Space debris orbit prediction model training method and space debris orbit prediction method

Abstract

The invention provides a space debris orbit prediction model training method and a space debris orbit prediction method, relates to the technical field of space debris orbit prediction, and is used for solving the technical problem that in the prior art, target information is difficult to effectively extract from sparse data, so that space situation changes cannot be accurately perceived. The space debris orbit prediction model training method comprises the following steps: acquiring historical orbit data of space debris; according to historical orbit data, extracting position residual errors between the prediction state vectors and the reference state vector of the space debris at different moments; generating a multi-dimensional feature data set according to the position residual error; performing data dimension reduction processing on the multi-dimensional feature data set to obtain a sample data set after dimension reduction; and training an initial random forest model through a machine learning algorithm by using the sample data set after dimension reduction, so that the initial random forest model establishes a mapping relationship between the orbit parameters of the space debris and future states, and a space debris orbit prediction model is obtained.

Description

Technical Field

[0001] The present invention relates to the field of space debris orbit prediction technology, which can be applied to the fields of machine learning and artificial intelligence technology, and more specifically, to a space debris orbit prediction model training method and a space debris orbit prediction method. Background Art

[0002] Since the first launch of artificial satellites, human beings have been exploring space for more than 60 years. With frequent space activities, space objects have spread all over the Earth's orbit. Except for a small number of spacecraft in orbit, most of them are space debris. According to published data, there are currently more than 36,500 space debris larger than 10 cm, and as many as 1 million space debris larger than 1 cm. These space debris are running at high speed in the Earth's orbit. Once they collide with spacecraft, they will have great destructive power. Therefore, effective monitoring of space debris is of great significance to the safety of spacecraft in orbit.

[0003] At present, human beings mainly rely on ground-based optical and radar equipment to monitor space debris. However, due to the constraints of construction costs and construction conditions, the layout of observation equipment is sparse. Therefore, when batch observing multiple targets at the same time, if there is a lack of reasonable optimization planning for space debris detection tasks, it will lead to insufficient observation time for some targets, small and discontinuous observation data. It is difficult for existing technologies to effectively extract target information from sparse data, which in turn affects the timely and accurate perception of space situation changes. Summary of the invention

[0004] In view of this, the present invention provides a space debris orbit prediction model training method and a space debris orbit prediction method to solve the technical problem that the prior art is difficult to effectively extract target information from sparse data, which leads to the inability to accurately perceive space situation changes.

[0005] One aspect of the present invention provides a method for training a space debris orbit prediction model, comprising: acquiring historical orbit data of space debris; extracting, based on the historical orbit data, the position residual between the predicted state vector and the reference state vector of the space debris at different times, wherein the position residual represents the position change of the space debris in the spatial coordinate system; generating a multidimensional feature data set based on the position residual, wherein the multidimensional feature data set includes a multidimensional feature vector and a label; performing data dimensionality reduction processing on the multidimensional feature data set, transforming the multidimensional feature vector therein into a low-dimensional space, and obtaining a sample data set after dimensionality reduction; using the sample data set after dimensionality reduction, training an initial random forest model through a machine learning algorithm, so that the initial random forest model establishes a mapping relationship between the orbit parameters of the space debris and the future state, and obtains a space debris orbit prediction model.

[0006] According to an embodiment of the present invention, based on historical orbit data, extracting the position residual between the predicted state vector and the reference state vector of the space debris at different times includes: setting a main reference time, wherein the state vector of the space debris at the main reference time is the reference state vector, and the position residual is the difference between the predicted state vector and the reference state vector from any time to the main reference time; in response to setting any one of the N times as the main reference time, respectively extracting the position residual between the predicted state vector and the corresponding reference state vector at each time before the main reference time, wherein N is a positive integer greater than 1.

[0007] According to an embodiment of the present invention, a multidimensional feature data set is subjected to data dimensionality reduction processing, and the multidimensional feature vectors therein are transformed into a low-dimensional space to obtain a sample data set after dimensionality reduction, including: standardizing the multidimensional feature data set; calculating a covariance matrix based on the standardized multidimensional feature data set, wherein the covariance matrix is ​​used to capture the correlation between the multidimensional feature vectors; performing eigenvalue decomposition on the covariance matrix to obtain eigenvectors and corresponding eigenvalues; according to the eigenvectors, determining the portion above the eigenvalue threshold as the principal component; and using a transformation matrix to map the principal component to the low-dimensional space to obtain the sample data set after dimensionality reduction.

[0008] According to an embodiment of the present invention, using the sample data set after dimensionality reduction, training the initial random forest model through a machine learning algorithm includes: determining the parameters to be optimized of the initial random forest model, wherein the parameters to be optimized include the maximum number of iterations, the maximum depth of the decision tree, the minimum number of separation samples, the minimum number of leaf node samples, and the sampling form; using a grid search algorithm to determine the optimal parameter combination of the initial random forest model, and configuring the optimal parameter combination in the initial random forest model; inputting the sample data set after dimensionality reduction into the initial random forest model after configuring the optimal parameter combination for training, so that the model captures the orbital change law of space debris.

[0009] According to an embodiment of the present invention, an initial random forest model is configured with a plurality of decision trees, wherein the sample data set after dimensionality reduction is input into the initial random forest model after configuring the optimal parameter combination for training, including: shuffling the multidimensional feature vectors and labels in the sample data set after dimensionality reduction, wherein the correspondence between the multidimensional feature vectors and labels after the shuffling operation remains unchanged; randomly dividing the sample data set after the shuffling operation into a training set and a test set according to a specified ratio, wherein the training set is used to train the model, and the test set is used to evaluate the model performance; randomly dividing the training set into a plurality of sub-sample sets in a replacement manner, wherein each decision tree randomly selects a sub-sample set for independent training to obtain an independent prediction result; and generating a space debris orbit prediction model based on the trained multiple decision trees.

[0010] According to an embodiment of the present invention, each decision tree randomly selects a subsample set for independent training, including: using a random subspace method to randomly extract multiple features from the subsample set for each decision tree to perform node splitting; using a Gini index algorithm to determine the best splitting feature and splitting point; in response to a node in the decision tree being unable to split, determining that the node is a leaf node, at which point, the training of the decision tree is characterized to be completed.

[0011] According to an embodiment of the present invention, the method further includes: performing a denormalization operation on the prediction results of each decision tree; and integrating the prediction results of multiple decision trees to obtain the orbit prediction results of the space debris.

[0012] According to an embodiment of the present invention, integrating the prediction results of multiple decision trees to obtain the orbit prediction result of the space debris includes: obtaining the prediction results of multiple decision trees; and performing mean calculation on the prediction results of the multiple decision trees to obtain the orbit prediction result of the space debris.

[0013] According to an embodiment of the present invention, it also includes: using a test set to independently evaluate the gap between the prediction result of each decision tree and the true value through multiple indicators, and optimizing the model parameters, wherein the multiple indicators include: root mean square error, mean absolute error and Pearson correlation coefficient.

[0014] Another aspect of the present invention provides a method for predicting the orbit of space debris, comprising: obtaining current orbit data of space debris; inputting the current orbit data into a space debris orbit prediction model to predict the orbit parameters of the space debris; wherein the space debris orbit prediction model is trained according to the above-mentioned space debris orbit prediction model training method.

[0015] Compared with the prior art, the space debris orbit prediction model training method and space debris orbit prediction method provided by the present invention have at least the following beneficial effects:

[0016] (1) The space debris orbit prediction model training method provided by the present invention no longer relies solely on real-time sparse observation data, but instead makes full use of the historical accumulated information of space debris. That is, by mining the orbit and feature historical data of space debris and using a machine learning random forest algorithm to establish a short-term prediction model to analyze the short-term evolution of space debris, it is able to maximize the detection efficiency of equipment under the condition of limited detection equipment resources, quickly and accurately grasp space debris information, and timely and effectively perceive changes in space situation.

[0017] (2) After the random forest prediction model is trained, the space debris orbit prediction model training method provided by the present invention can provide a reliable short-term prediction error for most observation targets by simply inputting their historical observation data. This method is simple and easy to implement. In addition, compared with traditional algorithms, the random forest algorithm has a fast training speed, is parallelizable, is suitable for processing large-scale data sets, and has strong adaptability to outliers and noise. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] The above and other objects, features and advantages of the present invention will become more apparent through the following description of the embodiments of the present invention with reference to the accompanying drawings, in which:

[0019] Figure 1 A flowchart of a space debris orbit prediction model training method according to an embodiment of the present invention is schematically shown;

[0020] Figure 2 A scree plot of attribute features of a multidimensional feature data set according to an embodiment of the present invention is schematically shown;

[0021] Figure 3 A scatter plot between the true value and the predicted value of the error number in the U direction according to an embodiment of the present invention is schematically shown;

[0022] Figure 4 A scatter plot between the true value and the predicted value of the error number in N directions according to an embodiment of the present invention is schematically shown;

[0023] Figure 5 A scatter plot between the true value and the predicted value of the error number in the W direction according to an embodiment of the present invention is schematically shown;

[0024] Figure 6 A schematic diagram showing a principle diagram of a space debris orbit prediction model training method according to an embodiment of the present invention;

[0025] Figure 7 A structural block diagram of a space debris orbit prediction model training device according to an embodiment of the present invention is schematically shown;

[0026] Figure 8 The structural block diagram of an electronic device suitable for implementing a space debris orbit prediction model training method according to an embodiment of the present invention is schematically shown. DETAILED DESCRIPTION

[0027] Below, embodiments of the present invention will be described with reference to the accompanying drawings. However, it should be understood that these descriptions are exemplary only and are not intended to limit the scope of the present invention. In the following detailed description, for ease of explanation, many specific details are set forth to provide a comprehensive understanding of embodiments of the present invention. However, it is apparent that one or more embodiments may also be implemented without these specific details. In addition, in the following description, descriptions of known structures and technologies are omitted to avoid unnecessary confusion of concepts of the present invention.

[0028] The terms used herein are only for describing specific embodiments and are not intended to limit the present invention. The terms "comprise", "include", etc. used herein indicate the existence of the features, steps, operations and / or components, but do not exclude the existence or addition of one or more other features, steps, operations or components.

[0029] All terms (including technical and scientific terms) used herein have the meanings commonly understood by those skilled in the art unless otherwise defined. It should be noted that the terms used herein should be interpreted as having a meaning consistent with the context of this specification and should not be interpreted in an idealized or overly rigid manner.

[0030] When using expressions such as "at least one of A, B, and C, etc.", they should generally be interpreted according to the meaning of the expression commonly understood by those skilled in the art (for example, "a system having at least one of A, B, and C" should include but is not limited to a system having A alone, B alone, C alone, A and B, A and C, B and C, and / or A, B, C, etc.).

[0031] In the embodiments of the present invention, the collection, updating, analysis, processing, use, transmission, provision, disclosure, storage and other aspects of the data involved (for example, including but not limited to user personal information) are in compliance with the provisions of relevant laws and regulations, are used for legitimate purposes, and do not violate public order and good morals. In particular, necessary measures are taken for user personal information to prevent illegal access to user personal information data and maintain the security of user personal information and network security.

[0032] Since the first launch of artificial satellites, human beings have been exploring space for more than 60 years. With frequent space activities, space objects have spread all over the Earth's orbit. Except for a small number of spacecraft in orbit, most of them are space debris. According to published data, there are currently more than 36,500 space debris larger than 10 cm, and as many as 1 million space debris larger than 1 cm. These space debris are running at high speed in the Earth's orbit. Once they collide with spacecraft, they will have great destructive power. Therefore, effective monitoring of space debris is of great significance to the safety of spacecraft in orbit.

[0033] At present, human beings mainly rely on ground-based optical and radar equipment to monitor space debris. However, due to the constraints of construction costs and construction conditions, the layout of observation equipment is sparse. Therefore, when batch observing multiple targets at the same time, if there is a lack of reasonable optimization planning for space debris detection tasks, it will lead to insufficient observation time for some targets, small and discontinuous observation data. It is difficult for existing technologies to effectively extract target information from sparse data, which in turn affects the timely and accurate perception of space situation changes.

[0034] Therefore, how to make full use of historical data on space debris, build a short-term prediction model for information changes, and then guide the optimization of space debris detection tasks is an effective way to maximize the detection efficiency of equipment under the condition of limited detection equipment resources, quickly and accurately grasp space debris information, and perceive space situation changes in a timely and effective manner.

[0035] Based on the above concept, the present invention provides a space debris orbit prediction model training method and a space debris orbit prediction method, which are used to solve the technical problem that the prior art is difficult to effectively extract target information from sparse data, thereby resulting in the inability to accurately perceive space situation changes. The space debris orbit prediction model training method includes:

[0036] Obtain historical orbital data of space debris;

[0037] According to the historical orbit data, the position residual between the predicted state vector and the reference state vector of the space debris at different times is extracted, wherein the position residual represents the position change of the space debris in the space coordinate system;

[0038] Generate a multidimensional feature data set according to the position residual, wherein the multidimensional feature data set includes a multidimensional feature vector and a label;

[0039] Performing data dimensionality reduction processing on the multidimensional feature data set, transforming the multidimensional feature vectors therein into a low-dimensional space, and obtaining a sample data set after dimensionality reduction;

[0040] Using the sample data set after dimensionality reduction, the initial random forest model is trained through a machine learning algorithm, so that the initial random forest model can establish a mapping relationship between the orbital parameters of space debris and its future state, and obtain a space debris orbit prediction model.

[0041] The space debris orbit prediction model training method provided by the embodiment of the present invention no longer relies solely on real-time sparse observation data, but instead makes full use of the historical accumulated information of space debris, that is, by mining the orbit and feature historical data of space debris, and using a machine learning random forest algorithm to establish a short-term prediction model to analyze the short-term evolution of space debris. Therefore, it is possible to maximize the detection efficiency of the equipment when the detection equipment resources are limited, quickly and accurately grasp the space debris information, and timely and effectively perceive the changes in the space situation.

[0042] In order to make the objectives, technical solutions and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with specific embodiments and with reference to the accompanying drawings.

[0043] Figure 1 A flowchart of a space debris orbit prediction model training method according to an embodiment of the present invention is schematically shown.

[0044] like Figure 1 As shown, the space debris orbit prediction model training method of this embodiment may include operations S1 to S5, for example.

[0045] In operation S1, historical orbit data of space debris is obtained.

[0046] In operation S2, the position residual between the predicted state vector and the reference state vector of the space debris at different times is extracted according to the historical orbit data, wherein the position residual represents the position change of the space debris in the space coordinate system.

[0047] In operation S3, a multi-dimensional feature data set is generated according to the position residual, wherein the multi-dimensional feature data set includes a multi-dimensional feature vector and a label.

[0048] In operation S4, a data dimension reduction process is performed on the multidimensional feature data set, and the multidimensional feature vectors therein are transformed into a low-dimensional space to obtain a sample data set after dimension reduction.

[0049] In operation S5, the sample data set after dimensionality reduction is used to train the initial random forest model through a machine learning algorithm, so that the initial random forest model establishes a mapping relationship between the orbital parameters of the space debris and the future state, and obtains a space debris orbit prediction model.

[0050] In this embodiment, operation S1 obtains historical orbital data of space debris. For example, space debris targets that updated more than 10 TLE orbits between September 1 and 30, 2017 can be selected for analysis to obtain their corresponding TLE root numbers. The TLE root numbers refer to various orbital parameters recorded in the TLE data, including orbital inclination, right ascension of ascending node, eccentricity, perigee parameters, orbital period, etc., which are used to accurately describe the motion state of the orbital object.

[0051] TLE (Two-Line Orbital Element): Satellite ephemeris, also known as two-line orbital data format, is used to describe the position and velocity of space objects. Satellite ephemeris can accurately calculate, predict, depict, and track the time, position, velocity, and other operating states of satellites and flying objects; it can express the precise parameters of flying objects such as celestial bodies, satellites, spacecraft, missiles, and space debris; it can place flying objects in three-dimensional space and use time to depict the past, present, and future of celestial bodies.

[0052] In this embodiment, in order to avoid the influence of satellite orbit maneuvers on the model prediction results, the orbits of active satellites are eliminated during the target selection process, and a total of 9493 targets meet the screening conditions, and the target orbits cover all areas of medium and low orbits.

[0053] After obtaining the required historical orbit data, the position residuals between the predicted state vector and the reference state vector of the space debris at different times are extracted based on the historical orbit data as the data set for subsequent model training and verification.

[0054] According to an embodiment of the present invention, operation S2 extracts the position residual between the predicted state vector and the reference state vector of the space debris at different times according to the historical orbit data, and may further include:

[0055] A main reference time is set, wherein the state vector of the space debris at the main reference time is the reference state vector, and the position residual is the difference between the predicted state vector from any time to the main reference time and the reference state vector;

[0056] In response to setting any one of the N moments as a main reference moment, the position residuals between the predicted state vector and the corresponding reference state vector at each moment before the main reference moment are extracted respectively, where N is a positive integer greater than 1.

[0057] In this embodiment, for example, the position residual may be calculated by a Pair-Wise difference method, specifically:

[0058] The Nth moment (the moment corresponding to the latest data among all TLE data in a given time period) is set as the main reference moment. At this moment, the released TLE data has not been predicted in any way, so the state vector (position and velocity) of the space debris is the most reliable.

[0059] The predicted state vector of the space debris from the N-1th moment to the Nth moment is calculated, and then the state vector at the Nth moment (reference state vector) is subtracted from the predicted state vector from the N-1th moment to the Nth moment to obtain the position residual.

[0060] Further calculate the predicted state vector of the space debris from the N-2th moment to the Nth moment minus the state vector at the Nth moment. Repeat this process until the predicted state vector from the 1st moment to the Nth moment minus the state vector at the Nth moment is calculated.

[0061] Then, the N-1th moment is reset as the new main reference moment, and the predicted state vector of the space debris from the N-2th moment to the N-1th moment is calculated minus the state vector at the N-1th moment. This process is repeated until the predicted state vector from the 1st moment to the N-1th moment is calculated minus the state vector at the N-1th moment.

[0062] Repeat the above operation until the N-(N-2)th moment is set as the new main reference moment, and calculate the predicted state vector of the space debris from the 1st moment to the N-(N-2)th moment minus the state vector at the N-(N-2)th moment.

[0063] The residual of the above state vector is calculated based on the geocentric coordinate system, from the position in the stellar coordinate system To the position in the geocentric equatorial coordinate system The conversion between them is as follows:

[0064]

[0065]

[0066] in, is the position of the space debris in the stellar coordinate system along the radial direction U, normal direction N, and tangential direction W, is the position of the space debris along the x, y, and z directions in the geocentric equatorial coordinate system, Ω is the right ascension of the ascending node, i is the orbital inclination, ω is the argument of perigee, f is the true anomaly, Rx(·), RY(·) and RZ(·) are the rotation matrices rotated by a certain angle along the X, Y, and Z axes of the right-hand coordinate system, respectively.

[0067] The position of the target in the geocentric equatorial coordinate system It can be expressed by its instantaneous Kepler root number Calculated, where a represents the semi-major axis of the orbit and e represents the eccentricity of the orbit.

[0068] In this embodiment, the above-mentioned Pair-Wise method is used to generate position residual data of 9493 targets based on the continuous TLE orbits of space debris, and the space environment data contemporaneous with the orbit data is obtained at the same time, including F10.7 and Ap index, to generate 9493 data set files, that is, to generate a multidimensional feature data set of operation S3.

[0069] Considering the limited computing power, 200 data files with the most data are selected for processing in this embodiment. The experimental data totals 341932 pieces, including seven attribute feature information and three directions of error information. Table 1 describes the statistical characteristics of each attribute feature in the sample as a whole.

[0070] Table 1 Statistical description of the sample population

[0071]

[0072] After the multidimensional feature data set is generated, the multidimensional feature data set is further subjected to data dimensionality reduction processing.

[0073] According to an embodiment of the present invention, operation S4 performs data dimensionality reduction processing on the multidimensional feature data set, transforms the multidimensional feature vectors therein into a low-dimensional space, and obtains a sample data set after dimensionality reduction, which may further include:

[0074] Standardize multidimensional feature data sets;

[0075] Based on the standardized multidimensional feature data set, a covariance matrix is ​​calculated, wherein the covariance matrix is ​​used to capture the correlation between multidimensional feature vectors;

[0076] Perform eigenvalue decomposition on the covariance matrix to obtain the eigenvectors and corresponding eigenvalues;

[0077] According to the eigenvector, the part above the eigenvalue threshold is determined as the principal component; the principal component is mapped to the low-dimensional space using the transformation matrix to obtain the sample data set after dimensionality reduction.

[0078] In this embodiment, a principal component analysis (PCA) method is used to perform data dimension reduction processing on attribute feature samples.

[0079] Assuming that the multidimensional feature data set has P-dimensional features and the output dimension after dimensionality reduction is p, the transposed matrix X of the multidimensional feature data set can be expressed as:

[0080] X=(X1,X2,…,X p ) T

[0081] Wherein, T represents the transposition operation. If X1 represents an N×1-dimensional column vector, then X represents an N×P matrix, and N represents the number of data in the multidimensional feature data set.

[0082] The specific data dimensionality reduction process is as follows:

[0083] Calculate the mean vector of a multidimensional feature dataset :

[0084]

[0085] Among them, X ki Represents the kith data in the transposed matrix X.

[0086] Calculate the zero-mean standard matrix of a multidimensional feature dataset :

[0087]

[0088] Calculate the covariance matrix D of the multidimensional feature data set:

[0089]

[0090] Compute the eigenvalues ​​λ1,λ2,...,λ of the covariance matrix p and its corresponding eigenvectors α1, α2, ..., α p The eigenvalues ​​are sorted and the parts above the eigenvalue threshold are determined as the main components.

[0091] Calculate the contribution rate σ of each principal component k And the cumulative contribution rate δ:

[0092]

[0093]

[0094] Select the d-dimensional features whose eigenvalues ​​meet the conditions and assign their corresponding eigenvectors α1, α2, ..., α d The transformation matrix A = [α1, α2, ..., α d ], calculate the dimension reduction data Y of the multidimensional feature data set X:

[0095]

[0096] Most of the information in the original data set is concentrated in the top few principal components. The number of principal components is selected by combining eigenvalue threshold and scree plot intuitive analysis.

[0097] In this embodiment, the PCA algorithm is used to reduce the dimension of the attribute features of the multidimensional feature data set. The sample population contains seven types of attribute features, which are called original components in the PCA algorithm. The characteristic value scree plot of each original component is as follows: Figure 2 shown.

[0098] Figure 2 A scree plot of attribute features of a multidimensional feature data set according to an embodiment of the present invention is schematically shown.

[0099] like Figure 2As shown, in the scree plot of the attribute features of the multidimensional feature data set of an embodiment of the present invention, the component number is the sequence number of the attribute feature. The scree plot can assist in extracting the number of principal components. It can be seen in the figure that the curve of the fourth component begins to flatten out.

[0100] In this embodiment, three principal components with eigenvalues ​​greater than 1 are extracted, which are represented as F1, F2 and F3 respectively. The extraction results are shown in Table 2 in conjunction with the construction of subsequent model training sets and test sets.

[0101] Table 2 Initial eigenvalues ​​and extracted principal component information

[0102]

[0103] Among the three principal components extracted by the above method, the first principal component (F1) is the dominant factor, and its variance accounts for 54.32% of the total variance, followed by the second principal component (F2), whose variance accounts for 16.79% of the total variance, and the cumulative information of the principal components can explain 85.95% of the total variance. The formulas for each principal component and its component coefficient are as follows:

[0104]

[0105]

[0106]

[0107] Among them, Vn represents the nth attribute feature in the original sample population.

[0108] After processing, the seven attribute characteristics in the original sample population are replaced by three principal components that can express more than 85% of its overall information. This process achieves data dimensionality reduction, retains most of the information in the original data, and to a certain extent eliminates the information redundancy in the subsequent random forest model construction process.

[0109] According to an embodiment of the present invention, operation S5 uses the sample data set after dimensionality reduction to train the initial random forest model through a machine learning algorithm, which may further include:

[0110] Determine the parameters to be optimized of the initial random forest model, where the parameters to be optimized include the maximum number of iterations, the maximum depth of the decision tree, the minimum number of separation samples, the minimum number of leaf node samples, and the sampling form;

[0111] The grid search algorithm is used to determine the optimal parameter combination of the initial random forest model, and the optimal parameter combination is configured in the initial random forest model;

[0112] The sample data set after dimensionality reduction is input into the initial random forest model after configuring the optimal parameter combination for training, so that the model can capture the orbital change law of space debris.

[0113] The random forest model is a classifier that uses multiple trees to train and predict samples. In machine learning, a random forest is a classifier that contains multiple decision trees, and its output category is determined by the majority of the categories output by individual trees. It can be used to handle classification and regression problems, as well as dimensionality reduction problems.

[0114] There are 17 commonly used parameters in the random forest model. In order to take into account the actual computing cost, in this embodiment, five indispensable parameters that determine the complexity of the decision tree and the sampling method are selected for hyperparameter optimization, namely: the maximum number of iterations, the maximum depth of the decision tree, the minimum number of separation samples, the minimum number of leaf node samples, and the sampling form.

[0115] The five parameters to be optimized are listed in a grid format, and the optimal parameter combination is screened by cross-validation. The k-fold cross-validation is used to select the combination with the highest score as the optimal combination for model construction.

[0116] Assume that the parameter combination to be optimized is , The objective function represents the score of building a random forest model, then the optimal solution of the objective function can be solved as follows:

[0117]

[0118] Before the grid search is performed, a wide range of values ​​for the parameters to be optimized, the so-called hyperparameters, needs to be set.

[0119] If the combination contains two optimal parameters and The value ranges of the two parameters are ( , )and( , ), defining it as a grid , the parameter combination can be expressed as:

[0120]

[0121]

[0122] Wherein, A = 1, 2, ..., M, B = 1, 2, ..., N. When the parameter grid is determined, the objective function is optimized, and the parameter combination that maximizes the objective function is output as the hyperparameter optimization result of this embodiment.

[0123] This embodiment optimizes five indispensable parameters that affect the complexity of the decision tree and the sampling method in the random forest model based on the hyperparameter optimization algorithm. The preset parameter range and the optimal parameter combination obtained by the hyperparameter optimization algorithm are shown in Table 3.

[0124] Table 3 Parameter ranges to be optimized and optimal solutions

[0125]

[0126] After obtaining the optimal parameter combination, the optimal parameter combination is configured in the initial random forest model.

[0127] According to an embodiment of the present invention, the initial random forest model is configured with a plurality of decision trees, wherein the sample data set after dimensionality reduction is input into the initial random forest model after configuring the optimal parameter combination for training, further comprising:

[0128] Performing a shuffling operation on the multidimensional feature vectors and labels in the sample data set after the dimension reduction, wherein the corresponding relationship between the multidimensional feature vectors and the labels after the shuffling operation remains unchanged;

[0129] The sample data set after the shuffle operation is randomly divided into a training set and a test set according to the specified ratio. The training set is used to train the model, and the test set is used to evaluate the model performance.

[0130] The training set is randomly divided into multiple sub-sample sets with replacement, where each decision tree randomly selects a sub-sample set for independent training to obtain an independent prediction result;

[0131] Based on the trained multiple decision trees, a space debris orbit prediction model is generated.

[0132] In this embodiment, the multidimensional feature vectors and labels in the sample data set that has undergone dimensionality reduction processing are shuffled at the same time, and the corresponding relationship is not changed in the process to ensure the generalization ability of the model. They are randomly divided into training sets and test sets, and the division ratio can be 4:1, for example.

[0133] Then, several sub-sample sets, namely root node samples, are randomly sampled with replacement from the training set. Each decision tree randomly selects a sub-sample set for independent training to obtain an independent prediction result.

[0134] According to an embodiment of the present invention, each decision tree randomly selects a subsample set for independent training, including:

[0135] Using the random subspace method, multiple features are randomly selected from the subsample set for each decision tree to perform node splitting;

[0136] Use the Gini index algorithm to determine the best splitting features and splitting points;

[0137] In response to a node in the decision tree being unable to split, the node is determined to be a leaf node, at which point the training of the decision tree is characterized to be complete.

[0138] In this embodiment, the random subspace method is used to randomly extract m features with replacement from the attribute feature set of the sample, and the Gini index algorithm is used to determine the split node attributes. When the nodes in the decision tree cannot be split, the training of each decision tree is completed without pruning, and finally a space debris orbit prediction model is generated.

[0139] According to an embodiment of the present invention, the space debris orbit prediction model training method further includes:

[0140] Denormalize the prediction results of each decision tree;

[0141] The prediction results of multiple decision trees are integrated to obtain the orbit prediction results of space debris.

[0142] In this embodiment, the prediction results are converted into three prediction parameter values ​​through denormalization, and then the prediction results of each decision tree are integrated.

[0143] According to an embodiment of the present invention, integrating the prediction results of multiple decision trees to obtain the orbit prediction result of space debris further includes:

[0144] Get the prediction results of multiple decision trees;

[0145] The prediction results of multiple decision trees are averaged to obtain the orbit prediction result of the space debris.

[0146] In this embodiment, according to the principle of random forest regression algorithm, the mean value predicted by each decision tree is used as the model prediction result:

[0147]

[0148] in, Represents the prediction result of the model, represents the output of each decision tree, represents the independent variable, and Z represents the number of decision trees.

[0149] According to an embodiment of the present invention, the space debris orbit prediction model training method further includes:

[0150] Using the test set, the gap between the prediction results of each decision tree and the true value is independently evaluated through multiple indicators to optimize the model parameters. The multiple indicators include: root mean square error, mean absolute error and Pearson correlation coefficient.

[0151] In this embodiment, three quantitative accuracy evaluation indicators are selected to estimate the gap between the algorithm estimation value and the true value of the test set to optimize and adjust the model parameters.

[0152] The three indicators are:

[0153] (1) Root Mean Squared Error (RMSE):

[0154]

[0155] (2) Mean Absolute Error (MAE):

[0156]

[0157] (3) Pearson Correlation Coefficient (PCC):

[0158]

[0159] in, represents the true value of the predicted parameter in the test set, represents the predicted value of the parameter to be estimated obtained by the model, Represents the mean of the data.

[0160] The index can intuitively describe the degree of linear correlation between the model prediction value and the true value. The closer the index value is to 1, the higher the linear correlation between the two.

[0161] and The parameter can reflect the difference between the true value and the predicted value of the parameter, and a lower index value represents a higher prediction accuracy.

[0162] In this embodiment, based on 273,546 training data sets and 68,386 experimental verification data sets, the short-term prediction errors du, dn, and dw of the space debris orbits in the radial U, normal N, and tangential W directions are predicted and analyzed.

[0163] (1) U direction error

[0164] The prediction results of the model on the U direction error on the validation data set are as follows: Figure 3 shown.

[0165] Figure 3 A scatter plot between the true value and the predicted value of the error number in the U direction according to an embodiment of the present invention is schematically shown.

[0166] like Figure 3 As shown in the figure, the scatter distribution between the true value of the U direction error of the validation data set and the model prediction value can be found:

[0167] The numerical range of the predicted value and the true value is basically consistent, between 0.1 and 0.65, and there is no obvious prediction range deviation; most of the scatter points are distributed near the 1:1 line, indicating that the gap between the predicted value and the true value is small, and there is no significant overestimation or underestimation; although the overall prediction deviation is small, most of the scatter points are below the 1:1 line, indicating that there is a slight underestimation overall.

[0168] The scatter plot can intuitively display the difference and approximate distribution between the predicted data and the real data, but it cannot quantitatively describe the prediction accuracy. Therefore, in this embodiment, three quantitative accuracy verification indicators are statistically analyzed at the same time as shown in Table 4.

[0169] Table 4. Verification of the accuracy of the prediction results of the U direction error

[0170]

[0171] Judging from the calculation results of the PCC index, the prediction accuracy of the U-direction error reached 0.9996, which means that there is a significant linear correlation between the true value and the predicted value of the parameter, and the prediction accuracy is high; and the RMSE and MAE parameter accuracies also reached 0.0049 and 0.0038 respectively, which are much lower than the statistical mean of the U-direction error sample, indicating that the model has a high prediction accuracy.

[0172] (2) N direction error

[0173] The model's prediction results for N-direction errors on the validation data set are as follows: Figure 4 shown.

[0174] Figure 4 A scatter plot between the true value and the predicted value of the error number in N directions according to an embodiment of the present invention is schematically shown.

[0175] like Figure 4 As shown in the figure, the gap between the model prediction and the true value and the data distribution can be intuitively understood. It can be seen from the figure:

[0176] Most of the predicted N-direction error data are distributed in the range of 0 to 0.5, and a small number of sample parameter values ​​are distributed between 0.5 and 1.8, which are highly consistent with the distribution of true value data; despite fluctuations, most of the scatter points are still distributed near the 1:1 line, and there is no significant underestimation or overestimation, and the prediction effect is good; overall, most of the scatter points are distributed below the 1:1 line, indicating that the model prediction results are underestimated, and the underestimation phenomenon is more prominent in the lower numerical range of 0 to 0.5 and the higher numerical range of 1 to 1.8.

[0177] The scatter plot can intuitively display the difference and approximate distribution between the predicted data and the real data, but it cannot quantitatively describe the prediction accuracy. Therefore, in this embodiment, three quantitative accuracy verification indicators are statistically analyzed at the same time as shown in Table 5.

[0178] Table 5. N-direction error prediction results accuracy verification

[0179]

[0180] The Pearson correlation coefficient can reflect the linear correlation between the model prediction value and the true value. The PCC index of the N-direction error predicted in this experiment reached 0.9961, which means that the two are highly linearly correlated, which to a certain extent reflects the good accuracy of the prediction results. The root mean square error and the mean absolute error can reflect the gap between the predicted value and the true value. The two types of evaluation parameter indicators in this embodiment reached 0.0248 and 0.0175 respectively, which are at a relatively low level, indicating that the gap between the model prediction N-direction error and the true value of the training set is low, and the model accuracy is high.

[0181] (3) W direction error

[0182] The model's prediction results for the W direction error on the validation data set are as follows: Figure 5 shown.

[0183] Figure 5 A scatter plot between the true value and the predicted value of the error number in the W direction according to an embodiment of the present invention is schematically shown.

[0184] like Figure 5 As shown in the figure, it can be seen that: the W direction forecast error value predicted by the model is in the range of 0 to 1.1, and the data distribution is relatively uniform, which is consistent with the data distribution of the true value; the scatter points of the predicted value and the true value are concentrated near the 1:1 line, indicating that there is a high consistency between the two, and the model provides reasonable estimates for most of the validation sets; the model does not have a significant prediction error in the W direction error data distribution interval, the scatter point fluctuation is small, and the model prediction results are highly stable.

[0185] The scatter plot can intuitively display the difference and approximate distribution between the predicted data and the real data, but it cannot quantitatively describe the prediction accuracy. Therefore, in this embodiment, three quantitative accuracy verification indicators are statistically analyzed at the same time as shown in Table 6.

[0186] Table 6. Accuracy verification of W direction error prediction results

[0187]

[0188] From the perspective of quantitative accuracy verification indicators, the random forest prediction model constructed in this experiment has high prediction accuracy. Specifically, the Pearson correlation coefficient of the model prediction W direction error is as high as 0.0092, which fully confirms the good linear correlation between the predicted value and the true value of the verification set. From the perspective of the two types of error verification indicators, the model accuracy also reached 0.0092 and 0.0069, which are far lower than the mean of the true value of the training set parameters and are at a relatively high level.

[0189] The space debris orbit prediction model training method provided by the embodiment of the present invention can provide reliable short-term prediction errors for most observation targets by only inputting their historical observation data after the random forest prediction model training is completed. In addition, compared with traditional algorithms, the random forest algorithm has fast training speed, can be parallelized, is suitable for processing large-scale data sets, and has strong adaptability to outliers and noise.

[0190] The embodiment of the present invention also provides a space debris orbit prediction method, comprising:

[0191] Get the current orbital data of space debris;

[0192] The current orbit data is input into the space debris orbit prediction model to predict the orbit parameters of the space debris; wherein the space debris orbit prediction model is trained according to the above-mentioned space debris orbit prediction model training method, and therefore, it will not be described in detail here.

[0193] The space debris orbit prediction method provided by the embodiment of the present invention no longer relies solely on real-time sparse observation data, but instead makes full use of the historical accumulated information of space debris, that is, by mining the orbit and characteristic historical data of space debris, and using a machine learning random forest algorithm to establish a short-term prediction model to analyze the short-term evolution of space debris. Therefore, it is possible to maximize the detection efficiency of the equipment when the detection equipment resources are limited, quickly and accurately grasp the space debris information, and timely and effectively perceive the changes in the space situation.

[0194] Figure 6 The principle diagram of the space debris orbit prediction model training method according to an embodiment of the present invention is schematically shown.

[0195] like Figure 6As shown, the principle of the space debris orbit prediction model training method of the embodiment of the present invention is implemented as follows:

[0196] 1. Generation of multi-dimensional feature dataset

[0197] The position residual between the TLE root number prediction data at time zero and the TLE root number prediction data at a new time is calculated to generate a multidimensional feature data set.

[0198] 2. Data Preprocessing

[0199] Clean and normalize data to ensure input data is in a uniform format and handle missing values ​​and outliers.

[0200] 3. Standardized Dataset

[0201] The dataset is standardized (mean is 0, variance is 1) to ensure that different features are on the same scale so that features with high variance do not dominate the selection of principal components.

[0202] 4. Data Dimensionality Reduction

[0203] The principal component analysis method is used to reduce the dimension of the attribute feature samples.

[0204] 5. Data Division

[0205] The sample data after dimensionality reduction is randomly divided into training set and test set in a ratio of 4:1. The training set is used to train the model, and the test set is used to evaluate the model performance.

[0206] 6. Parameter Optimization

[0207] The short-term prediction errors du, dn and dw of the orbits of space debris in the radial U, normal N and tangential W directions are analyzed, and the objective function is optimized to achieve the global optimal solution.

[0208] 7. Prediction Model Training

[0209] The random forest model (RF model) is trained using the optimized parameters and training data set to obtain the model's prediction results.

[0210] 8. Effect Evaluation

[0211] The accuracy of the model's prediction results was verified based on an independent validation dataset.

[0212] Figure 7 The structure block diagram of the space debris orbit prediction model training device according to an embodiment of the present invention is schematically shown.

[0213] like Figure 7As shown, the space debris orbit prediction model training device 700 of the embodiment of the present invention includes: an acquisition module 710, an extraction module 720, a generation module 730, a data dimension reduction module 740 and a training module 750.

[0214] The acquisition module 710 is used to acquire the historical orbit data of space debris.

[0215] The extraction module 720 is used to extract the position residual between the predicted state vector and the reference state vector of the space debris at different times according to the historical orbit data, wherein the position residual represents the position change of the space debris in the space coordinate system.

[0216] The generating module 730 is used to generate a multi-dimensional feature data set according to the position residual, wherein the multi-dimensional feature data set includes a multi-dimensional feature vector and a label.

[0217] The data dimension reduction module 740 is used to perform data dimension reduction processing on the multi-dimensional feature data set, transform the multi-dimensional feature vector therein into a low-dimensional space, and obtain a sample data set after dimension reduction;

[0218] The training module 750 is used to use the sample data set after dimensionality reduction to train the initial random forest model through a machine learning algorithm, so that the initial random forest model establishes a mapping relationship between the orbital parameters of space debris and the future state, and obtains a space debris orbit prediction model.

[0219] According to the embodiments of the present invention, any one or more of the modules, submodules, units, and subunits, or at least part of the functions of any one of them can be implemented in one module. According to the embodiments of the present invention, any one or more of the modules, submodules, units, and subunits can be split into multiple modules for implementation. According to the embodiments of the present invention, any one or more of the modules, submodules, units, and subunits can be at least partially implemented as hardware circuits, such as field programmable gate arrays (FPGAs), programmable logic arrays (PLAs), systems on chips, systems on substrates, systems on packages, application specific integrated circuits (ASICs), or can be implemented by hardware or firmware in any other reasonable way of integrating or packaging the circuit, or by any one of the three implementation methods of software, hardware, and firmware, or by a proper combination of any of them. Alternatively, according to the embodiments of the present invention, one or more of the modules, submodules, units, and subunits can be at least partially implemented as computer program modules, and when the computer program modules are run, the corresponding functions can be executed.

[0220] For example, any of the acquisition module 710, the extraction module 720, the generation module 730, the data dimension reduction module 740, and the training module 750 can be combined in one module / unit / sub-unit for implementation, or any of the modules / units / sub-units can be split into multiple modules / units / sub-units. Alternatively, at least part of the functions of one or more of these modules / units / sub-units can be combined with at least part of the functions of other modules / units / sub-units and implemented in one module / unit / sub-unit. According to an embodiment of the present invention, at least one of the acquisition module 710, the extraction module 720, the generation module 730, the data dimension reduction module 740, and the training module 750 can be at least partially implemented as a hardware circuit, such as a field programmable gate array (FPGA), a programmable logic array (PLA), a system on a chip, a system on a substrate, a system on a package, an application specific integrated circuit (ASIC), or can be implemented by hardware or firmware such as any other reasonable way of integrating or packaging the circuit, or by any one of the three implementation methods of software, hardware, and firmware, or by a suitable combination of any of them. Alternatively, at least one of the acquisition module 710, the extraction module 720, the generation module 730, the data dimension reduction module 740 and the training module 750 can be at least partially implemented as a computer program module, and when the computer program module is executed, the corresponding function can be performed.

[0221] It should be noted that the space debris orbit prediction model training device part in the embodiment of the present invention corresponds to the space debris orbit prediction model training method part in the embodiment of the present invention. The description of the space debris orbit prediction model training device part specifically refers to the space debris orbit prediction model training method part, which will not be repeated here.

[0222] Figure 8 The structural block diagram of an electronic device suitable for implementing a space debris orbit prediction model training method according to an embodiment of the present invention is schematically shown. Figure 8 The electronic device shown is only an example and should not bring any limitation to the functions and scope of use of the embodiments of the present invention.

[0223] like Figure 8As shown, the electronic device 800 according to an embodiment of the present invention includes a processor 801, which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) 802 or a program loaded from a storage part 808 to a random access memory (RAM) 803. The processor 801 may include, for example, a general-purpose microprocessor (e.g., a CPU), an instruction set processor and / or a related chipset and / or a special-purpose microprocessor (e.g., an application-specific integrated circuit (ASIC)), etc. The processor 801 may also include an onboard memory for caching purposes. The processor 801 may include a single processing unit or multiple processing units for performing different actions of the method flow according to an embodiment of the present invention.

[0224] In the storage part 808, various programs and data required for the operation of the electronic device 800 are stored. The processor 801, the ROM 802 and the storage part 808 are connected to each other through the bus 804. The processor 801 performs various operations of the method flow according to the embodiment of the present invention by executing the program in the ROM 802 and / or the storage part 808. It should be noted that the program can also be stored in one or more memories other than the ROM 802 and the storage part 808. The processor 801 can also perform various operations of the method flow according to the embodiment of the present invention by executing the program stored in the one or more memories.

[0225] According to an embodiment of the present invention, the electronic device 800 may further include an input / output (I / O) interface 805, which is also connected to the bus 804. The electronic device 800 may further include one or more of the following components connected to the input / output (I / O) interface 805: an input portion 806 including a keyboard, a mouse, etc.; an output portion 807 including a cathode ray tube (CRT), a liquid crystal display (LCD), etc., and a speaker, etc.; a storage portion 808 including a hard disk, etc.; and a communication portion 809 including a network interface card such as a LAN card, a modem, etc. The communication portion 809 performs communication processing via a network such as the Internet. A drive 810 is also connected to the input / output (I / O) interface 805 as needed. A removable medium 811, such as a magnetic disk, an optical disk, a magneto-optical disk, a semiconductor memory, etc., is installed on the drive 810 as needed, so that the computer program read therefrom is installed into the storage portion 808 as needed.

[0226] According to an embodiment of the present invention, the method flow according to an embodiment of the present invention can be implemented as a computer software program. For example, an embodiment of the present invention includes a computer program product, which includes a computer program carried on a computer-readable storage medium, and the computer program includes a program code for executing the method shown in the flowchart. In such an embodiment, the computer program can be downloaded and installed from the network through the communication part 809, and / or installed from the removable medium 811. When the computer program is executed by the processor 801, the above-mentioned functions defined in the system of the embodiment of the present invention are executed. According to an embodiment of the present invention, the system, equipment, device, module, unit, etc. described above can be implemented by a computer program module.

[0227] The present invention also provides a computer-readable storage medium, which may be included in the device / apparatus / system described in the above embodiment; or may exist independently without being assembled into the device / apparatus / system. The above computer-readable storage medium carries one or more programs, and when the above one or more programs are executed, the method according to the embodiment of the present invention is implemented.

[0228] According to an embodiment of the present invention, the computer-readable storage medium may be a non-volatile computer-readable storage medium. For example, it may include but is not limited to: a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination thereof. In the present invention, the computer-readable storage medium may be any tangible medium containing or storing a program, which may be used by or in combination with an instruction execution system, an apparatus or a device.

[0229] For example, according to an embodiment of the present invention, the computer-readable storage medium may include the ROM 802 and / or the storage portion 808 described above and / or one or more memories other than the ROM 802 and the storage portion 808 .

[0230] An embodiment of the present invention also includes a computer program product, which includes a computer program, and the computer program contains program code for executing the method provided by the embodiment of the present invention. When the computer program product runs on an electronic device, the program code is used to enable the electronic device to implement the method provided by the embodiment of the present invention.

[0231] When the computer program is executed by the processor 801, the above functions defined in the system / device of the embodiment of the present invention are executed. According to the embodiment of the present invention, the system, device, module, unit, etc. described above can be implemented by a computer program module.

[0232] In one embodiment, the computer program may rely on tangible storage media such as optical storage devices, magnetic storage devices, etc. In another embodiment, the computer program may also be transmitted and distributed in the form of signals on a network medium, and downloaded and installed through the communication part 809, and / or installed from a removable medium 811. The program code contained in the computer program may be transmitted using any appropriate network medium, including but not limited to: wireless, wired, etc., or any suitable combination of the above.

[0233] According to an embodiment of the present invention, the program code for executing the computer program provided by the embodiment of the present invention can be written in any combination of one or more programming languages. Specifically, these computing programs can be implemented using high-level process and / or object-oriented programming languages, and / or assembly / machine languages. Programming languages ​​include, but are not limited to, Java, C++, python, "C" language or similar programming languages. The program code can be executed entirely on the user computing device, partially on the user device, partially on the remote computing device, or entirely on the remote computing device or server. In the case of a remote computing device, the remote computing device can be connected to the user computing device through any type of network, including a local area network (LAN) or a wide area network (WAN), or can be connected to an external computing device (e.g., using an Internet service provider to connect through the Internet).

[0234] The flowcharts and block diagrams in the accompanying drawings illustrate the possible architecture, functions and operations of the systems, methods and computer program products according to various embodiments of the present invention. In this regard, each box in the flowchart or block diagram may represent a module, a program segment, or a part of a code, and the above-mentioned module, program segment, or a part of the code contains one or more executable instructions for implementing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the box may also occur in an order different from the order marked in the accompanying drawings. For example, two boxes represented in succession can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each box in the block diagram or flowchart, and the combination of boxes in the block diagram or flowchart, can be implemented with a dedicated hardware-based system that performs the specified function or operation, or can be implemented with a combination of dedicated hardware and computer instructions. It can be understood by those skilled in the art that the features recorded in the various embodiments of the present invention can be combined and / or combined in various ways, even if such a combination or combination is not explicitly recorded in the present invention. In particular, without departing from the spirit and teachings of the present invention, the features recorded in the various embodiments of the present invention can be combined and / or combined in various ways. All such combinations and / or combinations fall within the scope of the present invention.

[0235] The embodiments of the present invention are described above. However, these embodiments are only for the purpose of illustration, and are not intended to limit the scope of the present invention. Although each embodiment is described above, it does not mean that the measures in each embodiment cannot be used in combination advantageously. Without departing from the scope of the present invention, those skilled in the art may make various substitutions and modifications, which should all fall within the scope of the present invention.

Claims

1. A space debris orbit prediction model training method, characterized in that: The method comprises: Obtain historical orbital data of space debris; Extracting, based on the historical orbit data, position residuals between the predicted state vectors and the reference state vectors of the space debris at different times, wherein the position residuals represent position changes of the space debris in the space coordinate system; Generate a multidimensional feature data set according to the position residual, wherein the multidimensional feature data set includes a multidimensional feature vector and a label; Performing data dimensionality reduction processing on the multidimensional feature data set, transforming the multidimensional feature vectors therein into a low-dimensional space, and obtaining a sample data set after dimensionality reduction; The sample data set after dimensionality reduction is used to train an initial random forest model through a machine learning algorithm, so that the initial random forest model establishes a mapping relationship between the orbital parameters of the space debris and the future state, and obtains a space debris orbit prediction model.

2. The method according to claim 1, characterized in that The extracting, according to the historical orbit data, the position residual between the predicted state vector and the reference state vector of the space debris at different times comprises: Setting a main reference time, wherein the state vector of the space debris at the main reference time is the reference state vector, and the position residual is the difference between the predicted state vector from any time to the main reference time and the reference state vector; In response to setting any one of the N moments as a main reference moment, the position residuals between the predicted state vector and the corresponding reference state vector at each moment before the main reference moment are extracted respectively, where N is a positive integer greater than 1.

3. The method according to claim 1, characterized in that: The data dimensionality reduction process is performed on the multidimensional feature data set to transform the multidimensional feature vectors therein into a low-dimensional space to obtain a sample data set after dimensionality reduction, which includes: Performing standardization processing on the multidimensional feature data set; Calculating a covariance matrix based on the standardized multidimensional feature data set, wherein the covariance matrix is ​​used to capture the correlation between the multidimensional feature vectors; Performing eigenvalue decomposition on the covariance matrix to obtain eigenvectors and corresponding eigenvalues; According to the eigenvector, a portion higher than an eigenvalue threshold is determined as a principal component; The principal components are mapped to a low-dimensional space using a transformation matrix to obtain a sample data set after dimensionality reduction.

4. The method according to claim 1, characterized in that: The method of using the sample data set after dimensionality reduction to train the initial random forest model through a machine learning algorithm includes: Determine the parameters to be optimized of the initial random forest model, wherein the parameters to be optimized include the maximum number of iterations, the maximum depth of the decision tree, the minimum number of separation samples, the minimum number of leaf node samples, and the sampling form; Using a grid search algorithm to determine the optimal parameter combination of the initial random forest model, and configuring the optimal parameter combination in the initial random forest model; The sample data set after dimensionality reduction is input into the initial random forest model after configuring the optimal parameter combination for training, so that the model can capture the orbital change law of the space debris.

5. The method according to claim 4, characterized in that The initial random forest model is configured with a plurality of decision trees, wherein the inputting of the sample data set after dimensionality reduction into the initial random forest model configured with the optimal parameter combination for training comprises: Performing a shuffling operation on the multidimensional feature vectors and labels in the sample data set after the dimension reduction, wherein the corresponding relationship between the multidimensional feature vectors and the labels after the shuffling operation remains unchanged; The sample data set after the shuffle operation is randomly divided into a training set and a test set according to a specified ratio, wherein the training set is used to train the model and the test set is used to evaluate the model performance; The training set is randomly divided into a plurality of sub-sample sets by a method of replacement, wherein each decision tree randomly selects a sub-sample set for independent training to obtain an independent prediction result; Based on the trained multiple decision trees, a space debris orbit prediction model is generated.

6. The method according to claim 5, characterized in that Each decision tree randomly selects a subsample set for independent training, including: Using a random subspace method, a plurality of features are randomly selected from the subsample set for each decision tree to perform node splitting; Use the Gini index algorithm to determine the best splitting features and splitting points; In response to a node in the decision tree being unable to be split, the node is determined to be a leaf node, at which point it is indicated that the training of the decision tree is completed.

7. The method according to claim 5, characterized in that Also includes: Denormalize the prediction results of each decision tree; The prediction results of multiple decision trees are integrated to obtain the orbit prediction result of the space debris.

8. The method according to claim 7, characterized in that The integrating the prediction results of multiple decision trees to obtain the orbit prediction result of the space debris includes: Get the prediction results of multiple decision trees; The prediction results of multiple decision trees are averaged to obtain the orbit prediction result of the space debris.

9. The method according to claim 8, characterized in that Also includes: Using the test set, the gap between the prediction result of each decision tree and the true value is independently evaluated through multiple indicators to optimize the model parameters, wherein the multiple indicators include: root mean square error, mean absolute error and Pearson correlation coefficient.

10. A method for predicting the orbit of space debris, characterized in that: The method comprises: Get the current orbital data of space debris; Inputting the current orbit data into a space debris orbit prediction model to predict the orbit parameters of the space debris; The space debris orbit prediction model is trained according to the space debris orbit prediction model training method according to any one of claims 1 to 9.

Images