Multi-model fusion polar shift prediction method and system
By constructing a multi-model fusion polar motion prediction method, utilizing linear trend model, multi-channel singular spectrum analysis, and XGBoost model, the problem of insufficient polar motion prediction accuracy in existing technologies is solved, achieving higher accuracy and more stable polar motion prediction.
Patent Information
- Application Number
- CN202511412826.3
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-09-29
- Publication Date
- 2026-01-16
AI Technical Summary
Existing technologies struggle to fully utilize multi-scale, multi-source information in polar motion prediction and have limited processing capabilities for nonlinear and non-stationary signals, resulting in insufficient prediction accuracy and stability.
A multi-model fusion approach is adopted, including constructing a linear trend model, a multi-channel singular spectrum analysis model, and an XGBoost model. By detrending and training the historical time series of polar shifts, linear, periodic, and stochastic information are fused to predict polar shifts.
It improves the accuracy and stability of polar shift prediction, effectively handles multi-source information and coupling relationships in polar shift sequences, and enhances the ability to process nonlinear and non-stationary signals.
Smart Images
Figure CN121350387A_ABST
Abstract
Description
Technical Field
[0001] This application relates to the field of polar motion prediction technology, and in particular to a polar motion prediction method and system based on multi-model fusion. Background Technology
[0002] Polar motion refers to the periodic oscillation of the Earth's rotational axis relative to a reference frame on the Earth's surface. It is mainly caused by factors such as changes in the mass of the atmosphere, oceans, groundwater, and glaciers. Accurate prediction of polar motion is of great significance for satellite navigation, deep space exploration, geophysical research, and high-precision spatiotemporal reference systems such as the Global Positioning System (GPS).
[0003] Currently, polar motion forecasting mainly relies on time series analysis methods, such as least squares (LS), autoregressive moving average (ARMA) models, and neural networks (NN). These methods typically model and predict polar motion based on single-channel time series data, making it difficult to fully utilize the multi-scale, multi-source information present in polar motion data. Furthermore, traditional methods have limitations when handling nonlinear and non-stationary signals, and forecast accuracy and stability still need improvement. Summary of the Invention
[0004] The purpose of this application is to provide a multi-model fusion polar motion prediction method and system to improve the accuracy and stability of polar motion prediction.
[0005] To achieve the above objectives, this application provides the following solution.
[0006] Firstly, this application provides a multi-model fusion method for polar motion prediction, including:
[0007] Obtain the historical time series of polar migration;
[0008] Construct a linear trend model to characterize the linear change trend of polar motion in the historical time series of polar motion;
[0009] The linear trend model is used to detrend the historical time series of extreme shifts to obtain a primary residual series.
[0010] Construct a multi-channel singular spectrum analysis model to characterize the polar shift variation trend in a primary residual sequence;
[0011] The primary residual sequence is de-trended using the multi-channel singular spectrum analysis model to obtain the secondary residual sequence.
[0012] The XGBoost model is trained using the quadratic residual sequence to obtain a well-trained XGBoost model that characterizes the polar shift trend in the quadratic residual sequence.
[0013] Polar shift prediction is performed using the linear trend model, the multi-channel singular spectrum analysis model, and the trained XGBoost model.
[0014] Optionally, the linear trend model includes a linear trend model for the X component and a linear trend model for the Y component;
[0015] Constructing a linear trend model to characterize the linear change trend of polar motion in the historical time series of polar motion specifically includes:
[0016] Construct a linear function;
[0017] Based on the historical time series of polar shifts, the parameters in the linear function are solved using the least squares method to obtain the linear trend models of the X component and the Y component.
[0018] Optionally, the formula for solving the parameters of a linear function using the least squares method is as follows:
[0019]
[0020] Among them, a x and b x The slope and intercept of the linear trend model for the X component are respectively given by a. y and b y denoted as the slope and intercept of the linear trend model for the Y component, respectively; PMx(t) and PMy(t) are the observed values of the X and Y components at time t in the polar motion historical time series, respectively; and a and b are parameters in the linear function, namely the slope and intercept of the linear function, respectively.
[0021] Optionally, a multi-channel singular spectrum analysis model is constructed to characterize the polar shift variation trend in the primary residual sequence, specifically including:
[0022] The X-component and Y-component primary residual sequences in the primary residual sequence are decomposed using a multi-channel singular spectrum analysis method to obtain d eigenvectors; singular spectrum analysis is performed on the d eigenvectors, and the d eigenvectors are sorted in descending order of singular values to obtain an eigenvector sequence;
[0023] Principal mode reconstruction is performed based on the first r eigenvectors in the eigenvector sequence to obtain the principal mode sequence corresponding to the first residual sequence; r and d are both positive integers, where r is the first integer. <d;
[0024] A multi-channel singular spectrum analysis model is constructed based on the dominant mode sequence; the multi-channel singular spectrum analysis model is used to predict the dominant modes of the X component and the dominant modes of the Y component.
[0025] Optionally, the multi-channel singular spectrum analysis model is as follows:
[0026]
[0027] Wherein, PMx_MSSA(t+h) and PMy_MSSA(t+h) are the X-component prediction results and Y-component prediction results of the channel singular spectrum analysis model at time t+h, respectively. and These are the dominant modes of the X and Y components at time t+h, respectively. and Let be the principal modes of the X and Y components at time t+hj, respectively; L be the window size for multi-channel singular spectrum analysis; H be the prediction duration; and a be the main modes of the X and Y components. j The combination coefficients are the main modes of the j time steps prior to time t.
[0028] Optionally, the formula for calculating the combination coefficients of each principal mode in the multi-channel singular spectrum analysis model is as follows:
[0029]
[0030] R L =(a1,a2,Ka) L-1 );
[0031] Among them, R L Let a1, a2, and a be the combination coefficient vector. L-1 These are the combination coefficients for the 1st, 2nd, and (L-1)th principal modes, respectively, π. i Let be the last value of the i-th eigenvector in the eigenvector sequence. Let r be the first L-1 values of the i-th eigenvector in the eigenvector sequence, r be the reconstruction order, and ν be an intermediate variable.
[0032] Optionally, the formula for polar motion prediction using the linear trend model, the multi-channel singular spectrum analysis model, and the trained XGBoost model is as follows:
[0033]
[0034] Wherein, PMX_prediction(t+h) and PMY_prediction(t+h) are the X-component prediction results and Y-component prediction results of the polar shift at time t+h, respectively; PMx_LM(t+h) and PMY_LM(t+h) are the X-component prediction results and Y-component prediction results of the linear model at time t+h, respectively; PMx_MSSA(t+h) and PMY_MSSA(t+h) are the X-component prediction results and Y-component prediction results of the channel singular spectrum analysis model at time t+h, respectively; and PMx_XGB(t+h) and PMY_XGB(t+h) are the X-component prediction results and Y-component prediction results of the trained XGBoost model at time t+h, respectively.
[0035] Secondly, this application provides a multi-model fusion polar shift prediction system, comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the above-described multi-model fusion polar shift prediction method.
[0036] This application provides a multi-model fusion method and system for polar motion prediction. First, it obtains historical time series of polar motion; then, it constructs a linear trend model representing the linear trend of polar motion variation in the historical time series; it uses the linear trend model to de-trend the historical time series of polar motion, obtaining a first-order residual sequence; it constructs a multi-channel singular spectrum analysis model representing the trend of polar motion variation in the first-order residual sequence; it uses the multi-channel singular spectrum analysis model to de-trend the first-order residual sequence, obtaining a second-order residual sequence; it uses the second-order residual sequence to train an XGBoost model, obtaining a trained XGBoost model representing the trend of polar motion variation in the second-order residual sequence; finally, it uses the linear trend model, the multi-channel singular spectrum analysis model, and the trained XGBoost model to perform polar motion prediction. This application integrates a linear model, a multi-channel singular spectrum analysis model, and an XGBoost model to achieve polar motion prediction, thereby solving the problems of insufficient polar motion prediction accuracy, inadequate utilization of multi-source information, and limited processing capabilities for nonlinear and non-stationary signals in existing technologies, and improving the accuracy and stability of polar motion prediction. Attached Figure Description
[0037] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0038] Figure 1This is a flowchart illustrating a multi-model fusion polar motion prediction method provided in an embodiment of this application.
[0039] Figure 2 This is a schematic diagram of a multi-model fusion polar motion prediction method provided in an embodiment of this application.
[0040] Figure 3 This is a schematic diagram of the structure of a computer device provided in an embodiment of this application. Detailed Implementation
[0041] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.
[0042] To make the above-mentioned objectives, features and advantages of this application more apparent and understandable, the application will be further described in detail below with reference to the accompanying drawings and specific embodiments.
[0043] In one exemplary embodiment, a multi-model fusion polar shift prediction method is provided, such as... Figure 1 and Figure 2 The steps shown are 101-107.
[0044] Step 101: Obtain the historical time series of polar migration.
[0045] Step 102: Construct a linear trend model that characterizes the linear change trend of polar motion in the historical time series of polar motion.
[0046] Step 103: Use the linear trend model to de-trend the polar shift historical time series to obtain a primary residual series.
[0047] Step 104: Construct a multi-channel singular spectrum analysis model to characterize the polar shift trend in the primary residual sequence.
[0048] Step 105: Use the multi-channel singular spectrum analysis model to perform detrending processing on the primary residual sequence to obtain the secondary residual sequence.
[0049] Step 106: Train the XGBoost model using the quadratic residual sequence to obtain a trained XGBoost model that characterizes the polar shift trend in the quadratic residual sequence.
[0050] Step 107: Perform polar motion prediction using the linear trend model, the multi-channel singular spectrum analysis model, and the trained XGBoost model.
[0051] Implementing steps 101-107 above can improve the accuracy and stability of polar motion prediction.
[0052] In another exemplary embodiment, the linear trend model in step 102 above includes a linear trend model for the X component and a linear trend model for the Y component. Step 102 above can be replaced by steps 201-202 below.
[0053] Step 201: Construct a linear function.
[0054] Step 202: Based on the historical time series of polar shifts, the parameters in the linear function are solved using the least squares method to obtain the linear trend model of the X component and the linear trend model of the Y component.
[0055] In another exemplary embodiment, in step 201, the polar shift time series data (X component PMx(t) and Y component PMy(t)) are first preprocessed to remove outliers and missing values, and linear trend models are fitted to the polar shift X component PMx(t) and Y component PMy(t) respectively:
[0056]
[0057] Where PMx(t) and PMy(t) are the observed values of the polar motion X component and Y component at time t, respectively, and a x and a y b represents the slope of the linear trend model for the X and Y components, respectively. x and b y ε represents the intercepts of the linear trend models for the X and Y components, respectively. x (t) and ε y (t) represents the residual terms after removing the linear trend, and represents the nonlinear and periodic parts of the pole shift.
[0058] In another exemplary embodiment, step 202 above achieves parameter estimation of the linear function.
[0059] Linear parameters are estimated using the least squares (OLS) method:
[0060]
[0061] a and b are the slope and intercept of the linear function, respectively, and N is the length of the polar shift history time series.
[0062] In another exemplary embodiment, the obtained linear trend model is used for polar motion forecasting. For example, the prediction for the h-th future step using the linear trend model is as follows:
[0063]
[0064] Where PMx_LM(t+h) and PMy_LM(t+h) are the X-component prediction results and Y-component prediction results of the linear model at time t+h, respectively.
[0065] In another exemplary embodiment, in step 103 above, after removing the linear trend term from the polar shift X / Y components of the primary residual sequence, the remaining part is obtained:
[0066]
[0067] in, and These are the polar shift residuals in the first-order residual sequences of the X component and the Y component, respectively.
[0068] In recent years, Singular Spectrum Analysis (SSA) has been increasingly applied to polar motion forecasting due to its advantages in nonlinear time series analysis. However, existing SSA-based polar motion forecasting methods are mostly single-channel analyses, failing to effectively integrate the coupling relationship between the X and Y directions of polar motion, thus limiting the improvement of forecast performance.
[0069] Therefore, there is an urgent need for a polar motion prediction method that can integrate multi-channel information and improve forecast accuracy and stability to meet the needs of modern high-precision Earth reference frames and navigation systems for polar motion prediction.
[0070] In another exemplary embodiment, step 104 described above can be replaced by steps 301-305.
[0071] Step 301: Decompose the primary residual sequence of the X component and the primary residual sequence of the Y component in the primary residual sequence using the multi-channel singular spectrum analysis method to obtain d eigenvectors; perform singular spectrum analysis on the d eigenvectors and sort the d eigenvectors in descending order of singular values to obtain an eigenvector sequence.
[0072] Step 302: Based on the first r feature vectors in the feature vector sequence, perform principal mode reconstruction to obtain the principal mode sequence corresponding to the first residual sequence; r and d are both positive integers, r <d。
[0073] Step 303: Construct a multi-channel singular spectrum analysis model based on the master mode sequence; the multi-channel singular spectrum analysis model is used to predict the master mode of the X component and the master mode of the Y component.
[0074] In another exemplary embodiment, step 301 above is specifically implemented as follows:
[0075] (1) The linear residual sequence of the X component after removing the linear trend is: The first residual sequence of the Y component is: Where t is the time sequence number and N represents the length of the time series. Since polar motion is represented by polar coordinates and consists of two components in the X and Y directions, a dual-channel analysis is chosen.
[0076] (2) Select the window size for dual-channel singular spectrum analysis and extract the principal modes. L represents the window length, which usually needs to satisfy 1 < L < N / 2. For time series with known periods, in order to better separate the periodic part, L is generally taken as a multiple of the period. Therefore, according to Chandler and the number of days in the oscillation period of the annual term, 2190 is selected as the window length L.
[0077] (3) Trajectory Matrix Construction and Singular Value Decomposition: Since the window size is L, the total trajectory matrix X constructed from the trend-removed X component first-order residual sequence and the Y component first-order residual sequence is:
[0078]
[0079] Transform the total trajectory matrix X into the form of a sum of biorthogonal matrices of rank 1, S = XX T ;
[0080] Singular value decomposition yields the expanded components of Multivariate Singular Spectrum Analysis (MSSA):
[0081] X = X1 + X2 + L + X d ;
[0082] in, λ i Let d be the singular values of the trajectory matrix S, and d = max{i:λ} i >0},U i For the corresponding feature vector, It is a factor vector.
[0083] In another exemplary embodiment, in step 302 above, considering X i Related to the i-th maximum singular value of X, we select appropriate eigenvalues and corresponding eigenvectors to reconstruct X:
[0084]
[0085] Where r is the order of the principal component.
[0086] By analyzing X iEach block is averaged against the diagonal to obtain the reconstructed component (RC), which is a new N-time series. Therefore, the m-th reconstructed component RC... m Represented as:
[0087]
[0088] By following the steps above, the first r RCs can be selected, and the reconstructed principal modal components can be used as periodic prediction terms.
[0089] In another exemplary embodiment, step 303 above is specifically implemented as follows:
[0090] The linear residual sequences of the X and Y components, after removing the linear trend, are as follows:
[0091]
[0092] Decomposed by multichannel singular spectrum analysis (MSSA) and Combining the first r reconstructed principal modes determined by singular spectrum analysis as periodic prediction terms, the last L-1 values of the reconstructed principal mode sequence RS(t) are obtained, denoted as F, as follows:
[0093]
[0094] For the X component of the polar shift, Represented as Linear combination:
[0095]
[0096] In the formula, vector R L =(a1,a2,Ka) L-1 This can be represented as:
[0097]
[0098] Among them, R L Let a1, a2, and a be the combination coefficient vector. L-1 These are the combination coefficients for the 1st, 2nd, and (L-1)th principal modes, respectively, π. i Let be the last value of the i-th eigenvector in the eigenvector sequence. Let r be the first L-1 values of the i-th eigenvector in the eigenvector sequence, r be the reconstruction order, and ν be an intermediate variable.
[0099] For example, the predicted value of the dominant mode at time h in the future is:
[0100]
[0101] Wherein, PMx_MSSA(t+h) and PMy_MSSA(t+h) are the X-component prediction results and Y-component prediction results of the channel singular spectrum analysis model at time t+h, respectively. and These are the dominant modes of the X and Y components at time t+h, respectively. and Let be the principal modes of the X and Y components at time t+hj, respectively; L be the window size for multi-channel singular spectrum analysis; H be the prediction duration; and a be the main modes of the X and Y components. j The combination coefficients are the main modes of the j time steps prior to time t.
[0102] In another exemplary embodiment, in step 105 above, after removing the prediction portion of the linear model + MSSA model, the final residual, i.e., the quadratic residual sequence, is obtained:
[0103]
[0104] in, and These are the quadratic residual sequences of the X component and the quadratic residual sequences of the Y component, respectively, and the target variable to be modeled and predicted. The residuals contain nonlinear changes, abrupt changes, small disturbances, and other difficult-to-model components.
[0105] In another exemplary embodiment, step 106 above is specifically implemented as follows:
[0106] The autoregressive historical values of the residuals are used as the input feature vector of the XGBoost model:
[0107]
[0108] Where p is the lag order (typically 5 to 15), and Use the input feature vector for XGBoost.
[0109] Using the trained XGBoost model, predict the nonlinear perturbation at step h in the future:
[0110]
[0111] Where PMx_XGB(t+h) and PMy_XGB(t+h) are the X component prediction results and Y component prediction results of the trained XGBoost model at time t+h, respectively, and XGB x and XGB y All of them are pre-trained XGBoost models.
[0112] In another exemplary embodiment, step 107 is specifically implemented as follows:
[0113] The prediction results of the linear trend model, MSSA model, and trained XGBoost model are summed to obtain the final prediction result of the polar shift parameter:
[0114]
[0115] Wherein, PMX_prediction(t+h) and PMY_prediction(t+h) are the X-component prediction results and Y-component prediction results of the polar shift at time t+h, respectively; PMx_LM(t+h) and PMY_LM(t+h) are the X-component prediction results and Y-component prediction results of the linear model at time t+h, respectively; PMx_MSSA(t+h) and PMY_MSSA(t+h) are the X-component prediction results and Y-component prediction results of the channel singular spectrum analysis model at time t+h, respectively; and PMx_XGB(t+h) and PMY_XGB(t+h) are the X-component prediction results and Y-component prediction results of the trained XGBoost model at time t+h, respectively.
[0116] According to the specific embodiments provided in this application, this application has the following technical effects.
[0117] 1. Improve forecast accuracy. By integrating linear trend analysis, multi-channel singular spectrum analysis, and the XGBoost model, the forecast accuracy of polar motion parameters can be effectively improved by fully utilizing the trend, periodicity, and randomness information in the polar motion sequence.
[0118] 2. Make full use of multi-source information. By adopting the multi-channel singular spectrum analysis method, the coupling relationship between the X and Y directions of the polar motion sequence can be processed simultaneously, improving the model's ability to understand and express complex geophysical processes.
[0119] 3. Enhanced model stability and robustness. The MSSA method has a strong ability to handle nonlinear and non-stationary signals, while the XGBoost model can effectively capture random fluctuations in the residual sequence. The combination of the two significantly improves the stability and robustness of the model.
[0120] 4. Wide applicability. The method of this invention is not only applicable to polar motion parameter forecasting, but can also be extended to the forecasting and analysis of other geophysical time series, demonstrating good versatility and scalability.
[0121] In summary, the multi-model fusion polar motion prediction method proposed in this application can effectively overcome the shortcomings of existing technologies, significantly improve the prediction accuracy and stability of polar motion parameters, and has good application prospects and promotion value.
[0122] Based on the same inventive concept, this application also provides a multi-model fusion polar shift prediction system for implementing the multi-model fusion polar shift prediction method described above. The solution provided by this system is similar to the implementation scheme described in the above method. Therefore, the specific limitations of one or more multi-model fusion polar shift prediction system embodiments provided below can be found in the limitations of the multi-model fusion polar shift prediction method described above, and will not be repeated here.
[0123] In one exemplary embodiment, a multi-model fusion polar shift prediction system is provided. This system can be a server or a terminal, and its internal structure diagram can be as follows: Figure 3 As shown, this computer device includes a processor, memory, input / output (I / O) interfaces, and a communication interface. The processor, memory, and I / O interfaces are connected via a system bus, and the communication interface is also connected to the system bus via the I / O interfaces. The processor provides computational and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system, computer programs, and databases. The internal memory provides the environment for the operating system and computer programs stored in the non-volatile storage media to run. The I / O interfaces are used for exchanging information between the processor and external devices. The communication interface is used for communicating with external terminals via a network connection. When the computer program is executed by the processor, it implements a multi-model fusion polar shift prediction method.
[0124] Those skilled in the art will understand that Figure 3 The structures shown are merely block diagrams of some structures related to the present application and do not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than shown in the figures, or combine certain components, or have different component arrangements. In an exemplary embodiment, a computer device is provided, including a memory and a processor. The memory stores a computer program, and the processor executes the computer program to implement the steps in the above-described method embodiments.
[0125] In one exemplary embodiment, a computer-readable storage medium is provided storing a computer program that, when executed by a processor, implements the steps in the above-described method embodiments.
[0126] 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, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties.
[0127] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. Any references to memory, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can 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 can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM).
[0128] The databases involved in the embodiments provided in this application may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the embodiments provided in this application may be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, quantum computing-based data processing logic devices, etc., and are not limited to these.
[0129] The technical features of the above embodiments can be combined in any way. For the sake of brevity, 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.
[0130] This document uses specific examples to illustrate the principles and implementation methods of this application. The descriptions of the above embodiments are only for the purpose of helping to understand the methods and core ideas of this application. Furthermore, those skilled in the art will recognize that, based on the ideas of this application, there will be changes in the specific implementation methods and application scope. Therefore, the content of this specification should not be construed as a limitation of this application.
Claims
1. A method for predicting polar motion by multi-model fusion, characterized in that, The method comprises the following steps: acquiring a polar motion historical time sequence; constructing a linear trend model representing a linear change trend of the polar motion in the polar motion historical time sequence; performing a detrended processing on the polar motion historical time sequence by using the linear trend model to obtain a first residual sequence; constructing a multi-channel singular spectrum analysis model representing a change trend of the polar motion in the first residual sequence; performing a detrended processing on the first residual sequence by using the multi-channel singular spectrum analysis model to obtain a second residual sequence; training an XGBoost model by using the second residual sequence to obtain a trained XGBoost model representing a change trend of the polar motion in the second residual sequence; performing polar motion prediction by using the linear trend model, the multi-channel singular spectrum analysis model and the trained XGBoost model. 2.The pole shift prediction method of multi-model fusion according to claim 1, characterized in that, The linear trend model comprises an X-component linear trend model and a Y-component linear trend model. The step of constructing the linear trend model representing the linear change trend of the polar motion in the polar motion historical time sequence comprises the following steps: constructing a linear function; solving parameters in the linear function by using a least square method according to the polar motion historical time sequence to obtain the X-component linear trend model and the Y-component linear trend model. 3.The pole shift prediction method of multi-model fusion according to claim 2, characterized in that, The formula for solving the parameters in the linear function by using the least square method is: where a x and b x are the slope and intercept of the linear trend model for the X component, respectively, a y and b y are the slope and intercept of the linear trend model for the Y component, respectively, PMx(t) and PMy(t) are the observed values of the X component and the Y component at time t in the polar motion history time series, respectively, a and b are parameters in the linear function, a and b are the slope and intercept of the linear function, respectively, and N is the length of the polar motion history time series. 4.The pole shift prediction method of multi-model fusion according to claim 1, characterized in that, The step of constructing the multi-channel singular spectrum analysis model representing the change trend of the polar motion in the first residual sequence comprises the following steps: decomposing an X-component first residual sequence and a Y-component first residual sequence in the first residual sequence by using a multi-channel singular spectrum analysis method to obtain d characteristic vectors; performing singular spectrum analysis on the d characteristic vectors, and sorting the d characteristic vectors in a descending order of singular values to obtain a characteristic vector sequence; performing principal mode reconstruction based on the first r characteristic vectors in the characteristic vector sequence to obtain a principal mode sequence corresponding to the first residual sequence; r and d are positive integers, and r < d; constructing the multi-channel singular spectrum analysis model based on the principal mode sequence; the multi-channel singular spectrum analysis model is used for predicting the X-component principal mode and the Y-component principal mode.
5. The pole shift prediction method of claim 1 or 4, wherein, The multi-channel singular spectrum analysis model is: PMx_MSSA(t+h) and PMy_MSSA(t+h) are the prediction results of the X component and the Y component of the channel singular spectrum analysis model at the time t+h respectively, and are the principal modes of the X component and the Y component at the time t+h respectively, and are the principal modes of the X component and the Y component at the time t+h-j respectively, L is the window size of the multi-channel singular spectrum analysis, H is the prediction length, and a j is the combination coefficient of the principal modes at the j time points before the time t. 6.The pole shift prediction method of multi-model fusion according to claim 5, characterized in that, The formula for calculating combination coefficients of each principal mode in the multi-channel singular spectrum analysis model is: R L = (a1, a2, Ka L-1 ); wherein R L is a combination coefficient vector, a1, a2, and a L-1 are combination coefficients of the 1st, 2nd, and L-1th principal modes, respectively, π i is the last value of the i-th eigenvector in the eigenvector sequence, are the first L-1 values of the i-th eigenvector in the eigenvector sequence, r is a reconstruction order, and v is an intermediate variable, 7.The pole shift prediction method of multi-model fusion according to claim 1, characterized in that, The formula for performing the polar motion prediction by using the linear trend model, the multi-channel singular spectrum analysis model and the trained XGBoost model is: wherein, PMX_prediction(t+h) and PMY_prediction(t+h) are respectively an X-component prediction result and a Y-component prediction result of the polar motion at t+h time, PMx_LM(t+h) and PMy_LM(t+h) are respectively an X-component prediction result and a Y-component prediction result of the linear model at t+h time, PMx_MSSA(t+h) and PMy_MSSA(t+h) are respectively an X-component prediction result and a Y-component prediction result of the multi-channel singular spectrum analysis model at t+h time, and PMx_XGB(t+h) and PMy_XGB(t+h) are respectively an X-component prediction result and a Y-component prediction result of the trained XGBoost model at t+h time.
8. A multi-model fusion polar motion prediction system, comprising: A memory, a processor, and a computer program stored in the memory and capable of running on the processor, characterized in that the processor executes the computer program to implement the multi-model fusion polar motion prediction method of any one of claims 1-7.