improving the accuracy of clock error prediction. At the same time, data sampling, segmentation and normalization improve the efficiency of training and the generalization ability of the model. Through local training, the model can better adapt to data from different satellites or different time periods, thereby improving the overall prediction accuracy.

[0022] In one embodiment, calculating the time of occurrence of a gross error based on a median robust estimation method and a primary difference result includes: According to the median robust estimation method and the first difference result, the time when the gross error occurs is calculated as

[0023] in, for The middle number of is the result of a first difference at time t.

[0024] In one embodiment, locating the gross error by using the time when the gross error occurs includes: The clock error discrimination threshold is designed based on the time when the gross error occurs. ,when , the current differential result is determined to be a gross error.

[0025] In one embodiment, the sampled data is

[0026] in, The matrix is Matrix, front As a feature, Columns as data labels, represents the sequence length, represents the sampling interval, Indicates the data length, Indicates primary differential data.

[0027] In one embodiment, using the sampled data to segment the sequence includes: No. i The feature sequence is expressed as:

[0028] in, represents the length of the feature sequence, Indicates the length of the sequence.

[0029] In one embodiment, the segmented feature sequence is normalized to obtain a normalized feature sequence, further comprising: Normalize the segmented feature sequence, and get the normalized feature sequence as

[0030] in, Indicates i A sequence of features, represents the minimum value in the sequence. Represents the maximum value in a sequence.

[0031] In one embodiment, the normalized feature sequence is input into a trained dual-channel neural network model, and a prediction value is output, including: When the length of the normalized feature sequence of the input Less than sampling interval When points as the predicted value output; when the length of the sequence to be predicted is Greater than sampling interval When / And round it up to get the number of network iteration predictions.

[0032] In one embodiment, performing denormalization on the predicted value to obtain the denormalized predicted value includes: The predicted value is denormalized to obtain the denormalized predicted value

[0033] in, It is i The denormalized prediction value, Is the network i Unnormalized predicted values, represents the minimum value in the sequence. Represents the minimum value in a sequence.

[0034] In one embodiment, restoring the clock error according to the denormalized predicted value to obtain the clock error predicted value includes: Recover the clock error based on the denormalized predicted value and obtain the predicted clock error value as

[0035] in, To start predicting the clock difference at the previous moment, It is i The denormalized prediction value.

[0036] In a specific embodiment, Figure 3 As shown in the figure, the process of constructing a data set based on sampling can significantly improve the accuracy of long-term predictions of neural networks and reduce the error accumulation caused by a large number of iterative predictions.

[0037] like Figure 4 , Figure 5 The figure shows the 24h prediction residual curve drawn by using the designed network to conduct prediction experiments on GPS G02 and G07 satellites. In the experiment, p was set to 36 and L was set to 32, that is, sampling was performed every 36 points, and the length of the characteristic sequence was 32. Compared with classic algorithms such as LP, QP, GM, ARMA, RF, and LSTM, the dual-channel neural network has the smallest prediction residual, which illustrates the effectiveness of this application.

[0038] It should be understood that although Figure 1 The steps in the flowchart are shown in sequence as indicated by the arrows, but these steps are not necessarily executed in the order indicated by the arrows. Unless otherwise specified in this document, there is no strict order restriction for the execution of these steps, and these steps can be executed in other orders. Moreover, Figure 1 At least part of the steps may include multiple sub-steps or multiple stages. These sub-steps or stages are not necessarily executed at the same time, but can be executed at different times. The execution order of these sub-steps or stages is not necessarily sequential, but can be executed in turn or alternately with other steps or at least part of the sub-steps or stages of other steps.

[0039] In one embodiment, Figure 6 As shown, a dual-channel neural network clock error prediction device is provided, including: a clock error completion module 602, a sampling and sequence segmentation module 604, a dual-channel neural network model construction module 606 and a clock error prediction module 608; The clock error completion module 602 is used to obtain satellite clock error data; perform differential processing on the satellite clock error data to obtain a primary differential result; calculate the time when the gross error occurs according to the median robust estimation method and the primary differential result; locate the gross error at the time when the gross error occurs, remove the gross error at the corresponding position, and use the cubic spline interpolation method to complete the missing data to obtain the completed clock error data; The sampling and sequence segmentation module 604 is used to sample the completed clock error data using a preset sampling interval and sequence length, and use the sampled data to sample the sequence; normalize the segmented feature sequence to obtain a normalized feature sequence; A dual-channel neural network model building module 606 is used to build a dual-channel neural network model and train it using a preset training set; the dual-channel neural network model includes two channels, one channel is composed of LSTM units, which is used to extract the time features of the sequence, and the other channel is composed of a one-dimensional convolutional layer, which is used to extract the spatial features of the sequence. The outputs of the two channels are added and sent to a three-layer fully connected layer for fusing the time and space features; The clock error prediction module 608 is used to input the normalized feature sequence into the trained dual-channel neural network model and output the predicted value; perform denormalization on the predicted value to obtain the denormalized predicted value; restore the clock error according to the denormalized predicted value to obtain the clock error predicted value.

[0040] For the specific definition of a clock error prediction device of a dual-channel neural network, please refer to the definition of a clock error prediction method of a dual-channel neural network above, which will not be repeated here. Each module in the above-mentioned clock error prediction device of a dual-channel neural network can be implemented in whole or in part by software, hardware and a combination thereof. The above-mentioned modules can be embedded in or independent of the processor in the computer device in the form of hardware, or can be stored in the memory of the computer device in the form of software, so that the processor can call and execute the operations corresponding to the above modules.

[0041] Those of ordinary skill in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be implemented by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media used in the embodiments provided in this application may include non-volatile and / or volatile memory. Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory may include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in many forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.

[0042] The technical features of the above embodiments may be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0043] The above-mentioned embodiments only express several implementation methods of the present application, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the invention patent. It should be pointed out that, for a person of ordinary skill in the art, several variations and improvements can be made without departing from the concept of the present application, and these all belong to the protection scope of the present application. Therefore, the protection scope of the patent of the present application shall be subject to the attached claims.

Claims

1. A dual-channel neural network clock error prediction method, characterized in that: The method comprises: Acquire satellite clock error data; perform differential processing on the satellite clock error data to obtain a primary differential result; The time of occurrence of the gross error is calculated according to the median robust estimation method and the first difference result; the gross error is located at the time of occurrence of the gross error, the gross error at the corresponding position is removed, and the missing data is supplemented by the cubic spline interpolation method to obtain the supplemented clock error data; The completed clock error data is sampled using a preset sampling interval and sequence length, and the sequence is segmented using the sampled data; the segmented feature sequence is normalized to obtain a normalized feature sequence; A dual-channel neural network model is constructed and trained using a pre-set training set; the dual-channel neural network model includes two channels, one channel is composed of LSTM units, which is used to extract the temporal features of the sequence, and the other channel is composed of a one-dimensional convolutional layer, which is used to extract the spatial features of the sequence. The outputs of the two channels are added and sent to a three-layer fully connected layer for fusing the temporal and spatial features; The normalized feature sequence is input into a trained dual-channel neural network model to output a predicted value; the predicted value is denormalized to obtain a denormalized predicted value; the clock error is restored according to the denormalized predicted value to obtain a clock error predicted value.

2. The method according to claim 1, characterized in that The time when the gross error occurs is calculated according to the median robust estimation method and the first difference result, including: According to the median robust estimation method and the first difference result, the time when the gross error occurs is calculated as in, for The middle number of is the result of a first difference at time t.

3. The method according to claim 1, characterized in that Use the time when the gross error occurs to locate the gross error, including: The clock error discrimination threshold is designed based on the time when the gross error occurs. ,when , the current differential result is determined to be a gross error.

4. The method according to claim 1, characterized in that: The sampled data is in, The matrix is Matrix, front As a feature, Columns as data labels, represents the sequence length, represents the sampling interval, Indicates the data length, Indicates primary differential data.

5. The method according to claim 1, characterized in that Use the sampled data to split the sequence, including: No. i The feature sequence is expressed as: in, represents the length of the feature sequence, Indicates the length of the sequence.

6. The method according to claim 1, characterized in that Normalizing the segmented feature sequence to obtain a normalized feature sequence also includes: Normalize the segmented feature sequence, and get the normalized feature sequence as in, Indicates i A sequence of features, represents the minimum value in the sequence. Represents the maximum value in a sequence.

7. The method according to claim 1, characterized in that The normalized feature sequence is input into the trained dual-channel neural network model, and the predicted value is output, including: When the length of the normalized feature sequence of the input Less than sampling interval When points as the predicted value output; when the length of the sequence to be predicted is Greater than sampling interval When / And round it up to get the number of network iteration predictions.

8. The method according to claim 1, characterized in that Denormalizing the predicted value to obtain a denormalized predicted value includes: The predicted value is denormalized to obtain a denormalized predicted value: in, It is i The denormalized predicted value, Is the network i Unnormalized predicted values, represents the minimum value in the sequence. Represents the minimum value in a sequence.

9. The method according to claim 1, characterized in that The clock error is restored according to the denormalized predicted value to obtain the predicted clock error value, including: Recover the clock error based on the denormalized predicted value and obtain the predicted clock error value as in, To start predicting the clock difference at the previous moment, It is i The denormalized prediction value.

10. A dual-channel neural network clock error prediction device, characterized in that: The device comprises: A clock error completion module is used to obtain satellite clock error data; perform differential processing on the satellite clock error data to obtain a primary differential result; calculate the time when the gross error occurs according to the median robust estimation method and the primary differential result; locate the gross error at the time when the gross error occurs, remove the gross error at the corresponding position, and use the cubic spline interpolation method to complete the missing data to obtain the completed clock error data; A sampling and sequence segmentation module is used to sample the completed clock error data using a preset sampling interval and sequence length, and use the sampled data to sample the sequence; normalize the segmented feature sequence to obtain a normalized feature sequence; Construct a dual-channel neural network model module, which is used to construct a dual-channel neural network model and train it using a preset training set; the dual-channel neural network model includes two channels, one channel is composed of LSTM units, which is used to extract the time features of the sequence, and the other channel is composed of a one-dimensional convolutional layer, which is used to extract the spatial features of the sequence. The outputs of the two channels are added and sent to a three-layer fully connected layer for fusing the time and space features; The clock error prediction module is used to input the normalized feature sequence into a trained dual-channel neural network model and output a predicted value; perform denormalization on the predicted value to obtain a denormalized predicted value; and restore the clock error according to the denormalized predicted value to obtain a clock error predicted value.

Images