UT1 prediction method based on adaptive selection of angular momentum according to prediction length

Abstract

This invention discloses a UT1 forecasting method based on adaptive selection of angular momentum according to forecast duration, belonging to the field of geodesy and satellite navigation technology. The method includes: simultaneously acquiring NTSC autonomous EOP observation data and GFZ-released EAM data; filtering the EAM data by removing HAM and SLAM components; dividing the overall forecast window into multiple intervals according to the forecast duration and adaptively selecting the optimal angular momentum combination for each interval; deducting tidal terms and leap seconds from the UT1 sequence and converting it into a GAM sequence using the Liouville equation; constructing the forecast model corresponding to each interval using LS and AR algorithms to generate UT1 forecast data for the corresponding interval; and stitching together the UT1 forecast data from all intervals to output complete UT1 forecast data. This method can adaptively select the angular momentum combination according to the forecast duration, suppressing model noise at the source, improving prediction accuracy, and fully adapting to the NTSC dataset.

Description

Technical Field

[0001] This invention belongs to the field of geodesy and satellite navigation technology, specifically relating to a UT1 prediction method based on adaptive selection of angular momentum for prediction duration. Background Technology

[0002] Universal Time (UT1) is a time scale defined based on the Earth's rotation. It is a core parameter for coordinate transformation between celestial and terrestrial reference systems and is widely used in major national engineering projects such as satellite navigation, deep space exploration, and precision time service. Due to the large volume of observational data and the complexity of the calculation process, measured UT1 products have a fixed time delay, typically several days behind, which cannot meet the needs of real-time and near-real-time operations. Therefore, the industry generally needs to conduct UT1 forecasts spanning 1-90 days. High-precision UT1 forecasts are a key technology for ensuring the real-time positioning accuracy of the BeiDou Navigation Satellite System and supporting the determination of deep spacecraft orbits. Currently, the National Time Service Center of the Chinese Academy of Sciences (NTSC) has built an independent UT1 measurement and service system, achieving independent supply of core UT1 data and laying a data foundation for the development of domestically produced UT1 forecasting technology.

[0003] Studies have shown that the effective angular momentum (EAM) of Earth's fluids is the primary physical source of changes in Earth's rotation, comprising four components: atmospheric angular momentum (AAM), oceanic angular momentum (OAM), hydrological angular momentum (HAM), and sea level angular momentum (SLAM). Introducing EAM as a physical constraint into the UT1 forecast model can effectively improve forecast accuracy and stability.

[0004] Currently, the mainstream UT1 forecasting schemes at home and abroad mainly adopt the basic forecasting framework of least squares (LS) combined with autoregressive (AR), which integrates all four types of angular momentum components, namely AAM, OAM, HAM and SLAM, into a full component EAM as a unified physical constraint, and applies it to UT1 forecasts for the entire period from 1 to 90 days.

[0005] However, these approaches still suffer from three major technical shortcomings: First, the rigid configuration of angular momentum combinations leads to high system noise. Existing technologies consistently use the full-component EAM, failing to remove high-error components such as HAM and SLAM. Since HAM and SLAM models have inherent errors, blindly adding these components introduces additional noise into the forecast system, lowering overall forecast accuracy. Second, they fail to differentiate between time periods, resulting in poor adaptability: short-term (within ten days) Earth rotation disturbances are primarily driven by combined atmospheric and oceanic movements, while medium- to long-term (ten to ninety days) disturbances are dominated by long-period atmospheric variations. A single full-component combination cannot match the geophysical characteristics of different time periods, causing significant accuracy degradation in some timeframes. Third, existing algorithm parameters and logic are largely developed based on the international IERS dataset, lacking specific adaptation strategies for NTSC autonomous observation data. This results in poor localization performance, and the performance of domestically produced forecast products lags significantly behind mainstream international products.

[0006] Therefore, there is an urgent need to develop a high-precision UT1 forecasting method that can adaptively select angular momentum combinations based on forecast duration, suppress model noise at the source, and fully adapt to the NTSC dataset. Summary of the Invention

[0007] To address the aforementioned problems in the existing technology, this invention provides a UT1 prediction method based on adaptive selection of angular momentum during prediction duration. The technical problem to be solved by this invention is achieved through the following technical solution: This invention proposes a UT1 prediction method based on adaptive selection of angular momentum for prediction duration, comprising the following steps: Simultaneously acquire NTSC's autonomous EOP (Earth Orientation Parameters) observation data and GFZ (German Research Centre for Geosciences) EAM data; the EOP observation data includes the UT1 sequence; the EAM data includes the EAM analysis sequence and the EAM finite-duration forecast sequence, and both the EAM analysis sequence and the EAM (Effective Angular Momentum) finite-duration forecast sequence include four components: AAM, OAM, HAM, and SLAM. The EAM data was preprocessed by component filtering to remove the HAM and SLAM components, resulting in preprocessed EAM analysis sequences and preprocessed EAM finite-duration forecast sequences containing only AAM and OAM components. The overall forecast window is divided into multiple intervals based on the forecast duration, and the optimal angular momentum combination is adaptively selected for each interval based on the AAM and OAM components. Subtracting the tidal term and leap seconds from the UT1 sequence yields the UT1R (UT1 with tidal corrections) sequence, which is then converted into a GAM (Geodetic Angular Momentum) sequence using the Liouville equation. Based on the optimal angular momentum combination corresponding to each interval, combined with the preprocessed EAM analysis sequence, the preprocessed EAM finite duration forecast sequence, and the GAM sequence, the LS and AR algorithms are used to construct the forecast model corresponding to each interval, and the UT1 forecast data of the corresponding interval is generated according to the forecast model. The UT1 forecast data for all intervals are stitched together to output complete UT1 forecast data covering the entire forecast window.

[0008] Compared with the prior art, the beneficial effects of the present invention are as follows: This invention provides a UT1 forecasting method based on adaptive selection of angular momentum for forecast duration. First, it simultaneously collects NTSC autonomous EOP observation data and GFZ-released EAM data. The EOP observation data includes the UT1 sequence; the EAM data includes the EAM analysis sequence and the EAM finite-duration forecast sequence, both of which include four components: AAM, OAM, HAM, and SLAM. Second, the HAM and SLAM components in the EAM data are removed. Then, the overall forecast window is divided into multiple intervals according to the forecast duration, and the optimal angular momentum combination is adaptively selected for each interval based on the AAM and OAM components. Next, the tidal term and leap seconds are subtracted from the UT1 sequence to obtain the UT1R sequence, which is then converted into a GAM sequence using the Liouville equation. Then, based on the optimal angular momentum combination corresponding to each interval, and combined with the previous data, a forecast model corresponding to each interval is constructed using a LS combined with AR algorithm, generating UT1 forecast data for the corresponding interval. Finally, the UT1 forecast data for all intervals are stitched together to output complete UT1 forecast data covering the entire forecast window. This method actively eliminates two types of high-error angular momentum components, HAM and SLAM, reducing noise interference from the data source and significantly improving forecast accuracy. On the other hand, it designs a segmented adaptive selection of angular momentum combinations based on the forecast duration, solving the problem of poor adaptability of a single combination across all time periods and achieving balanced forecast performance across the medium and long term. Furthermore, this method is compatible with NTSC's independent EOP observation data, eliminating dependence on foreign models and adapting to domestic operational scenarios.

[0009] The present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Attached Figure Description

[0010] Figure 1A flowchart illustrating a UT1 prediction method based on adaptive selection of angular momentum for prediction duration, provided for an embodiment of the present invention; Figure 2 A flowchart illustrating another UT1 prediction method based on adaptive selection of angular momentum for prediction duration provided in an embodiment of the present invention; Figure 3 This is a comparison chart of the forecast accuracy of the short-term forecast model of the present invention, the forecast accuracy of the traditional EAM full-component combined model, and the forecast accuracy of the existing IERS concurrent forecast. Figure 4 This is a comparison chart of the forecast accuracy of the medium- and long-term forecast model of this invention, the forecast accuracy of the traditional EAM full-component combined model, and the forecast accuracy of the existing IERS concurrent forecast. Figure 5 This is a comparison chart showing the forecast accuracy of the method of the present invention on IERS data, the forecast accuracy of the EAM full component combined model, and the forecast accuracy of existing IERS concurrent data. Detailed Implementation

[0011] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the embodiments described below are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0012] A first aspect of the present invention provides a UT1 prediction method based on adaptive selection of angular momentum for prediction duration. (See also: [link to related information]) Figure 1 , Figure 1 A flowchart illustrating a UT1 prediction method based on adaptive selection of angular momentum for prediction duration, provided for an embodiment of the present invention; Figure 2 A flowchart illustrating another UT1 prediction method based on adaptive selection of angular momentum for prediction duration, provided in an embodiment of the present invention. This method mainly includes the following steps: Step 1: Synchronously acquire NTSC autonomous EOP observation data and GFZ-released EAM data; the EOP observation data includes the UT1 sequence; the EAM data includes the EAM analysis sequence and the EAM finite duration forecast sequence, and both the EAM analysis sequence and the EAM finite duration forecast sequence include four components: AAM, OAM, HAM, and SLAM.

[0013] In this embodiment, two types of data can be collected simultaneously as the basic data sources for subsequent forecasts. The first type of data is NTSC autonomous EOP observation data, which selects continuous historical observation data from the past four years, including UT1 time series values ​​and LOD (Length of Day) time series values, with a sampling interval of 24 hours, serving as the core input sample for the forecast model. The second type of data is EAM data released by GFZ, including EAM analysis sequences of the four major components: atmospheric angular momentum (AAM), ocean angular momentum (OAM), land water angular momentum (HAM), and sea level angular momentum (SLAM), as well as EAM finite-duration forecast sequences, serving as physical constraint inputs. Optionally, the EAM finite-duration forecast sequence can be a 10-day forecast sequence with a sampling interval of 3 hours.

[0014] Step 2: Perform component filtering preprocessing on the EAM data, removing the HAM and SLAM components to obtain preprocessed EAM analysis sequences and preprocessed EAM finite duration forecast sequences containing only AAM and OAM components.

[0015] Specifically, based on the error characteristics of geophysical models, the HAM and SLAM components have inherent errors. Therefore, this invention permanently removes the two high-error components, HAM and SLAM, from all EAM data, retaining only AAM and OAM as effective angular momentum components. This avoids the inherent noise introduced into the forecast system by model lag errors and amplitude bias errors at the source. After the removal operation, the EAM data is simplified to {AAM, OAM}. In other words, the preprocessed EAM analysis sequence and the preprocessed EAM finite-duration forecast sequence obtained after component screening both include only AAM and OAM components.

[0016] This invention actively removes two types of high-error angular momentum components, HAM and SLAM, by filtering high-error components from EAM data, retaining only AAM and OAM as effective angular momentum components. This reduces interference from the data source, significantly reduces the inherent noise of the model, and significantly improves the forecast accuracy. The short-term forecast accuracy is significantly improved compared to the traditional full EAM scheme.

[0017] Furthermore, after completing the component filtering preprocessing of the EAM data, the following steps are also included: A weighted average algorithm was used to reconstruct the preprocessed EAM analysis sequence and the preprocessed EAM finite-duration forecast sequence daily, so that both the preprocessed EAM analysis sequence and the preprocessed EAM finite-duration forecast sequence were aligned with the time scale of the UT1 sequence.

[0018] Specifically, to address the resolution difference between the 3-hour interval sampling of AAM and OAM components and the 24-hour interval sampling of the UT1 sequence, a weighted average algorithm is used to reconstruct the AAM and OAM components daily.

[0019] For example, for each natural day, the data at eight full hours (00:00, 03:00, 06:00, 09:00, 12:00, 15:00, 18:00, and 21:00) are used as input. The daily weighted average is calculated using weighting coefficients of 0.0455, 0.0455, 0.1364, 0.1818, 0.2273, 0.1818, 0.1364, and 0.0455. This is then uniformly reconstructed into a 24-hour time resolution EAM analysis sequence and an EAM finite-duration forecast sequence, ensuring that the two EAM sequences are strictly aligned with the UT1 time series at every full hour.

[0020] Thus, we obtained the EAM analysis sequence and the EAM finite duration forecast sequence, which are aligned with the UT1 time series time scale and contain only AAM and OAM components.

[0021] Step 3: Divide the overall forecast window into multiple intervals according to the forecast duration, and adaptively select the optimal angular momentum combination for each interval based on the AAM and OAM components.

[0022] First, the overall forecast window is divided into short-term and medium-to-long-term intervals based on the forecast duration.

[0023] Optionally, taking the 90-day overall forecast window as an example, days 1-12 can be divided into a short-term interval, and days 13-90 into a medium- to long-term interval.

[0024] Then, for the short-term interval, the combined angular momentum formed by the superposition of AAM and OAM components is selected as the optimal angular momentum combination; at the same time, for the medium and long-term interval, a single AAM component is selected as the optimal angular momentum combination.

[0025] Specifically, since short-term Earth rotation disturbances are dominated by atmospheric and oceanic motions, this invention selects the combination of AAM and OAM as the effective angular momentum in the short-term interval. In other words, the EAM analysis sequence and the EAM finite-duration forecast sequence used in the short-term interval are both AAM+OAM combination sequences, which can accurately fit the short-term fluctuation characteristics.

[0026] Since the Earth's rotation in the medium to long term is dominated by long-term cycles such as annual and seasonal variations, atmospheric motion plays a dominant role as the core excitation source, and the stability of a single AAM component is significantly better than that of a multi-component combination. Therefore, this invention selects a single AAM component as the effective angular momentum in the medium to long term, meaning that both the EAM analysis sequence and the EAM finite-duration forecast sequence used in the medium to long term are single AAM sequences.

[0027] Step 4: Subtract the tidal term and leap second from the UT1 sequence to obtain the UT1R sequence, and then convert the UT1R sequence into a GAM sequence using the Liouville equation.

[0028] Specifically, the acquired UT1 sequence contains abrupt changes caused by artificially inserted leap seconds, and is superimposed with multi-period solid tidal disturbances induced by lunar and solar gravity. Neither of these signals belongs to the natural rotational fluctuations caused by the atmospheric and oceanic fluid spheres, and both would disrupt the temporal stationarity. Therefore, in this embodiment, before using the UT1 sequence, the tidal term and leap seconds are uniformly subtracted to obtain a stationary UT1R sequence. After subsequent interval-based adaptive angular momentum modeling and forecasting are completed, the tidal and leap second corrections are restored, and the final UT1 forecast result conforming to the standards is output.

[0029] Optionally, in this embodiment, the formula for converting the UT1R sequence into a GAM sequence using the Liouville equation is as follows: ; In the formula, It is an EAM sequence. The average angular velocity of Earth's rotation is 7.292115e-5 rad / sec. This represents the UT1R sequence obtained by subtracting the tidal term and leap seconds from the UT1 sequence. Indicates time, Represents the differential symbol.

[0030] It is understandable that the above formula can also be used to convert GAM sequences into UT1R sequences. Furthermore, the UT1R sequence in the above formula can be replaced with UT1 data to achieve mutual conversion between EAM sequences and UT1 sequences.

[0031] Step 5: Based on the optimal angular momentum combination corresponding to each interval, and combining the preprocessed EAM analysis sequence, the preprocessed EAM finite duration forecast sequence, and the GAM sequence, use the LS and AR algorithms to construct the forecast model corresponding to each interval, and generate the UT1 forecast data for the corresponding interval according to the forecast model.

[0032] In this embodiment, both the forecast model constructed for the short-term interval (also known as the short-term forecast model) and the forecast model constructed for the medium- and long-term interval (also known as the medium- and long-term forecast model) are implemented based on an algorithm framework combining LS and AR. The core structure of this algorithm framework is least squares fitting, and its mathematical expression is as follows: ; In the formula, Represents the time-varying fitting function. Indicates time, Let represent the parameters of the linear term in the least squares fit, respectively. and Both represent periodic parameters. Indicates the number of cycles. Indicates the first Each cycle duration.

[0033] For example, the number of cycles in this embodiment The value can be 10, and the corresponding 10 cycle durations can be 9.13, 13.7, 23.9, 27.4, 91.3, 121.75, 182.62, 365.24, 1095.72, and 3396.73 respectively.

[0034] Optionally, in the short term, a short-term forecast model is constructed based on the optimal angular momentum combination formed by the superposition of AAM and OAM components, as well as the LS and AR algorithms. The short-term forecast model uses the preprocessed EAM analysis sequence and GAM sequence as inputs to the LS and AR algorithms, and introduces the preprocessed EAM finite-duration forecast sequence as a priori constraint to generate short-term UT1 forecast data in stages.

[0035] Specifically, the forecasting process of the short-term forecasting model is divided into two stages: In the first stage, based on the optimal angular momentum combination formed by the superposition of AAM and OAM components, the LS algorithm and AR algorithm are used to extrapolate the residuals of the GAM sequence and the preprocessed EAM analysis sequence for the corresponding duration of the first stage to obtain residual forecast data. The residual forecast data is added to the preprocessed EAM finite-duration forecast sequence to obtain GAM forecast data. The GAM forecast data is then converted into short-term first-stage UT1 forecast data by correcting for tidal and leap seconds using the Liouville equation.

[0036] In the second stage, the UT1 sequence is combined with the short-term first-stage UT1 forecast data, and the LS and AR algorithms are used to extrapolate the combined data for the corresponding duration of the second stage to obtain the short-term second-stage UT1 forecast data. By merging the short-term first-stage UT1 forecast data and the short-term second-stage UT1 forecast data, a complete short-term UT1 forecast data is obtained.

[0037] For example, by combining a 10-day EAM finite-duration forecast sequence, the 1-12 day short-term interval can be divided into two stages: the first stage is 1-10 days, and the second stage is 11-12 days. For the first stage (1-10 days), the specific calculation logic is as follows: First, calculate the residuals between the GAM sequence obtained in step 4 and the EAM analysis sequence containing AAM and OAM components; then, use the LS and AR algorithms to extrapolate the residuals for 10 days to obtain residual forecast data; next, add the residual forecast data point-by-point to the 10-day forecast sequence containing AAM and OAM components to obtain the 1-10 day GAM forecast data; finally, convert the GAM forecast data into 1-10 day UT1 forecast data using the Liouville equation and tidal correction. For the second stage 11-12 day forecast, the specific calculation logic is as follows: the UT1 sequence obtained in step 1 is combined with the 1-10 day UT1 forecast data obtained in the first stage as input samples. The LS and AR algorithms are used to perform extrapolation predictions for 11-12 days to obtain the UT1 forecast data for 11-12 days. Finally, the 1-10 day UT1 forecast data and the 11-12 day UT1 forecast data are merged to obtain the complete short-term UT1 forecast data.

[0038] Furthermore, a medium- to long-term forecast model is constructed based on the optimal angular momentum combination formed by a single AAM component and the LS and AR algorithms. The medium- to long-term forecast model uses preprocessed EAM analysis sequences and GAM sequences as inputs to the LS and AR algorithms, and generates medium- to long-term UT1 forecast data by extracting medium- to long-term interval forecast segments after global extrapolation.

[0039] Specifically, the forecasting process of the medium- and long-term forecasting model is divided into two stages. In the first stage, based on the optimal angular momentum combination formed by a single AAM component, the LS and AR algorithms are used to extrapolate the residuals of the GAM sequence and the preprocessed EAM analysis sequence for the corresponding duration of the first stage to obtain residual forecast data. The residual forecast data is added to the preprocessed EAM finite-duration forecast sequence to obtain GAM forecast data. The GAM forecast data is then converted into medium- and long-term first-stage UT1 forecast data by applying the Liouville equation and tidal and leap second corrections. In the second stage, the UT1 sequence is combined with the medium- and long-term first-stage UT1 forecast data. The LS and AR algorithms are used to extrapolate the combined data for the corresponding duration of the second stage, and the data corresponding to the medium- and long-term intervals are extracted to obtain the medium- and long-term UT1 forecast data.

[0040] For example, the medium-to-long-term interval of 13-90 days can also be divided into two stages: the first stage is 1-10 days, and the second stage is 11-90 days. For the forecast of the first stage of 1-10 days, the calculation logic is the same as that of the first stage of the short-term interval. First, the residuals are calculated for the GAM sequence obtained in step 4 and the EAM analysis sequence containing only AAM components. Then, the LS and AR algorithms are used to extrapolate the residuals for 10 days to obtain the residual forecast data. Next, the residual forecast data is added point by point to the 10-day forecast sequence containing only AAM components to obtain the 1-10 day GAM forecast data. Finally, the GAM forecast data is converted into 1-10 day UT1 forecast data through the Liouville equation and tidal correction. For the second stage of the 11-90 day forecast, the specific calculation logic is as follows: the UT1 sequence obtained in step 1 is combined with the 1-10 day UT1 forecast data obtained in the first stage as input samples, and the LS and AR algorithms are used to perform extrapolation forecasts for 11-90 days. The forecast data for 13-90 days is then extracted as the medium- and long-term UT1 forecast data.

[0041] Thus, we have obtained short-term and medium- to long-term UT1 forecast data.

[0042] This invention, by setting fixed segmentation rules, divides the overall forecast window into short-term and medium-to-long-term intervals according to the forecast duration, achieving automatic duration division. It also matches the optimal angular momentum combination for the geophysical characteristics of the forecasts in different intervals. The short-term forecast uses a dual-component combination of AAM and OAM, while the medium-to-long-term forecast uses a single AAM combination, which solves the problem of poor adaptability of a single combination across the entire time period. The forecast performance is balanced across the entire forecast period, and the accuracy of the medium-to-long-term forecast is significantly improved compared to the traditional full EAM scheme.

[0043] Step 6: Concatenate the UT1 forecast data for all intervals to output complete UT1 forecast data covering the entire forecast window.

[0044] Specifically, the 1-12 day short-term UT1 forecast data and the 13-90 day medium- and long-term UT1 forecast data obtained in step 5 are spliced ​​together to form a complete 1-90 day UT1 forecast data.

[0045] This invention addresses the issue of poor adaptability of single combinations across all time periods by matching geophysical features in short-term and medium-to-long-term forecasts, achieving balanced forecast performance across the entire 1-90 day range. The resulting complete UT1 forecast data is used for satellite navigation, deep space exploration, and precise time synchronization for space-ground reference system conversion calculations. The entire algorithm and parameter rules are based on NTSC's independently developed EOP sequence debugging, eliminating reliance on foreign models and adapting to domestic operational scenarios.

[0046] Optionally, in another embodiment of the present invention, the NTSC autonomous EOP observation data in the entire method can be replaced with IERS data to extend to the global EOP forecasting scenario. The IERS data also includes UT1 time series values ​​and LOD time series values, fully adapting to the entire methodological process.

[0047] The method of the present invention will be verified by prediction experiments on the NTSC dataset and the IERS dataset respectively, so as to verify and illustrate the beneficial effects of the present invention.

[0048] Experiment 1: Validation of short-term forecasts based on the NTSC dataset.

[0049] This experiment uses the method of this invention to perform short-term forecasts on the NTSC dataset, while using the traditional EAM full-component combination and contemporaneous IERS daily files as controls. The forecast accuracy is quantified using the mean absolute error, as detailed below: S101. Select historical EOP observation data, EAM analysis sequence, and 10-day forecast sequence from NTSC.

[0050] S102. Remove the HAM and SLAM components from the EAM sequence, combine the remaining AAM and OAM into new EAM data, and convert it into EAM analysis and forecast sequences with daily intervals.

[0051] S103. Using the forecast experiment span from June 2024 to September 2025, more than 420 forecast experiments were conducted using the short-term forecast model of this invention, and a control experiment was conducted using the traditional EAM full component combination model.

[0052] S104. The forecast accuracy is quantified using the mean absolute error (MAE). The forecast accuracy of the short-term forecast model of this invention is compared with the forecast accuracy of the traditional EAM full-component combined model and the concurrent IERS forecast accuracy. The results are as follows: Figure 3 As shown.

[0053] Figure 3 This is a comparison chart showing the forecast accuracy of the short-term forecast model of this invention, the forecast accuracy of the traditional EAM full-component combined model, and the forecast accuracy of the existing IERS concurrent forecast. Figure 3 In the figure, the horizontal axis represents the forecast period in days, and the vertical axis represents the mean absolute error in milliseconds. NTSC_oldEAM in the figure represents the forecast accuracy based on the traditional full-component EAM model combined with NTSC data; NTSC_newEAM represents the forecast accuracy based on the combined AAM+OAM model with NTSC data, i.e., the forecast accuracy of the short-term forecast model of this invention; and IERS represents the forecast accuracy of the IERS daily files for the same period.

[0054] from Figure 3As can be seen, the forecast accuracy of the short-term forecast model UT1 of this invention is improved by 10.70% and 9.97% compared with the traditional complete EAM and IERSdaily, respectively, indicating that the short-term forecast model can effectively improve the forecast accuracy of UT1, proving the effectiveness of the short-term forecast model proposed in this invention.

[0055] Experiment 2: Validation of medium- and long-term forecasts based on the NTSC dataset.

[0056] This experiment uses the method of this invention to perform medium- and long-term forecasts on the NTSC dataset. It also uses the traditional EAM full-component combination and contemporaneous IERS daily files as controls. The forecast accuracy is quantified using the mean absolute error, as detailed below: S201. Select historical EOP observation data, EAM analysis sequence, and 10-day forecast sequence from NTSC.

[0057] S202. Remove the HAM, SLAM, and OAM components from the EAM sequence, combine the remaining AAM to form new EAM data, and convert it into EAM analysis and forecast sequences with daily intervals.

[0058] S203. Using the forecast experiment span from June 2024 to September 2025, more than 420 forecast experiments were conducted using the medium- and long-term forecast model of this invention, and a control experiment was conducted using the traditional EAM full-component combination model.

[0059] S204. Using MAE to quantify forecast accuracy, the forecast accuracy of the medium- and long-term forecast model of this invention is compared with the forecast accuracy of the traditional EAM full-component combined model and the concurrent IERS forecast accuracy. The results are as follows: Figure 4 As shown.

[0060] Figure 4 This is a comparison chart showing the forecast accuracy of the medium- and long-term forecast model of this invention, the forecast accuracy of the traditional EAM full-component combined model, and the forecast accuracy of the existing IERS concurrent forecast. Figure 4 In the figure, the horizontal axis represents the forecast period in days, and the vertical axis represents the mean absolute error in milliseconds. NTSC_oldEAM in the figure represents the forecast accuracy based on the combined model of NTSC data and the traditional full-component EAM model; NTSC_newEAM represents the forecast accuracy based on the combined model of NTSC data and AAM, i.e., the forecast accuracy of the medium- and long-term forecast model of this invention; and IERS represents the forecast accuracy of the IERS daily files for the same period.

[0061] from Figure 4As can be seen, the UT1 forecast accuracy of the medium- and long-term forecast model of this invention is improved by 9.79% and 50.71% compared with the traditional complete EAM and IERSdaily, respectively. This indicates that the medium- and long-term forecast model can effectively improve the forecast accuracy of UT1, proving the effectiveness of the medium- and long-term forecast model proposed in this invention.

[0062] Experiment 3: Universality verification based on the IERS dataset.

[0063] This experiment uses the method of this invention to make forecasts on the IERS dataset, while using the traditional EAM full component combination and the IERS daily files from the same period as controls. The forecast accuracy is quantified using the mean absolute error, as detailed below: S301. Select historical EOP observation data from IERS, along with EAM analysis sequences and 10-day forecast sequences.

[0064] S302. Remove HAM and SLAM data from the EAM sequence, combine AAM and OAM into new EAM data, and convert it into EAM analysis and forecast sequences with daily intervals; at the same time, convert AAM into EAM analysis and forecast sequences with daily intervals.

[0065] S303, with the forecast experiment spanning from June 2024 to September 2025, conducted more than 420 forecast experiments using AAM+OAM and AAM, and used the EAM full component combination model for control experiments.

[0066] S304. Using MAE to quantify forecast accuracy, the forecast accuracy of this invention is compared with the forecast accuracy of the EAM full-component combined model and the concurrent IERS forecast accuracy. The results are as follows: Figure 5 As shown.

[0067] Figure 5 This is a comparison chart showing the forecast accuracy of the method of the present invention on IERS data, the forecast accuracy of the EAM full component combined model, and the forecast accuracy of existing IERS data for the same period. The upper chart represents short-term forecast accuracy, and the lower chart represents medium- and long-term forecast accuracy. The horizontal axis represents the forecast period in days, and the vertical axis represents the mean absolute error in milliseconds (ms). In the chart, IERS_oldEAM represents the forecast accuracy based on IERS data and the traditional EAM full component combined model; IERS_AAM+OAM represents the forecast accuracy based on IERS data and the AAM+OAM model, i.e., the forecast accuracy of the short-term forecast model of the present invention; IERS_AAM represents the forecast accuracy based on IERS data and the AAM model, i.e., the forecast accuracy of the medium- and long-term forecast model of the present invention; and IERS represents the forecast accuracy of the IERS daily files for the same period.

[0068] from Figure 5As can be seen, for short-term forecasts, the combined AAM and OAM forecasts are more effective; for medium- and long-term forecasts, AAM forecasts are more effective.

[0069] This demonstrates that the strategy proposed in this invention, which adaptively selects angular momentum based on forecast duration, uses a combination of AAM and OAM in the short term and a single AAM in the medium and long term. This strategy is applicable not only to the NTSC dataset but also to the IERS dataset.

[0070] In summary, the UT1 forecasting method based on adaptive selection of angular momentum for forecast duration provided by this invention has the following technical advantages compared with the prior art through three core improvements: high error component removal, forecast interval segmentation, and adaptive selection of angular momentum. (1) Suppressing inherent model noise: Actively removing high-error angular momentum components such as HAM and SLAM reduces interference from the data source, and the forecast accuracy is greatly improved; (2) Precise optimization by time period: The optimal EAM combination is matched with the geophysical characteristics of short-term, medium-term and long-term forecasts, which solves the problem of poor adaptability of a single combination throughout the time period and achieves balanced forecast performance across the entire span of 1 to 90 days; (3) Deep adaptation to domestic datasets: The entire set of algorithms and parameter rules are based on NTSC's independent EOP sequence debugging, getting rid of dependence on foreign models and adapting to domestic business operation scenarios; (4) Significant improvement in accuracy: Short term (1~12 days): The accuracy is improved by an average of 10.70% compared to the traditional full EAM solution and by 9.97% compared to IERS daily accuracy; Medium and long term (13~90 days): The accuracy is improved by an average of 9.79% compared to the traditional full EAM solution and by 50.71% compared to IERS daily accuracy.

[0071] (5) High versatility: The technical rules can be adapted to both NTSC and IERS datasets and can be extended to global EOP forecasting scenarios.

[0072] The above description, in conjunction with specific preferred embodiments, provides a further detailed explanation of the present invention. It should not be construed that the specific implementation of the present invention is limited to these descriptions. For those skilled in the art, various simple deductions or substitutions can be made without departing from the concept of the present invention, and all such modifications and substitutions should be considered within the scope of protection of the present invention.

Claims

1. A UT1 prediction method based on adaptive selection of angular momentum for prediction duration, characterized in that, Includes the following steps: The system simultaneously acquires NTSC autonomous EOP observation data and GFZ-released EAM data; wherein, the EOP observation data includes UT1 sequences; the EAM data includes EAM analysis sequences and EAM finite-duration forecast sequences, and both the EAM analysis sequences and the EAM finite-duration forecast sequences include four components: AAM, OAM, HAM, and SLAM. The EAM data is preprocessed by component filtering to remove HAM and SLAM components, resulting in a preprocessed EAM analysis sequence and a preprocessed EAM finite duration forecast sequence containing only AAM and OAM components. The overall forecast window is divided into multiple intervals based on the forecast duration, and the optimal angular momentum combination is adaptively selected for each interval based on the AAM and OAM components. The tidal term and leap second are subtracted from the UT1 sequence to obtain the UT1R sequence, and the UT1R sequence is converted into a GAM sequence using the Liouville equation; Based on the optimal angular momentum combination corresponding to each interval, combined with the preprocessed EAM analysis sequence, the preprocessed EAM finite duration forecast sequence, and the GAM sequence, the LS and AR algorithms are used to construct the forecast model corresponding to each interval, and the UT1 forecast data of the corresponding interval is generated according to the forecast model. The UT1 forecast data for all intervals are stitched together to output complete UT1 forecast data covering the entire forecast window.

2. The UT1 prediction method based on adaptive selection of angular momentum for prediction duration according to claim 1, characterized in that, After completing the component filtering preprocessing of the EAM data, the process also includes: The preprocessed EAM analysis sequence and the preprocessed EAM finite-duration forecast sequence are reconstructed daily using a weighted average algorithm so that both the preprocessed EAM analysis sequence and the preprocessed EAM finite-duration forecast sequence are aligned with the time scale of the UT1 sequence.

3. The UT1 prediction method based on adaptive selection of angular momentum for prediction duration according to claim 1, characterized in that, The overall forecast window is divided into multiple intervals based on the forecast duration, and the optimal angular momentum combination is adaptively selected for each interval based on AAM and OAM components, including: The overall forecast window is divided into short-term and medium-to-long-term intervals based on the forecast duration. For the short-term interval, the combined angular momentum formed by the superposition of AAM and OAM components is selected as the optimal angular momentum combination; at the same time, for the medium- and long-term interval, a single AAM component is selected as the optimal angular momentum combination.

4. The UT1 prediction method based on adaptive selection of angular momentum for prediction duration according to claim 3, characterized in that, The overall forecast window is 90 days; the short-term range is 1-12 days; and the medium-to-long-term range is 13-90 days.

5. The UT1 prediction method based on adaptive selection of angular momentum for prediction duration according to claim 1, characterized in that, The formula for converting the UT1R sequence into a GAM sequence using the Liouville equation is as follows: ； In the formula, It is an EAM sequence. The average angular velocity of Earth's rotation. This represents the UT1R sequence obtained by subtracting the tidal term and leap seconds from the UT1 sequence. Indicates time, Represents the differential symbol.

6. The UT1 prediction method based on adaptive selection of angular momentum for prediction duration according to claim 3, characterized in that, Based on the optimal angular momentum combination corresponding to each interval, and combining the preprocessed EAM analysis sequence, the preprocessed EAM finite-duration forecast sequence, and the GAM sequence, the LS and AR algorithms are used to construct the forecast model corresponding to each interval, and the UT1 forecast data for the corresponding interval is generated according to the forecast model, including: In the short-term interval, a short-term forecast model is constructed based on the optimal angular momentum combination formed by the superposition of AAM and OAM components and the LS and AR algorithms. The short-term forecast model uses the preprocessed EAM analysis sequence and the GAM sequence as inputs to the LS and AR algorithms, and introduces the preprocessed EAM finite-duration forecast sequence as a priori constraint to generate short-term UT1 forecast data in stages. In the medium- to long-term interval, a medium- to long-term forecast model is constructed based on the optimal angular momentum combination formed by a single AAM component and the LS and AR algorithms. The medium- to long-term forecast model uses the preprocessed EAM analysis sequence and the GAM sequence as inputs to the LS and AR algorithms, and generates medium- to long-term UT1 forecast data by extracting medium- to long-term interval forecast segments after global extrapolation.

7. The UT1 prediction method based on adaptive selection of angular momentum for prediction duration according to claim 6, characterized in that, The short-term forecast model uses the preprocessed EAM analysis sequence and the GAM sequence as inputs to the LS and AR algorithms, and simultaneously introduces the preprocessed EAM finite-duration forecast sequence as a priori constraint to generate short-term UT1 forecast data in stages, including: The forecasting process of the short-term forecasting model is divided into two stages. In the first stage, based on the optimal angular momentum combination formed by the superposition of AAM and OAM components, the LS and AR algorithms are used to extrapolate the residuals of the GAM sequence and the preprocessed EAM analysis sequence for the corresponding duration of the first stage to obtain residual forecast data. The residual forecast data is added to the preprocessed EAM finite-duration forecast sequence to obtain GAM forecast data. The GAM forecast data is then converted into short-term first-stage UT1 forecast data by applying the Liouville equation and tidal and leap second corrections. In the second stage, the UT1 sequence is combined with the short-term first-stage UT1 forecast data, and the LS and AR algorithms are used to extrapolate the combined data for the corresponding duration of the second stage to obtain the short-term second-stage UT1 forecast data. By merging the short-term first-stage UT1 forecast data and the short-term second-stage UT1 forecast data, a complete short-term UT1 forecast data is obtained.

8. The UT1 prediction method based on adaptive selection of angular momentum for prediction duration according to claim 6, characterized in that, The medium- and long-term forecast model uses the preprocessed EAM analysis sequence and the GAM sequence as inputs to the LS and AR algorithms, and generates medium- and long-term UT1 forecast data by extracting medium- and long-term interval forecast segments after global extrapolation, including: The forecasting process of the medium- and long-term forecasting model is divided into two stages. In the first stage, based on the optimal angular momentum combination formed by a single AAM component, the LS and AR algorithms are used to extrapolate the residuals of the GAM sequence and the preprocessed EAM analysis sequence for the corresponding duration of the first stage to obtain residual forecast data. The residual forecast data is added to the preprocessed EAM finite-duration forecast sequence to obtain GAM forecast data. The GAM forecast data is then converted into medium- and long-term first-stage UT1 forecast data by applying the Liouville equation and tidal and leap second corrections. In the second stage, the UT1 sequence is combined with the medium- and long-term first-stage UT1 forecast data. The LS and AR algorithms are used to extrapolate the combined data for the corresponding duration of the second stage, and the data corresponding to the medium- and long-term intervals are extracted to obtain the medium- and long-term UT1 forecast data.

9. The UT1 prediction method based on adaptive selection of angular momentum for prediction duration according to claim 6, characterized in that, The mathematical expression for LS is: ； In the formula, Represents the time-varying fitting function. Indicates time, Let represent the parameters of the linear term in the least squares fit, respectively. and Both represent periodic parameters. Indicates the number of cycles. Indicates the first Each cycle duration.

10. A UT1 prediction method based on adaptive selection of angular momentum for prediction duration according to any one of claims 1-9, characterized in that, The complete UT1 forecast data is used for satellite navigation, deep space exploration, and precise time synchronization for space-ground reference system conversion calculations; the NTSC autonomous EOP observation data can be replaced with IERS data.

Images