LS + CNN combined prediction UT1-UTC method considering effective angular momentum

Abstract

The invention discloses an LS + CNN combined prediction UT1-UTC method considering effective angular momentum, and relates to the technical field of geophysics and related computer science, and the method comprises the steps: obtaining a UT1-UTC basic data set and an effective angular momentum data set, constructing an LS model and a CNN model, combining an LS fitting extrapolation result with an output result of the CNN model, and obtaining a high-precision UT1-UTC prediction value. The model provided by the invention has relatively high precision and reliability in UT1-UTC medium-and-long-term forecasting, and compared with a traditional method, the model provided by the invention can overcome the defects of a traditional linear model on UT1-UTC parameter forecasting, and meanwhile, the forecasting precision and stability of the UT1-UTC are improved. According to the method, medium-and-long-term high-precision forecasting of the UT1-UTC is realized, and a brand-new method for forecasting the UT1-UTC parameters is provided for the fields of navigation satellite orbit determination, astronomical observation, geophysical research and the like.

Description

Technical Field

[0001] The present application relates to the fields of geophysics and related computer science and technology, and in particular to a LS+CNN combined prediction method for UT1-UTC that takes into account effective angular momentum. Background Art

[0002] UT1-UTC (the difference between Universal Time and Coordinated Universal Time) is a key indicator in the Earth Rotation Parameters (ERP), reflecting the unevenness of the Earth's rotation rate. UT1 represents astronomical time based on the Earth's rotation, while UTC is the international standard time based on atomic time. Changes in UT1-UTC directly affect the accuracy and reliability of fields such as the Global Positioning System, satellite navigation, astronomical observations, and geophysical research. For example, in satellite navigation, the accuracy of UT1-UTC directly affects the precise calculation of satellite orbits and positioning accuracy; in geophysical research, changes in UT1-UTC can be used to analyze the Earth's internal dynamic processes and the impact of climate change.

[0003] Currently, UT1-UTC forecasts mainly rely on linear models and numerical analysis methods. Linear models are usually based on historical observational data and predict future values by fitting long-term trends and cyclical changes. For example, the UT1-UTC prediction model released by the International Earth Rotation and Reference Systems Service (IERS) is mainly based on the least squares (LS) method and time series analysis methods. The Earth's rotation is affected by multiple factors, including the exchange of angular momentum in the atmosphere, ocean, mantle, and interior of the Earth. However, existing UT1-UTC prediction models often ignore the combined effects of these angular momentum exchanges. Although some studies have attempted to incorporate atmospheric angular momentum (AAM) and oceanic angular momentum (OAM) into the models, these studies generally only consider a single factor and lack a comprehensive analysis of angular momentum exchange. Summary of the Invention

[0004] The purpose of this application is to provide a LS+CNN combined forecasting method for UT1-UTC that takes into account effective angular momentum, which can fully consider the comprehensive effects of various angular momentum factors and effectively improve the accuracy of UT1-UTC forecast.

[0005] To achieve the above objectives, this application provides the following solutions:

[0006] This application provides a LS+CNN combined forecasting method for UT1-UTC that takes into account effective angular momentum, including:

[0007] Acquire a UT1-UTC basic data set and an effective angular momentum data set; the effective angular momentum data set includes: AAM data, OAM data, hydrospheric angular momentum (HAM) data, and sea-level angular momentum (SLAM) data;

[0008] Converting the UT1-UTC basic data set into an excitation domain to obtain a geodetic angular momentum sequence;

[0009] subtracting the effective angular momentum data set from the geodetic angular momentum sequence to obtain a geodetic angular momentum residual;

[0010] performing least square fitting on the geodetic angular momentum residual and the effective angular momentum data set respectively to obtain a geodetic angular momentum residual fitting sequence and an effective angular momentum fitting sequence;

[0011] Subtracting the geodetic angular momentum residual fitting sequence from the geodetic angular momentum residual to obtain a geodetic residual difference;

[0012] subtracting the effective angular momentum fitting sequence from the effective angular momentum data set to obtain an effective angular momentum residual sequence;

[0013] Extrapolating the geodetic angular momentum residual fitting sequence and the effective angular momentum fitting sequence respectively by the least square method to obtain a geodetic angular momentum residual fitting extrapolation sequence and an effective angular momentum fitting extrapolation sequence;

[0014] Inputting the effective angular momentum residual sequence and the geodetic residual difference into a trained convolutional neural network (CNN) model respectively to obtain effective angular momentum residual prediction data and geodetic residual difference prediction data;

[0015] Adding the effective angular momentum fitting extrapolation sequence to the effective angular momentum residual prediction data to obtain an effective angular momentum prediction result;

[0016] Adding the geodetic angular momentum residual fitting extrapolation sequence and the geodetic residual difference prediction data to obtain a geodetic angular momentum residual prediction result;

[0017] The effective angular momentum prediction result is added to the geodetic angular momentum residual prediction result to obtain a UT1-UTC prediction result.

[0018] Optionally, converting the UT1-UTC basic data set into an excitation domain to obtain a geodetic angular momentum sequence specifically includes:

[0019] Subtracting leap seconds from the UT1-UTC basic dataset to obtain a subtracted dataset;

[0020] Correcting the subtraction data set for the earth's solid tide to obtain a UT1R-TAI parameter sequence;

[0021] The UT1R-TAI parameter sequence is converted into the excitation domain to obtain the geodetic angular momentum sequence.

[0022] Optionally, obtaining the geodetic angular momentum residual specifically includes:

[0023] The effective angular momentum data set is subtracted from the geodetic angular momentum sequence to obtain a geodetic angular momentum residual.

[0024] Optionally, performing least squares fitting on the geodetic angular momentum residual and the effective angular momentum data set respectively to obtain a geodetic angular momentum residual fitting sequence and an effective angular momentum fitting sequence, and obtaining a geodetic angular momentum residual difference and an effective angular momentum residual sequence, specifically includes:

[0025] interpolating the effective angular momentum dataset into an interpolated dataset having the same time interval as the UT1-UTC basic dataset;

[0026] interpolating the effective angular momentum dataset into an interpolated dataset having the same time interval as the UT1-UTC basic dataset;

[0027] performing least square fitting on the geodetic angular momentum residual and the interpolation data set respectively to obtain a geodetic angular momentum residual fitting sequence and an effective angular momentum fitting sequence;

[0028] The geodetic angular momentum residual is subtracted from the geodetic angular momentum residual fitting sequence to obtain the geodetic angular momentum residual difference value. The effective angular momentum data is subtracted from the effective angular momentum fitting sequence to obtain the effective angular momentum residual sequence.

[0029] Optionally, performing least squares extrapolation on the geodetic angular momentum residual fitting sequence and the effective angular momentum fitting sequence to obtain an extrapolated sequence of the geodetic angular momentum residual fitting sequence and the effective angular momentum fitting sequence mainly includes constructing a least squares extrapolation model:

[0030]

[0031] Among them, χ(t) represents the least squares fitting result, ak represents a constant, a k+1 represents the linear term, t represents the UTC time, k is [0,7], i represents the segment number of the least squares fitting, C i and D i represents the coefficient corresponding to the periodic term, R i represents the periodic term, e i represents zero-mean white noise.

[0032] Extrapolation is performed using the above-mentioned geodetic angular momentum residual fitting sequence and effective angular momentum fitting sequence.

[0033] Optionally, predicting the geodetic angular momentum residual difference and the geodetic angular momentum residual sequence specifically includes:

[0034] The network neural network model has three convolution layers, and the size of the convolution kernel of each layer is 3×3, so as to predict the above-mentioned geodetic angular momentum residual difference and geodetic angular momentum residual sequence.

[0035] Optionally, adding the effective angular momentum prediction result to the geodetic angular momentum residual prediction result to obtain a UT1-UTC prediction result specifically includes:

[0036] Adding the effective angular momentum prediction result to the geodetic angular momentum residual prediction result to obtain a geodetic angular momentum prediction result;

[0037] converting the geodetic angular momentum prediction result into a UT1R-TAI prediction result using the Liouville equation;

[0038] The effects of leap seconds and the Earth's solid tides are added to the UT1R-TAI forecast result to obtain the final UT1-UTC forecast result.

[0039] According to the specific embodiments provided in this application, this application discloses the following technical effects:

[0040] The present application provides a LS+CNN combined forecasting method for UT1-UTC taking into account effective angular momentum, the method comprising: obtaining a UT1-UTC basic data set and an effective angular momentum data set; the effective angular momentum data set comprising: atmospheric angular momentum data, ocean angular momentum data, land water angular momentum data and sea level angular momentum data; converting the UT1-UTC basic data set into an excitation domain to obtain a geodetic angular momentum sequence; subtracting the effective angular momentum data set from the geodetic angular momentum sequence to obtain a geodetic angular momentum residual; performing least squares fitting on the geodetic angular momentum residual and the effective angular momentum data set respectively to obtain a geodetic angular momentum residual fitting sequence and an effective angular momentum fitting sequence; subtracting the geodetic angular momentum residual fitting sequence from the geodetic angular momentum residual to obtain a geodetic residual difference; and subtracting the effective angular momentum data set from the geodetic angular momentum residual to obtain a geodetic residual difference. The effective angular momentum fitting sequence is obtained to obtain an effective angular momentum residual sequence; the geodetic angular momentum residual fitting sequence and the effective angular momentum fitting sequence are extrapolated by the least squares method to obtain a geodetic angular momentum residual fitting extrapolation sequence and an effective angular momentum fitting extrapolation sequence; the effective angular momentum residual sequence and the geodetic residual difference are respectively input into a trained CNN model to obtain effective angular momentum residual prediction data and geodetic residual difference value prediction data; the effective angular momentum fitting extrapolation sequence is added to the effective angular momentum residual prediction data to obtain an effective angular momentum prediction result; the geodetic angular momentum residual fitting extrapolation sequence and the geodetic residual difference value prediction data are added to obtain a geodetic angular momentum residual prediction result; and the effective angular momentum prediction result is added to the geodetic angular momentum residual prediction result to obtain a UT1-UTC prediction result. Since the UT1-UTC basic dataset is converted to the excitation domain in this application, all influencing factors can be reflected. Moreover, the geodetic angular momentum sequence obtained by converting it to the excitation domain and the effective angular momentum dataset are on the same level. As a result, the combined effects of various angular momentum factors can be fully considered in the subsequent least squares extrapolation and convolutional neural network prediction processes, effectively improving the accuracy of the UT1-UTC forecast. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0042] Figure 1 A flowchart of a LS+CNN combined UT1-UTC forecasting method taking into account effective angular momentum provided by one embodiment of the present application;

[0043] Figure 2 A schematic diagram of the basic sequence of the UT1-UTC basic dataset from EOP 20C04 and its preprocessing process provided in one embodiment of the present application;

[0044] Figure 3 A schematic diagram of the interpolation processing results of the axial components of AAM, OAM, HAM, and SLAM and the total amount of EAM provided in one embodiment of the present application;

[0045] Figure 4 A schematic diagram of CNN model parameter training for a residual difference sequence of a geodetic angular momentum sequence provided in one embodiment of the present application;

[0046] Figure 5 A schematic diagram showing the accuracy statistics of CNN model parameter training for the residual difference sequence of the geodetic angular momentum sequence provided in one embodiment of the present application;

[0047] Figure 6 A schematic diagram of CNN model parameter training for the residual sequence of an effective angular momentum dataset provided in one embodiment of the present application;

[0048] Figure 7 A statistical diagram of the accuracy of CNN model parameter training for the residual sequence of the effective angular momentum dataset provided in one embodiment of the present application. DETAILED DESCRIPTION

[0049] The following will be combined with the drawings in the embodiments of this application to clearly and completely describe the technical solutions in the embodiments of this application. Obviously, the embodiments described are only part of the embodiments of this application, not all of the embodiments. Based on the embodiments in this application, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of this application.

[0050] The impact of land-water angular momentum (HAM) and sea-level angular momentum (SLAM) on UT1-UTC forecasts has not been fully considered in existing technologies. This incomplete consideration of angular momentum exchange limits the accuracy and reliability of UT1-UTC forecasts.

[0051] With the continuous advancement of Earth rotation observation technology, the amount and accuracy of UT1-UTC observation data have increased significantly. However, existing forecasting methods have significant shortcomings in adapting to high-precision, high-frequency observation data. For example, while the least squares method can effectively extract long-term trends and cyclical changes, it performs poorly when dealing with nonlinear characteristics and short-term fluctuations. In addition, traditional time series analysis methods (such as the ARIMA model) have difficulty accurately capturing the changing patterns of UT1-UTC when faced with the complex dynamics of Earth's rotation, resulting in large forecast errors.

[0052] In recent years, deep learning technology has made significant progress in time series forecasting, particularly in handling complex nonlinear relationships and large-scale data. Convolutional neural networks (CNNs), as powerful feature extraction tools, can automatically learn hidden patterns and regularities in input data and have been successfully applied to time series forecasting in various fields, including meteorology and finance. However, research on the application of CNN technology to UT1-UTC forecasting is still in its infancy and has not yet fully integrated the physical background of Earth's rotational dynamics.

[0053] Effective angular momentum (EAM) is a comprehensive reflection of the exchange of angular momentum between the atmosphere, ocean, and mantle within the Earth system. It reflects the influence of interactions between Earth's various spheres on its rotation rate. Research has shown that changes in atmospheric angular momentum (AAM) are primarily caused by atmospheric circulation and air pressure distribution, while changes in ocean angular momentum (OAM) are closely related to ocean circulation and tidal processes. Land-water angular momentum (HAM) and sea-level angular momentum (SLAM) are related to rheological processes within the Earth's interior. These angular momentum exchange processes significantly drive variations in UT1-UTC.

[0054] Although existing research has recognized the impact of angular momentum exchange on UT1-UTC, most studies focus solely on a single angular momentum factor (such as AAM or OAM), ignoring the combined effects of these factors. Furthermore, existing research lacks a systematic analysis and optimization of the angular momentum weighting coefficients when incorporating angular momentum exchange into UT1-UTC forecast models. This incomplete consideration of angular momentum results in existing models failing to accurately reflect the dynamics of Earth's rotation, which in turn affects the accuracy of UT1-UTC forecasts.

[0055] To overcome the shortcomings of the prior art, this application proposes a method for forecasting UT1-UTC using a LS+CNN combination that takes into account effective angular momentum. This method combines the least squares method (LS) and a convolutional neural network (CNN), and fully considers the combined effects of the atmosphere, ocean, land water, and sea level to establish a more accurate UT1-UTC forecast model. Specifically, this application incorporates the physical background of angular momentum exchange within the Earth system into the model by calculating effective angular momentum (EAM), while utilizing the nonlinear feature extraction capabilities of CNN to mine hidden patterns in the UT1-UTC time series. Through this combined method, this application aims to improve the forecast accuracy of UT1-UTC and provide a more reliable Earth rotation parameter forecast technology for applications in related fields.

[0056] This application provides a method for predicting parameters of the Earth's rotation rate (UT1-UTC) by combining the least squares (LS) method and a convolutional neural network (CNN) involving effective angular momentum, aiming to improve the forecast accuracy of UT1-UTC. This method combines the LS and CNN models and fully considers the impact of the EAM on UT1-UTC to improve the forecast accuracy and reliability of UT1-UTC parameters. The forecast method includes: obtaining UT1-UTC parameters and EAM sequences, performing data preprocessing on the UT1-UTC sequence, including subtracting leap seconds and tidal terms, to obtain the UT1R-TAI base sequence. Subsequently, an LS model and a CNN model are constructed. Finally, the LS fitting results are combined with the output of the CNN model to obtain a high-precision UT1-UTC forecast value. The LS+CNN combined model described in this application has high accuracy and reliability in medium- and long-term UT1-UTC forecasts. Compared with traditional methods, it can effectively overcome the shortcomings of traditional linear models in predicting UT1-UTC parameters and effective angular momentum, while also improving the forecast accuracy and stability of UT1-UTC. This method achieves high-precision medium- and long-term prediction of UT1-UTC, and provides a new method for predicting UT1-UTC parameters for fields such as navigation satellite orbit determination, astronomical observation, and geophysical research.

[0057] In order to make the above-mentioned purposes, features and advantages of the present application more obvious and easy to understand, the present application is further described in detail below with reference to the accompanying drawings and specific implementation methods.

[0058] In an exemplary embodiment, Figure 1 As shown, a LS+CNN combined UT1-UTC forecasting method taking into account effective angular momentum is provided, including the following steps S1 to S11.

[0059] S1. Acquire a UT1-UTC basic data set and an effective angular momentum data set; the effective angular momentum data set includes: atmospheric angular momentum data, ocean angular momentum data, land water angular momentum data, and sea level angular momentum data.

[0060] In this example, the UT1-UTC basic data set and the related effective angular momentum data set are collected and organized. The UT1-UTC sequence comes from the IERS official website, and the effective angular momentum comes from the GFZ official website. It mainly includes atmospheric angular momentum (AAM), ocean angular momentum (OAM), land water angular momentum (HAM), and sea level angular momentum (SLAM) data.

[0061] First, the effective angular momentum data of AAM, OAM, HAM and SLAM are collected, and the UT1-UTC sequence with a time interval of 1 day (UT1-UTC basic dataset) is downloaded from the International Earth Rotation and Reference System Service (IERS) website.

[0062] It is worth noting that historical data is used when training the model.

[0063] As an optional implementation, this embodiment may use the following method to collect data to collect the UT1-UTC basic dataset and the effective angular momentum dataset:

[0064] UT1-UTC basic dataset: obtained from the International Earth Rotation Service (IERS), with a data time span from January 1999 to April 2023, and a sampling interval of once a day.

[0065] Atmospheric angular momentum (AAM) data: Derived from the ERA5 global atmospheric reanalysis product, the data time span is consistent with the UT1-UTC data.

[0066] Ocean angular momentum (OAM) data: obtained from the GOCE ocean circulation model, the data time span is consistent with the UT1-UTC data.

[0067] Land water angular momentum (HAM) data and sea level angular momentum (SLAM) data: calculated through geophysical models, the data time span is consistent with the UT1-UTC basic data set. Figure 2 and Figure 3 shown.

[0068] Then, the data preprocessing steps are carried out, including deducting leap seconds from the UT1-UTC basic data set and calculating the effects of the Earth's solid tides, as well as interpolating the effective angular momentum data set. The specific steps include the following:

[0069] Step 1: Before forecasting the UT1-UTC series, it is necessary to deduct leap seconds and correct for the Earth's solid tides. Leap seconds are corrected using the value provided by the Earth's rotation center. The Earth's solid tide correction model is:

[0070]

[0071] Among them, δUT1 is the Earth solid tide correction term, B i ,C i They are given in the 7th and 8th columns of Table 8.1 of the 2010 IERS Convention Standards, α ij is the ith tide α in Table 8.1 j Integer multiples of α ij represents the basic angular coefficient of nutation, α j is the helio-lunar nutation parameter, j is listed from 1 to 5.

[0072] Step ②: Interpolate the collected effective angular momentum dataset into the same interval and time as the UT1-UTC basic dataset.

[0073] Interpolation of the effective angular momentum dataset involves interpolating the axial components of AAM, OAM, HAM, and SLAM to time intervals consistent with the UT1-UTC base dataset; and merging the interpolated angular momentum components to form a unified time series. Preprocessing of the UT1-UTC base dataset includes deducting leap seconds and correcting for Earth's solid Earth tides.

[0074] S2. Convert the UT1-UTC basic data set into the excitation domain to obtain a geodetic angular momentum sequence.

[0075] In this example, the preprocessed UT1R-TAI parameter sequence (preprocessed results of the UT1-UTC basic data set; UT1R-TAI parameters = (UT1-UTC) - leap second - Earth solid tide correction) is converted to the excitation domain to obtain a geodetic angular momentum sequence. The UT1R-TAI parameter sequence can be converted to the excitation domain using the Liouville equation, as shown below:

[0076]

[0077] UT1R is the universal time after further correcting the periodic tidal changes on the basis of UT1, TAI represents atomic time, UT1R-TAI is the sequence obtained by preprocessing UT1-UTC, Ψ is the geodetic angular momentum function, Ω is the average angular velocity of the Earth, which is 7.292115×10-5rands-1; t represents UTC time.

[0078] S3. Subtract the effective angular momentum data set from the geodetic angular momentum sequence to obtain a geodetic angular momentum residual.

[0079] S4. Performing least square fitting on the geodetic angular momentum residual and the effective angular momentum data set respectively to obtain a geodetic angular momentum residual fitting sequence and an effective angular momentum fitting sequence.

[0080] Based on the construction of the linear model: In this embodiment, the geodetic angular momentum residual and the data set after the interpolation of the effective angular momentum data set are respectively subjected to least squares fitting to obtain the geodetic angular momentum residual fitting sequence and the effective angular momentum fitting sequence, and the formula is:

[0081]

[0082] Among them, χ(t) represents the least squares fitting result, a k represents a constant, a k+1represents the linear term, t represents the UTC time, k is [0,7], i represents the segment number of the least squares fitting, C i and D i represents the coefficient corresponding to the periodic term, R i Indicates the periodic term. The periodic terms considered in the fitting process include annual swing, semi-annual swing, 1 / 3 annual swing, and 13.7-day swing. The values are usually 1 year, 1 / 2 year, 1 / 3 year, and 13.7 / 365 years. i represents zero-mean white noise.

[0083] In this embodiment, when constructing a linear model and using LS to fit the effective angular momentum, the constructed linear model periodic terms include annual swing, semi-annual swing, 1 / 3 annual swing, and 13.7-day swing, which are usually taken as 1 year, 1 / 2 year, 1 / 3 year, and 13.67 / 365 years.

[0084] S5. Subtract the geodetic angular momentum residual fitting sequence from the geodetic angular momentum residual to obtain a geodetic residual difference value.

[0085] S6. Subtract the effective angular momentum fitting sequence from the effective angular momentum data set to obtain an effective angular momentum residual sequence.

[0086] S7. Extrapolate the geodetic angular momentum residual fitting sequence and the effective angular momentum fitting sequence respectively by using the least square method to obtain a geodetic angular momentum residual fitting extrapolation sequence and an effective angular momentum fitting extrapolation sequence.

[0087] S8. Input the effective angular momentum residual sequence and the geodetic residual difference into the trained neural network model respectively to obtain effective angular momentum residual prediction data and geodetic residual difference prediction data.

[0088] The training process of the convolutional neural network model includes:

[0089] A training UT1-UTC basic dataset and a training effective angular momentum dataset are obtained; the training UT1-UTC basic dataset is converted to an excitation domain to obtain a geodetic angular momentum sequence; the geodetic angular momentum sequence is subtracted from the effective angular momentum dataset to obtain a geodetic angular momentum residual sequence; the geodetic angular momentum residual sequence and the effective angular momentum sequence are fitted using least squares to obtain corresponding fitting terms and residual terms, respectively; the obtained residual terms are used as input sequences for a CNN model; the training residual sequence is normalized to obtain normalized data; and the CNN model is trained using the normalized data to obtain a suitable CNN model.

[0090] In this embodiment, the residual sequence is used to construct a CNN model and perform parameter training. The specific training process of the CNN model is as follows:

[0091] (1) Normalization: Normalize the residual sequence to a suitable range (such as [0, 1] or [-1, 1]) to improve the convergence speed of the model.

[0092] (2) Construct a CNN model. Based on the characteristics of the residual sequence of geodetic angular momentum and effective angular momentum fitting, a suitable CNN model structure is constructed through network training. The model usually includes the following parts:

[0093] Convolutional layer: used to extract local features from the residual sequence. Multiple convolutional layers can be set, each using a small-sized convolution kernel (such as 3×3).

[0094] Pooling layer: used to reduce the dimension of features and reduce the amount of computation. Usually, maximum pooling or average pooling is used.

[0095] Fully connected layer: combines the features extracted by the convolutional layer and the pooling layer to output the final prediction result.

[0096] The convolution layer of the CNN model in this embodiment is set to 3 layers, the size of each convolution kernel is 3×3, the pooling layer adopts maximum pooling, and the number of neurons in the fully connected layer is adjusted according to the cross-validation results. The CNN model is trained using historical data, and the network parameters are optimized by the back propagation algorithm. The mean square error (MSE) is used as the loss function, the model performance is evaluated through cross-validation, and the hyperparameters (such as learning rate, number of convolution kernels, number of layers, etc.) are adjusted to improve the generalization ability of the model. The error statistics after training in this embodiment are as follows Figure 4 、 Figure 5 、 Figure 6 and Figure 7 shown.

[0097] S9. Add the effective angular momentum fitting extrapolation sequence to the effective angular momentum residual prediction data to obtain an effective angular momentum prediction result.

[0098] S10. Add the geodetic angular momentum residual fitting extrapolation sequence and the geodetic residual difference prediction data to obtain a geodetic angular momentum residual prediction result.

[0099] S11. Add the effective angular momentum prediction result to the geodetic angular momentum residual prediction result to obtain a geodetic angular momentum prediction result; and use the Liouville equation and the effects of leap seconds and the Earth's solid tides to obtain a final UT1-UTC prediction result.

[0100] Finally, the accuracy of the forecast results is evaluated by comparing the mean absolute error (MAE) with the international standard sequence Bulletin A. The accuracy evaluation standard of the UT1-UTC parameter is based on the following formula:

[0101]

[0102] Among them, MAE j is the accuracy of the jth time series, n is the length of the time series, i is the length of the predicted time series, P i is the i-th forecast value, O i is the i-th true value.

[0103] This embodiment uses a combined LS+CNN model to fully leverage the advantages of both, taking full advantage of the interpretability of linear models while also introducing deep learning's powerful modeling capabilities for complex time series data. Specific effects and benefits include:

[0104] Improving UT1-UTC Forecast Accuracy: This embodiment significantly improves UT1-UTC forecast accuracy by utilizing a combined LS+CNN method that takes effective angular momentum into account. Traditional methods (such as least squares and time series analysis) have limitations when dealing with complex nonlinear changes and short-term fluctuations. This embodiment combines the linear fitting capabilities of the least squares method with the nonlinear feature extraction capabilities of convolutional neural networks to more accurately capture the complex variations in UT1-UTC time series. Experimental results demonstrate that the method of this embodiment performs well in medium- and long-term (e.g., quarterly to annual) forecasts, significantly reducing forecast errors compared to traditional methods.

[0105] Enhance the generalization ability of the model: This embodiment significantly enhances the generalization ability of the model through technical means such as data preprocessing, model training optimization and cross-validation. Specifically: data normalization eliminates the dimensional differences between different data sources and improves the training efficiency and stability of the model. The convolution layer and pooling layer of CNN can automatically extract local features and global features in the input data, and effectively capture short-term fluctuations and long-term trends in time series. Hyperparameters (such as learning rate, number of convolution kernels, number of layers, etc.) are optimized through cross-validation to ensure that the model has good performance on an independent test set. Experimental results show that the method of this embodiment performs better than the existing technology on an independent test set, and can adapt to changes in different time scales and different data characteristics.

[0106] Promoting the Development of Earth Rotation Parameter Prediction Technology: This embodiment proposes a new UT1-UTC prediction method that combines the advantages of traditional numerical analysis and deep learning techniques, providing a new technical means for high-precision prediction of Earth rotation parameters. This method is not only innovative in theory but also performs well in practical applications. By introducing the concept of effective angular momentum, this embodiment fully considers the impact of angular momentum exchange within the Earth system on UT1-UTC, overcoming the shortcomings of existing technologies that have insufficient consideration of angular momentum exchange. Furthermore, this embodiment provides new ideas and methods for research in related fields, promoting the development of Earth rotation parameter prediction technology.

[0107] In summary, this paper proposes a combined LS+CNN method for UT1-UTC forecasting that takes effective angular momentum into account. By combining the advantages of traditional linear models and deep learning techniques, this method significantly improves the UT1-UTC forecast accuracy and enhances the model's generalization capabilities. This paper not only provides high-precision Earth rotation parameter forecasting technology support for related fields but also promotes its development, possessing important scientific significance and practical application value.

[0108] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of relevant data must comply with relevant regulations.

[0109] Those skilled in the art will 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, database or other media used in the embodiments provided in this application may include at least one of non-volatile and volatile memory. Non-volatile memory may include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory may include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM may be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM).

[0110] The databases involved in the various embodiments provided herein may include at least one of a relational database and a non-relational database. Non-relational databases may include, but are not limited to, distributed databases based on blockchains. The processors involved in the various embodiments provided herein may include, but are not limited to, general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic units, data processing logic units based on quantum computing, and the like.

[0111] The technical features of the above embodiments can 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.

[0112] This document uses specific examples to illustrate the principles and implementation methods of this application. The description of the above examples is only intended to help understand the method and core concept of this application. At the same time, for those skilled in the art, based on the concept of this application, there may be changes in the specific implementation methods and application scope. In summary, the content of this specification should not be understood as limiting this application.

Claims

1. A LS+CNN combined UT1-UTC forecasting method taking into account effective angular momentum, characterized in that: include: Obtain the UT1-UTC basic dataset and effective angular momentum dataset; The effective angular momentum data set includes: atmospheric angular momentum data, ocean angular momentum data, land water angular momentum data and sea level angular momentum data; Converting the UT1-UTC basic data set into an excitation domain to obtain a geodetic angular momentum sequence; subtracting the effective angular momentum data set from the geodetic angular momentum sequence to obtain a geodetic angular momentum residual; performing least square fitting on the geodetic angular momentum residual and the effective angular momentum data set respectively to obtain a geodetic angular momentum residual fitting sequence and an effective angular momentum fitting sequence; Obtaining a geodetic residual difference by subtracting the geodetic angular momentum residual fitting sequence from the geodetic angular momentum residual; subtracting the effective angular momentum fitting sequence from the effective angular momentum data set to obtain an effective angular momentum residual sequence; Extrapolating the geodetic angular momentum residual fitting sequence and the effective angular momentum fitting sequence respectively by the least square method to obtain a geodetic angular momentum residual fitting extrapolation sequence and an effective angular momentum fitting extrapolation sequence; Inputting the effective angular momentum residual sequence and the geodetic residual difference into the trained CNN model respectively to obtain effective angular momentum residual prediction data and geodetic residual difference prediction data; Adding the effective angular momentum fitting extrapolation sequence to the effective angular momentum residual prediction data to obtain an effective angular momentum prediction result; Adding the geodetic angular momentum residual fitting extrapolation sequence and the geodetic residual difference prediction data to obtain a geodetic angular momentum residual prediction result; The effective angular momentum prediction result is added to the geodetic angular momentum residual prediction result to obtain a UT1-UTC prediction result.

2. The LS+CNN combined UT1-UTC forecasting method taking into account effective angular momentum according to claim 1, characterized in that: The UT1-UTC basic data set is converted into the excitation domain to obtain the geodetic angular momentum sequence, specifically including: Subtracting leap seconds from the UT1-UTC basic dataset to obtain a subtracted dataset; Correcting the subtraction data set for the earth's solid tide to obtain a UT1R-TAI parameter sequence; The UT1R-TAI parameter sequence is converted into the excitation domain to obtain the geodetic angular momentum sequence.

3. The LS+CNN combined UT1-UTC forecasting method taking into account effective angular momentum according to claim 1, characterized in that: performing least squares fitting on the geodetic angular momentum residual and the effective angular momentum data set respectively to obtain a geodetic angular momentum residual fitting sequence and an effective angular momentum fitting sequence; subtracting the geodetic angular momentum residual fitting sequence from the geodetic angular momentum residual to obtain a geodetic residual difference; and subtracting the effective angular momentum fitting sequence from the effective angular momentum data set to obtain an effective angular momentum residual sequence, specifically comprising: interpolating the effective angular momentum dataset into an interpolated dataset having the same time interval as the UT1-UTC basic dataset; performing least square fitting on the geodetic angular momentum residual and the interpolation data set respectively to obtain a geodetic angular momentum residual fitting sequence and an effective angular momentum fitting sequence; subtracting the geodetic angular momentum residual from the geodetic angular momentum residual fitting sequence to obtain a geodetic angular momentum residual difference; The effective angular momentum data is subtracted from the effective angular momentum fitting sequence to obtain an effective angular momentum residual sequence.

4. The LS+CNN combined UT1-UTC forecasting method taking into account effective angular momentum according to claim 1, characterized in that: The process of extrapolating the geodetic angular momentum residual fitting sequence and the effective angular momentum fitting sequence respectively by the least square method to obtain the geodetic angular momentum residual fitting extrapolation sequence and the effective angular momentum fitting extrapolation sequence includes: Construction of least squares extrapolation model: Among them, χ(t) represents the least squares fitting result, a k represents a constant, a k+1 represents the linear term, t represents the UTC time, k is [0,7], i represents the segment number of the least squares fitting, C i and D i represents the coefficient corresponding to the periodic term, R i represents the periodic term, e i represents zero-mean white noise; Extrapolation is performed using the above-mentioned geodetic angular momentum residual fitting sequence and effective angular momentum fitting sequence.

5. The LS+CNN combined UT1-UTC forecasting method taking into account effective angular momentum according to claim 1, characterized in that: The effective angular momentum residual sequence and the geodetic residual difference are respectively input into the trained CNN model to obtain effective angular momentum residual prediction data and geodetic residual difference prediction data, specifically including: The neural network model has three convolution layers, and the size of the convolution kernel in each layer is 3×3, which predicts the above-mentioned geodetic residual difference and effective angular momentum residual series.

6. The LS+CNN combined UT1-UTC forecasting method taking into account effective angular momentum according to claim 1, characterized in that: The effective angular momentum prediction result is added to the geodetic angular momentum residual prediction result to obtain the UT1-UTC prediction result, which specifically includes: Adding the effective angular momentum prediction result to the geodetic angular momentum residual prediction result to obtain a geodetic angular momentum prediction result; The geodetic angular momentum prediction result is converted into a UT1R-TAI prediction result using the Liouville equation, and the effects of leap seconds and the Earth's solid tides are added to the UT1R-TAI prediction result to obtain the final UT1-UTC prediction result.

Images