A method for calculating self-gravity based on polynomial regression

By employing a self-gravity calculation method based on polynomial regression, and using the changes in the two test mass displacements and the MOSA angle as feature inputs, a finite element model is built and function fitting is performed. This solves the problems of large computational load and long calculation time for self-gravity, and achieves fast and accurate self-gravity calculation, which is suitable for ground and on-orbit applications.

CN117744422BActive Publication Date: 2026-11-17INNOVATION ACAD FOR MICROSATELLITES OF CAS +1
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202311505667.2
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2023-11-13
Publication Date
2026-11-17
Estimated Expiration
2043-11-13

AI Technical Summary

Technical Problem

Existing technologies require large computational loads and long processing times for calculating self-gravity in gravitational wave detection, which cannot meet the rapid computational needs of space-based gravitational wave detection missions.

Method used

A self-gravity calculation method based on polynomial regression was adopted. By selecting the changes in the two test mass displacements and the MOSA angle as feature inputs, a finite element model of the spacecraft was built to generate sample data. The self-gravity function was fitted and calculated using the polynomial regression algorithm. Linear normalization, inverse normalization and cross-validation were combined to improve the model performance.

Benefits of technology

It achieves fast and efficient self-gravity calculation, reduces computational load and time requirements, meets the requirements of dynamic full-link simulation tasks, and is applicable to both ground and on-orbit self-gravity calculation.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN117744422B_ABST
    Figure CN117744422B_ABST
Patent Text Reader

Abstract

This invention relates to a method for calculating self-gravity based on polynomial regression. The method includes the following steps: S1, selecting the displacement of two test masses and the angle change between the moving optical subassemblies (MOSA) as feature inputs, and the triaxial acceleration of the two test masses as the target output; S2, constructing a finite element model of the spacecraft, and generating sample data by changing the feature inputs based on the gravity formula using finite element modeling analysis; S3, using a polynomial regression algorithm to fit the self-gravity function to the feature inputs and target outputs based on the sample data, linearly normalizing the input data of the corresponding dimensions of the polynomial and inversely normalizing the output data, and using cross-validation to evaluate the model performance during the regression process; S4, using the regressed self-gravity function to quickly and effectively calculate the self-gravity caused by the change in the spatial position of the two test masses. The beneficial effect is that it allows for a faster and more effective calculation of the self-gravity experienced by the test masses.
Need to check novelty before this filing date? Find Prior Art

Description

[Technical Field]

[0001] This invention relates to the field of space gravitational wave detection technology, specifically to a method for calculating self-gravity based on polynomial regression. [Background Technology]

[0002] In physics, gravitational waves are ripples in the curvature of spacetime that propagate outward from a radiation source in the form of waves, transmitting energy through gravitational radiation. In other words, gravitational waves are a type of matter wave generated by the violent motion and changes in matter and energy.

[0003] According to the Taiji Program, China will launch a constellation of three satellites located at the apex of an equilateral triangle around 2030 to directly detect gravitational waves in the mid-to-low frequency band (0.1 mHz-1.0 Hz) using laser interferometry. The main scientific objective of the Taiji Program is to observe the gravitational wave radiation processes generated by events such as the merger of binary black holes and the merger of objects with extremely high mass ratios.

[0004] The Taiji Program, a space-based gravitational wave detection project, uses laser interferometry to measure the picometer-level optical path change between two test masses, thereby retrieving gravitational wave information. This requires the total residual acceleration of the test masses along the sensitive axis to be less than [a certain value]. The gravitational force between the spacecraft and the test mass is called self-gravity. It is mainly affected by factors such as the spatial location of the test mass and the mass distribution of the spacecraft. The acceleration generated by self-gravity is an important source of residual acceleration noise. Accurately calculating the self-gravity at the location of the test mass and analyzing the acceleration noise it generates are necessary to meet the requirements of space gravitational wave detection missions.

[0005] Related literature 1: "Calculation and compensation of self-gravity of gravitational reference sensor based on FEM [J / OL], China Space Science and Technology 1-9 [2023-04-17]", which uses finite element method to calculate the linear acceleration, angular acceleration and gravitational gradient of the gravitational source in the gravitational reference sensor on the test mass. In order to shorten the calculation time, a "quasi-adaptive" mesh generation method is proposed to reduce the number of meshes, and a counterweight is designed to compensate for self-gravity.

[0006] Related document 2: "A Self-Gravity Simulation Method Based on Distance-Controlled Mesh Refinement, CN202211258923.8[P], 2023-01-24", proposes a self-gravity simulation method based on distance-controlled mesh refinement. This method effectively reduces the total number of cells in the mass body surrounding the test mass, improves computational efficiency, and the generated mesh has the characteristic of "dense near and sparse far", which improves the accuracy of self-gravity simulation results and reduces the computational load.

[0007] Related literature 3: "Structural, thermal, optical and gravitational modeling for LISA[J], Classical and Quantum Gravity, 2004, 21(5): S603." Figure 1 This is a simplified finite element model diagram of the LISA gravitational wave detector spacecraft. Figure 1 As shown, a complete STOP-G analysis model is proposed to meet the design requirements of the LISA gravitational wave detection program. This model uses a single model topology to build a finite element model, and considers factors such as thermal disturbances and structural changes on the spacecraft to analyze and calculate the spacecraft's self-gravity.

[0008] In the field of gravitational wave detection, the finite element method (FEM) is currently the primary method for calculating the self-gravity of the test mass. The general approach involves establishing a geometric model of the spacecraft, meshing it, defining the mass and position of each node using the finite element method, calculating the self-gravity at each node based on Newton's law of universal gravitation, and finally summing the results to obtain the overall self-gravity. To ensure accuracy, this method imposes strict limitations on mesh generation (e.g., a 14101-element, 10563-node three-dimensional finite element mesh is used in reference 3). Theoretically, smaller mesh sizes result in higher computational accuracy; however, as the number of meshes increases, the computational complexity also increases. Even with optimization of mesh generation to improve computational efficiency, the problem of excessive computational load and long computation time remains.

[0009] This invention addresses the technical problems of excessive computational load and long computation time associated with self-gravity by making technical improvements to the self-gravity calculation method. [Summary of the Invention]

[0010] The purpose of this invention is to propose a faster and more efficient method for calculating the self-gravitational force on the quality of a test sample.

[0011] To achieve the above objectives, the technical solution adopted by this invention is a self-gravity calculation method based on polynomial regression, comprising the following steps:

[0012] S1. First, select the displacement of the two test masses and the angle change of the moving optical subassemblies (MOSA) as the feature input, and the triaxial acceleration of the two test masses as the target output.

[0013] S2. Then, a finite element model of the spacecraft is built, and sample data is generated by changing the feature inputs based on the gravity formula using the finite element modeling and analysis method.

[0014] S3. Then, a multinomial regression algorithm is used to perform self-gravity function fitting calculation on the feature input and target output based on the sample data. The input data of the corresponding dimension of the multinomial is linearly normalized to [-1, 1], and the output data of the corresponding dimension is inversely normalized. To prevent overfitting, cross-validation is used to evaluate the model performance during the regression process.

[0015] S4. Use the regressed self-gravity function to quickly and effectively calculate the change in self-gravity due to the change in the spatial position of the two test masses.

[0016] Preferably, step S1:

[0017] Two inspection quality coordinate systems are established. TM1 represents inspection quality 1, with the corresponding coordinate system being O1x1y1z1 and the origin O1 located at the center of inspection quality 1. TM2 represents inspection quality 2, with the corresponding coordinate system being O2x2y2z2 and the origin O2 located at the center of inspection quality 2. θ is the MOSA angle.

[0018] The feature input X and the target output Y are defined as follows:

[0019] X = [l x1 , l y1 , l z1 , l x2 , l y2 , l z2 ,α] T

[0020] Y = [a x1 a y1 a z1 a x2 a y2 a z2 ] T

[0021] Where α represents the change in the MOSA angle, l ij a represents the displacement deviation of the inspection mass j in the i direction. ij Let i represent the acceleration of the test mass j in the direction i caused by its own gravity, where i = x, y, z, and j = 1, 2.

[0022] Preferably, step S2 specifically includes the following sub-steps:

[0023] For any point P(x, y, z) inside a uniformly dense inspection mass, and a point Q(x0, y0, z0) of mass M in space, the formula for calculating the gravitational force of point Q(x0, y0, z0) on the inspection mass is as follows:

[0024]

[0025] A finite element model of the spacecraft platform and payload was built, meshed, and the self-gravity of each node on the test mass was calculated according to the gravity calculation formula. Finally, the self-gravity of the entire spacecraft was obtained by superimposing the results.

[0026] Change feature input l ij ∈[-0.001, 0.001], α∈[-0.01, 0.01], multiple sets of sample data are generated through simulation calculation.

[0027] Preferably, step S3:

[0028] According to the following formula:

[0029]

[0030] The input data corresponding to the polynomial dimension is linearly normalized to [-1, 1];

[0031] According to the following formula:

[0032]

[0033] Perform inverse normalization on the output data of the corresponding dimension of the polynomial.

[0034] Preferably, step S3, the polynomial regression algorithm, is a nonlinear regression method that fits the data by performing linear regression on the high-order polynomials of the input features, capturing the nonlinear relationship between features, and providing a more flexible function fitting capability. The order of the polynomial regression determines the final effect of the function fitting and affects the feature combination. The order refers to the degree of the highest-order term in the polynomial regression, and the feature combination refers to combining the various powers in the original feature X to expand the feature space.

[0035] Preferably, step S3, the multinomial regression algorithm, specifically includes the following sub-steps:

[0036] When the polynomial order is 1, polynomial regression is equivalent to linear regression. Combining the definitions of X and Y, we can obtain:

[0037]

[0038] Among them, w pq The coefficients of the polynomial function are represented by p = 1, 2, ..., 6, and q = 0, 1, ..., 7.

[0039] At this point, the feature combination matrix is ​​simply the input matrix with a constant term added, i.e., S1 = [1, X]. T ] T ;

[0040] When the polynomial order is N, the number of elements in the characteristic combination matrix is:

[0041]

[0042] That is, the number of combinations of selecting N elements from N+d elements, where d = row(X) = 7 represents the feature input dimension;

[0043] When the polynomial order is N, the characteristic combination matrix can be written as:

[0044]

[0045] For Nth-order polynomial regression, the equation can be obtained as follows:

[0046]

[0047] If m sets of sample data are obtained through experimental simulation in step S2, then there are a total of m Nth-order polynomial regression equations, denoted as...

[0048]

[0049]

[0050]

[0051] Wherein, the superscript (i) represents the relevant parameters corresponding to the i-th group of sample data, W is the weight matrix, i.e. the parameters that need to be fitted and solved, S and R correspond to the feature combination matrix and target output matrix related to the feature input, m represents the number of experimental sampling data groups, and n = row(Y) = 6 represents the target output dimension;

[0052] The system of equations obtained from m Nth-order polynomial regressions can be expressed in matrix form as follows:

[0053]

[0054] The least squares solution to this matrix equation is:

[0055]

[0056] Preferably, step S3 is a multinomial regression algorithm:

[0057] In actual fitting processes, the root mean square error (RMSE) is used to measure the difference between predicted and true values.

[0058]

[0059] Preferably, step S3 uses K-fold cross-validation to evaluate model performance:

[0060] The m sets of sample data obtained from the simulation are divided into K parts. K-1 parts are used as training data and 1 part is used as validation data in turn for the experiment.

[0061] Each experiment yields corresponding model performance metrics, and overfitting is determined by observing whether there is a significant performance gap between the training dataset and the validation dataset.

[0062] Finally, the average of the results from the K-fold cross-validation algorithm is used as the accuracy estimate.

[0063] This invention provides a self-gravity calculation method based on polynomial regression, which has the following advantages: Based on the factors influencing self-gravity, the displacements of two test masses and the changes in the MOSA angle are selected as feature inputs, and the triaxial accelerations of the two test masses are used as target outputs. A finite element model of the spacecraft is built, and a large amount of sample data is generated by changing the feature inputs based on the gravity formula. A polynomial regression algorithm is used to fit the self-gravity function to the feature inputs and target outputs based on the sample data. Considering the numerical characteristics of the sample data, the inputs and outputs are linearly normalized and inversely normalized. Furthermore, to prevent overfitting, cross-validation is used to evaluate the model performance during the regression process, improving the accuracy and reliability of the function regression. The regressed function can quickly and effectively calculate the self-gravity caused by changes in the spatial position of the test masses. Compared with the traditional finite element modeling and analysis method, this method is intuitive, simple, accurate, reliable, computationally efficient, and time-saving, meeting the requirements of dynamic full-link simulation tasks. Since the selected feature inputs can be measured in real time on the spacecraft using high-precision sensors, it is applicable to both ground-based and on-orbit self-gravity calculations. The method is feasible and has practical application significance. [Attached Image Description]

[0064] Figure 1 This is a simplified finite element model of the LISA gravitational wave detector spacecraft.

[0065] Figure 2 This is a flowchart illustrating a self-gravity calculation method based on polynomial regression.

[0066] Figure 3 This is a schematic diagram of the spacecraft's self-gravity. In the diagram, TM1 and TM2 represent two test masses, corresponding to coordinate systems O1x1y1z1 and O2x2y2z2. The origins of the coordinate systems O1 and O2 are located at the center of the two test masses, and θ is the MOSA angle.

Detailed Implementation Methods

[0067] The present invention will be further described below with reference to the embodiments and the accompanying drawings.

[0068] Example

[0069] This embodiment implements a method for calculating self-gravity based on polynomial regression.

[0070] Space gravitational wave detection places extremely high demands on system noise performance, with self-gravity being a significant source of noise. Many factors influence self-gravity, such as changes in spacecraft mass, the motion of moving components, and thermal deformation of the structure. In engineering practice, dynamic end-to-end noise model simulation of the spacecraft is necessary to verify whether the system performance meets the corresponding performance requirements, necessitating the calculation of the self-gravity experienced by the test mass at every moment. Traditional finite element modeling and analysis methods require extensive mesh generation and the calculation of space gravity using Newton's law of universal gravitation, resulting in high computational complexity and time consumption, which cannot meet the requirements of dynamic end-to-end simulation. Therefore, to calculate the self-gravity experienced by the test mass more quickly and effectively, this embodiment designs a self-gravity calculation method based on polynomial regression.

[0071] Figure 2 This is a schematic diagram of a self-gravity calculation method based on multinomial regression. (For example...) Figure 2 As shown in this embodiment, the method first selects the displacement of the two test masses and the change in the MOSA angle as feature inputs, and the triaxial acceleration of the two test masses as the target output. Then, a finite element model of the spacecraft is built, and sample data is generated by changing the feature inputs based on the gravity formula. Subsequently, a multinomial regression algorithm is used to fit the feature inputs and target outputs to the sample data. Considering the numerical characteristics of the sample data, the inputs and outputs are linearly normalized and denormalized. In addition, cross-validation is used to evaluate the model performance during the regression process to prevent overfitting. The self-gravity caused by the change in the spatial position of the test masses can be quickly and effectively calculated using the regressed function.

[0072] The following is a detailed implementation of the method in this embodiment.

[0073] Figure 3 This is a schematic diagram of the spacecraft's self-gravity. In the diagram, TM1 and TM2 represent two test masses, corresponding to coordinate systems O1x1y1z1 and O2x2y2z2. The origins of the coordinate systems O1 and O2 are located at the centers of the two test masses, and θ is the MOSA angle. Figure 3 As shown, in this embodiment, considering the spacecraft's operating mode and actual state, the rotation of the test mass is ignored, and its spatial position change can be represented only by displacement. In addition, since the MOSA will rotate during the actual operation of the spacecraft, its included angle will change, thereby affecting the position of the two test masses. Therefore, this embodiment selects the displacement of the two test masses and the change of the MOSA included angle as feature inputs. Self-gravity mainly affects the residual acceleration noise index in the gravitational wave detection system, so the triaxial acceleration of the two test masses is selected as the target output.

[0074] The feature input X and the target output Y are defined as follows:

[0075] X = [l x1, l y1 , l z1 , l x2 , l y2 , l z2 ,α] T #(1)

[0076] Y = [a x1 a y1 a z1 a x2 a y2 a z2 ] T #(2)

[0077] Where α represents the change in the MOSA angle, l ij a represents the displacement deviation of the inspection mass j in the i direction. ij Let i represent the acceleration of the test mass j in the direction i caused by its own gravity, where i = x, y, z, and j = 1, 2.

[0078] For any point P(x, y, z) inside a uniformly dense inspection mass V, and a point Q(x0, y0, z0) of mass M in space, the formula for calculating the gravitational force exerted by this point on the inspection mass is as follows:

[0079]

[0080] A finite element model of the spacecraft platform and payload was built, and the mesh was divided. The self-gravity of each node to the inspection quality was calculated according to Equation (3), and finally the overall self-gravity was obtained by superposition.

[0081] Change feature input l ij Given α ∈ [-0.001, 0.001] and α ∈ [-0.01, 0.01], multiple sets of sample data are generated through simulation calculations. Combining engineering practice and finite element model calculation results, the acceleration noise caused by self-gravity is distributed in 10... -10 ~10 -8 The order of magnitude, while the input is in the order of 10. -3 ~10 -2 In order to eliminate the influence of dimensions and make the final fitting result more intuitive and concise, avoid numerical errors caused by coefficients, and improve fitting speed and accuracy, this embodiment linearly normalizes the input and output data of the corresponding dimensions in the total sample to [-1, 1] and denormalizes them according to Equations (4) and (5).

[0082]

[0083]

[0084] Multinomial regression is a nonlinear regression method that fits data by performing linear regression on a high-order polynomial of input features. It can capture nonlinear relationships between features and provides more flexible function fitting capabilities. The order of the multinomial regression determines the final effect of the function fitting and also affects the feature combination. The order refers to the degree of the highest-order term in the multinomial regression, while the feature combination refers to combining the powers of the original features X to expand the feature space. This is the key to introducing nonlinear relationships and improving the model's fitting ability in multinomial regression. Choosing an appropriate polynomial order and feature combination is crucial in problem-solving. Too low an order and feature combination may lead to underfitting, while too high an order and feature combination may lead to overfitting.

[0085] When the polynomial order is 1, polynomial regression is equivalent to linear regression. Combining the definitions of X and Y, we can obtain...

[0086]

[0087] Among them, w pq Let q represent the coefficients of the polynomial function, p = 1, 2, ..., 6, q = 0, 1, ..., 7.

[0088] At this point, the feature combination matrix is ​​simply the input matrix with a constant term added, i.e., S1 = [1, X]. T ] T .

[0089] When the polynomial order is N, the number of elements in the characteristic combination matrix is:

[0090]

[0091] That is, the number of combinations of selecting N elements from N+d elements, where d = row(X) = 7 represents the feature input dimension.

[0092] At this point, the feature combination matrix can be written as:

[0093]

[0094] Similar to equation (6), for Nth-order polynomial regression, we can obtain the equation...

[0095]

[0096] If m sets of sample data are obtained through experimental simulation, then there are a total of m sets of equations as shown in equation (9).

[0097] remember

[0098]

[0099]

[0100]

[0101] Wherein, the superscript (i) represents the relevant parameters corresponding to the i-th group of sample data, W is the weight matrix, i.e. the parameters that need to be fitted and solved, S and R are the feature combination matrix and target output matrix related to the feature input, m represents the number of experimental sampling data groups, and n = row(Y) = 6 represents the target output dimension.

[0102] The above system of m equations can then be expressed in matrix form as follows:

[0103]

[0104] The least squares solution to this matrix equation is

[0105]

[0106] In actual fitting processes, the root mean square error (RMSE) can be used to measure the difference between predicted and true values.

[0107]

[0108] To prevent overfitting in multinomial regression, K-fold cross-validation is used to evaluate the model. The m sets of data obtained from simulation are divided into K parts, and K-1 parts are used alternately as training data and 1 part as validation data for trials. Performance metrics are obtained for each trial, and overfitting is determined by observing whether there is a significant performance gap between the training and validation sets. Finally, the average of the K results is used as an estimate of the algorithm's accuracy.

[0109] This embodiment of the method selects appropriate feature inputs and target outputs based on considerations of factors influencing self-gravity. It utilizes finite element modeling analysis to generate a large amount of sample data by controlling variables, and employs a multinomial regression algorithm to fit and calculate self-gravity. Simultaneously, it improves the accuracy and reliability of the function regression through data normalization, inverse normalization, and cross-validation. Compared with traditional finite element modeling analysis, this method is intuitive, simple, accurate, reliable, computationally efficient, and time-saving, meeting the requirements of dynamic full-link simulation. Furthermore, since the selected feature inputs can be measured in real-time on the satellite using high-precision sensors, this method is applicable to both ground-based and on-orbit self-gravity calculations.

[0110] The above description is only a preferred embodiment of the present invention. It should be noted that those skilled in the art can make several improvements and additions without departing from the principle of the present invention, and these improvements and additions should also be considered within the scope of protection of the present invention.

Claims

1. A method for calculating self-gravity based on polynomial regression, characterized in that... Includes the following steps: S1. First, select the displacement of the two test masses and the change of the MOSA angle as feature inputs, and the triaxial acceleration of the two test masses as target outputs. S2. Then, a finite element model of the spacecraft is built, and sample data is generated by changing the feature inputs based on the gravity formula using the finite element modeling and analysis method. S3. Then, a multinomial regression algorithm is used to perform self-gravity function fitting calculation on the feature input and target output based on the sample data. The input data of the corresponding dimension of the multinomial is linearly normalized to [-1,1], and the output data of the corresponding dimension is inversely normalized. To prevent overfitting, cross-validation is used to evaluate the model performance during the regression process. S4. Use the regressed self-gravity function to quickly and effectively calculate the self-gravity that changes due to the spatial position of the two test masses. Establish two inspection quality coordinate systems: TM1 represents inspection quality 1, with a corresponding coordinate system O1x1y1z1 and the origin O1 located at the center of inspection quality 1; TM2 represents inspection quality 2, with a corresponding coordinate system O2x2y2z2 and the origin O2 located at the center of inspection quality 2. The angle between the two sides is MOSA; Feature Input and target output The definition is as follows: in, Indicates the change in the MOSA angle. Indicates inspection quality exist Displacement deviation in the direction, Indicates inspection quality Subject to self-gravity acceleration in the direction, ; Step S3, the multinomial regression algorithm, specifically includes the following sub-steps: When the polynomial order is 1, polynomial regression is equivalent to linear regression. and The definition can be obtained as follows: in, The coefficients of the polynomial function are represented. ; At this time, the feature combination matrix It is in the input matrix Add a constant term to the existing term, i.e. ; When the order of the polynomial is When the number of elements in the characteristic combination matrix is: That is, from Select from elements The number of combinations of elements. Indicates the dimension of the feature input; The order of the polynomial is The time feature combination matrix can be written as: for The equation can be obtained by polynomial regression of order 1: If obtained through experimental simulation in step S2 Group sample data, then there are a total of indivual The system of equations obtained by polynomial regression of order 1 is denoted as follows: Among them, superscript This represents the relevant parameters corresponding to the i-th group of sample data. This is the weight matrix, which represents the parameters that need to be fitted and solved. and The feature combination matrix and target output matrix are related to the corresponding feature inputs. Indicates the number of groups of experimental sample data. Indicates the target output dimension; but indivual The system of equations obtained from the polynomial regression of order 1 is expressed in matrix form as follows: The least squares solution to this matrix equation is:

2. The self-gravity calculation method based on polynomial regression according to claim 1, characterized in that... Step S2 specifically includes the following sub-steps: For any point inside a uniformly dense inspection mass There is a mass in the space. point Then point The formula for calculating the gravitational pull on the quality of the inspection is as follows: A finite element model of the spacecraft platform and payload was built, meshed, and the self-gravity of each node on the test mass was calculated according to the gravity calculation formula. Finally, the self-gravity of the entire spacecraft was obtained by superimposing the results. Change feature input , Multiple sets of sample data are generated through simulation calculations.

3. The method for calculating self-gravity based on polynomial regression according to claim 2, characterized in that... Step S3: According to the following formula: The input data of the corresponding dimension of the polynomial is linearly normalized to [-1, 1]. According to the following formula: Perform inverse normalization on the output data of the corresponding dimension of the polynomial.

4. The self-gravity calculation method based on polynomial regression according to claim 3, characterized in that... Step S3: Polynomial Regression Algorithm. This is a non-linear regression method that fits data by performing linear regression on a high-order polynomial of the input features, capturing the non-linear relationships between features and providing more flexible function fitting capabilities. The order of the polynomial regression determines the final effect of the function fitting and affects the feature combination. The order refers to the degree of the highest-order term in the polynomial regression, and the feature combination refers to combining the original features... The various powers in the equation are combined to expand the feature space.

5. The self-gravity calculation method based on polynomial regression according to claim 4, characterized in that... Step S3: Polynomial Regression Algorithm In actual fitting processes, the root mean square error (RMSE) is used to measure the difference between predicted and true values.

6. The method for calculating self-gravity based on polynomial regression according to claim 5, characterized in that... Step S3 adopts Folded cross-validation is used to evaluate model performance. The simulation results Group sample data divided into Divide, take turns One set is used as training data, and one set is used as validation data for the experiment; Each experiment yields corresponding model performance metrics, and overfitting is determined by observing whether there is a significant performance gap between the training dataset and the validation dataset. Finally The average of the results is used as Accuracy estimation of the folded cross-validation algorithm.