A method for quickly identifying intention of space target based on early warning information

Abstract

The application provides a space target intention rapid identification method based on early warning information, and belongs to the technical field of space debris collision early warning and space threat early warning. The method uses the historical early warning information of the space target to analyze and extract the possible characteristics of the orbit data, so as to obtain a characteristic database, and then uses a supervised classification algorithm to establish the mapping relationship between the characteristic database and the target intention and the mapping relationship between the key characteristics and the target intention twice, and finally realizes the intention identification of real-time early warning information through a space target intention database. The application solves the problem of low timeliness and accuracy of manual identification in the existing space target intention identification method, and compared with the prior art, the application can realize accurate identification of the target intention in orbit in real time, and can be better applied to space debris collision avoidance, satellite autonomous management, space threat early warning and other task scenes.

Description

Technical Field

[0001] This invention relates to the fields of space debris collision early warning and space threat early warning technology, specifically to a method for rapid identification of space target intent based on early warning information. Background Technology

[0002] As human space activities continue, the number of non-cooperative targets in space is increasing, with space debris, which accounts for the vast majority, numbering in the hundreds of thousands. This exacerbates the congestion of the space environment and greatly increases the risk of collisions between spacecraft in orbit and other targets.

[0003] Meanwhile, with the continuous development of space technology, the orbital maneuvering capabilities of non-cooperative space targets are becoming increasingly enhanced, and their maneuvering behavior is becoming more complex and variable. Therefore, it is necessary to conduct research on autonomous, rapid, and accurate algorithms for identifying the orbital behavior intentions of space targets, in order to achieve early warning functions for high-risk orbital behaviors of space targets.

[0004] Currently, my country's space target behavior intent recognition technology is weak, still relying on expert experience and knowledge for manual intent analysis and recognition. This makes it difficult to quickly and effectively identify space target behavior intent in advance, and thus difficult to adapt to the increasingly complex and rapidly changing space environment. Summary of the Invention

[0005] The technical problem solved by this invention is that existing spatial target intent recognition methods suffer from low timeliness and accuracy in manual identification.

[0006] To solve the above problems, the technical solution of the present invention is as follows:

[0007] A method for rapid identification of spatial target intent based on early warning information includes the following steps:

[0008] S1. Extract historical orbital data of space targets using historical early warning information and label the target's historical true intentions to obtain the original library construction data;

[0009] S2. Analyze and extract possible features from the orbital data. By performing coordinate transformation and calculation on the original database construction data, possible feature values ​​are obtained. These possible feature values ​​are used as feature database construction data. The feature database construction data is stored in chronological order to obtain the feature database.

[0010] S3. Establish a mapping relationship between the feature database and the target intent through a supervised classification algorithm, optimize feature selection using the particle swarm optimization algorithm, extract key features for target intent recognition, and store the key feature data in chronological order to obtain a key feature database.

[0011] S4. By establishing a mapping relationship between the key feature database and the target intent through a supervised classification algorithm, a spatial target intent database is constructed, thereby establishing a mapping relationship between key features and target intent.

[0012] S5. Real-time warning information intent recognition is achieved through a spatial target intent database.

[0013] Furthermore, historical early warning information includes orbital parameters and intentions, including kinetic interception, close-range reconnaissance, rendezvous and docking, and normal operation.

[0014] Further, step S1 includes the following steps:

[0015] S1-1. Obtain historical early warning information of space targets, and solve the historical early warning information using the SGP4 model / SDP4 model to obtain the solved historical early warning information;

[0016] S1-2. The calculated historical early warning information is then preprocessed to obtain preprocessed historical early warning information. The complete orbital parameters, orbital behavior sequence, and target intent corresponding to the complete orbital behavior sequence of the space target are recorded as the original library construction data.

[0017] Note: Multiple orbital behaviors constitute a sequence of orbital behaviors, which include: Lambert transfer, Hohmann transfer, and low-thrust hovering.

[0018] Furthermore, preprocessing includes wavelet denoising and data transformation.

[0019] Further, possible features include: average relative distance, average relative velocity, rate of change of relative distance modulus, rate of change of relative velocity modulus, predicted nearest distance, predicted nearest velocity, orbital semi-major axis, orbital eccentricity, orbital inclination, relative orbital semi-major axis, relative orbital eccentricity, relative orbital inclination, ellipse fitting error of XY data, and trigonometric function fitting error of YZ data.

[0020] Furthermore, step S3 includes the following:

[0021] A mapping relationship between possible features and target intent is established using the random forest algorithm to obtain a first decision tree random forest model. The first decision tree random forest model includes multiple first decision forests. The input of the first decision tree forest is the possible features, and the output is the target intent corresponding to the possible features. Then, the evaluation criterion is the prediction performance evaluation function value of each intent. The particle swarm optimization algorithm is used to optimize the key feature combination and the number of random forest decision trees in the training of the random forest model. The feature combination with the best classification effect is used as the key feature for target intent recognition.

[0022] Preferably, using the performance evaluation function value of each intention prediction as the evaluation criterion, the particle swarm optimization algorithm is used to optimize the key feature combination for training the random forest model and the number of random forest decision trees, including the following:

[0023] Let the set of independent variables for particle swarm optimization be x = (x1, x2, ..., xn). n ,x tree ), where n is the number of possible features, x i ,i = 1,2,…,n represents whether to select the i-th feature as the key feature, with a value of 0 or 1, x tree The number of decision trees in the random forest is denoted by , with values ​​ranging from (0, 1000). The feature selected from the optimal combination of independent variables that maximizes the intention prediction performance evaluation function value is used as the key feature for target intention recognition.

[0024] The intention prediction effectiveness evaluation function is as follows:

[0025] Value=α1V1+α2V2+α3V3+α4V4+α5V5+α6L+α7n

[0026] In the above formula, α i Let i = 1, 2, ..., 7 be the weights of the variables, satisfying... V1 represents the accuracy rate of identifying close-range reconnaissance intent; V2 represents the accuracy rate of identifying kinetic energy interception intent; V3 represents the accuracy rate of identifying rendezvous and acquisition intent; and V4 represents the accuracy rate of identifying hovering jamming intent. Let L be the average accuracy rate for recognizing four different intentions, and L be the ratio of the number of training features to the total number of features. Normalized values ​​of decision trees for training a random forest model.

[0027] Preferably, step S4 includes the following:

[0028] A second random forest algorithm is used to establish a mapping relationship between target key feature data and target intent, and a spatial target intent library is constructed. The spatial target intent library is a second decision tree random forest model trained by using the key feature combination with the maximum value of the intent prediction performance evaluation function and the number of decision trees. The second decision tree random forest model includes multiple second decision tree forests. The input of the second decision tree forest is the key features of the spatial target, and the output is the spatial target intent corresponding to the key features.

[0029] Preferably, step S5 includes the following:

[0030] Acquire real-time early warning information of space targets, extract key features required for the space target intent database, input the key features into the space target intent database for data processing, and output the orbital behavior intent corresponding to the key features.

[0031] Preferably, both the first decision tree forest and the second decision tree forest include multiple decision trees. Each decision tree includes: a root node at the top level of the decision tree, several internal nodes at the middle level of the decision tree, and several leaf nodes at the bottom level of the decision tree. The root node can be connected to the internal nodes through branches or to the leaf nodes through branches. The root node serves as the input to the decision tree, and the leaf nodes serve as the output of the decision tree.

[0032] Explanation: The root node is located at the top level of the tree and contains all input data samples; internal nodes are located in the middle of the tree. Each internal node contains a portion of the data samples after feature classification from the data samples of the previous node, and classifies the node data through its own node features before passing it to the next node; leaf nodes are located at the bottom level of the tree and represent the results of the data.

[0033] The beneficial effects of this invention are:

[0034] Compared to existing space target intent recognition methods, this invention proposes a method for processing historical early warning information of space targets, extracts key features of typical orbital behaviors of space targets, proposes methods for constructing space target orbital behavior and intent databases, and presents a space target intent recognition algorithm based on random forests. It can extract and identify key features from target orbital data and match them with existing intent databases, achieving accurate on-orbit real-time target intent recognition. This improves the intelligence level of on-orbit spacecraft, making them more suitable for the increasingly complex and rapidly changing space environment of the future, and ensuring their safe operation and reliable functioning. Attached Figure Description

[0035] Figure 1 This is a flowchart of a method for rapid identification of spatial target intent based on early warning information, as described in Embodiment 1.

[0036] Figure 2 This is a wavelet denoising result of the first 400 TLE historical data points after launch and orbit insertion in Example 1;

[0037] Figure 3 This is a relative trajectory diagram of the kinetic energy interception target behavior in Example 1;

[0038] Figure 4 This is the absolute trajectory diagram of the kinetic energy interception target behavior in Example 1;

[0039] Figure 5 This is a relative trajectory diagram of the target's behavior during close-range reconnaissance in Example 1;

[0040] Figure 6 This is the absolute trajectory diagram of the target's behavior during close-range reconnaissance in Example 1;

[0041] Figure 7This is a relative trajectory diagram of the target behavior intended for rendezvous and docking in Example 1;

[0042] Figure 8 This is the absolute trajectory diagram of the target behavior of the rendezvous and docking in Example 1;

[0043] Figure 9 This is a relative trajectory diagram of the intended target behavior during normal operation in Example 1;

[0044] Figure 10 This is the absolute trajectory diagram of the intended target behavior during normal operation in Example 1;

[0045] Figure 11 This is a graph showing the relationship between the model prediction accuracy and the number of trees in the random forest when all features are selected as classification features in Example 1.

[0046] Figure 12 This is a graph showing the relationship between the model prediction accuracy and the number of trees in the random forest when the average relative distance, the rate of change of the relative distance modulus, the relative orbital eccentricity, the predicted nearest distance, the predicted nearest velocity, and the YZ trigonometric function fitting error are selected as classification features. Detailed Implementation

[0047] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this invention, and not all of them. Based on the embodiments of this invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this invention.

[0048] The terminology used in the embodiments of this invention is for the purpose of describing particular embodiments only and is not intended to limit the invention. The singular forms “a,” “the,” and “the” as used in the embodiments of this invention and the appended claims are also intended to include the plural forms, and “multiple” generally includes at least two unless the context clearly indicates otherwise.

[0049] It should be understood that although the terms first, second, third, etc., may be used to describe... in the embodiments of the present invention, these... should not be limited to these terms. These terms are only used to distinguish... For example, first... may also be referred to as second... without departing from the scope of the embodiments of the present invention, and similarly, second... may also be referred to as first...

[0050] Example 1

[0051] This embodiment describes a method for rapid identification of spatial target intent based on early warning information, such as... Figure 1 As shown, it includes the following steps:

[0052] S1. Extract historical orbital data of space targets from historical early warning information and label the target's historical true intentions to obtain the original database construction data. The historical early warning information includes orbital parameters and intentions, including kinetic interception, close-range reconnaissance, rendezvous and docking, and normal operation. This includes the following steps:

[0053] S1-1. Obtain historical early warning information of space targets, and solve the historical early warning information using the SGP4 model / SDP4 model to obtain the solved historical early warning information;

[0054] S1-2. The calculated historical early warning information is then preprocessed to obtain preprocessed historical early warning information. The complete orbital parameters, orbital behavior sequence, and target intent corresponding to the complete orbital behavior sequence of the space target are recorded as the original library construction data.

[0055] The preprocessing steps described above include wavelet denoising and data transformation.

[0056] In this embodiment, the historical early warning information of space targets includes n satellites for each of i satellites. i The TLE data sets are arranged chronologically for each satellite, and the format of each set is shown in the example below:

[0057] Table 1 TLE Data Table

[0058]

[0059] The first line of data contains: 36413 is the space target number given by NORAD, where U indicates unclassified; 10009A is the international designation, where 10 represents 2010, 009 indicates the 9th launch in 2010, and A represents the space target numbered A; 17185.22546559 represents the time point of this TLE data set, where 17 represents 2017, and 185.22546559 represents the 185th day of 2017 at 0.22546559 hours; -.00000097 is the first derivative of the mean motion with respect to time; +00000-0 is the second derivative of the mean motion with respect to time; -93235-5 is the BSTAR drag coefficient; 0 indicates the orbital model, i.e., the SGP4 or SDP4 orbital model is used; 999 represents the data number; and 2 is the check digit. The second line of data: 063.3979 is the orbital inclination; 147.6704 is the right ascension of the ascending node; 0248459 is the fractional part of the orbital eccentricity; 004.1465 is the argument of perigee; 356.1523 is the mean perigee; 13.45166106 is the number of orbits the satellite makes around the Earth per day; 36021 is the number of orbits since launch; 7 is still a check digit.

[0060] Random errors often exist in track measurement data, so it is necessary to filter these noisy data. In the data preprocessing layer, the "wavelet denoising" method is used to find the best approximation of the sample data in order to remove the noisy data.

[0061] Figure 2 This is the wavelet denoising result of the first 400 TLE historical data points after "TIANHUI" was launched into orbit: The TLE data in the figure represents the semi-major axis variation deconstructed through the data inversion layer. The threshold selection adopts a fixed threshold principle and is adjusted according to the noise level estimate of the first layer wavelet decomposition. It can be seen that after wavelet denoising, the noise in the original TLE data is effectively filtered out.

[0062] The denoised TLE data will be processed and converted into orbital parameters X consisting of 25 variables. i :

[0063]

[0064] In the above formula, Δt i The forecast duration for this point is given by: a is the semi-major axis, e is the eccentricity, i is the orbital inclination, Ω is the right ascension of the ascending node, ω is the argument of perigee, M is the mean perigee angle, r is the satellite's position vector in the inertial frame, v is the satellite's velocity vector in the inertial frame, the subscripts x, y, z represent the three-axis components in the inertial frame, the prescript Δ represents the difference between the error orbit and the accurate orbit, and the subscript i represents the state number.

[0065] In step S1, the time interval between each set of TLE data for each satellite is relatively long, and the orbital parameters vary greatly, which is not conducive to extracting features to describe the target orbital behavior. Therefore, it is necessary to use numerical prediction methods to calculate the satellite orbital parameters X between the two sets of TLE data. ij , where j is the j-th predicted data point of the i-th TLE data prediction, and the numerical prediction method is the fourth-order Runge-Kutta method.

[0066] Figure 3 , Figure 4 , Figure 5 , Figure 6 , Figure 7 , Figure 8 , Figure 9 , Figure 10 These are inertial space target motion trajectory diagrams drawn after TLE data processing for close-range reconnaissance, kinetic interception, rendezvous and capture, and hovering interference trajectory behavior, respectively.

[0067] S2. Analyze and extract possible features from the orbital data. Obtain possible feature values ​​by performing coordinate transformation and calculation on the original database construction data. Use these possible feature values ​​as feature database construction data. Store the feature database construction data in chronological order to obtain the feature database, which includes the following:

[0068] Orbital data can be categorized into two parts: absolute features and relative features. Absolute features refer to the orbital parameters of the target spacecraft described in the geocentric inertial coordinate system, including the semi-major axis, eccentricity, and inclination. Relative features refer to the orbital parameters of the target spacecraft described in the first orbital coordinate system of our satellite, including average relative distance, average relative velocity, rate of change of relative distance modulus, rate of change of relative velocity modulus, predicted closest distance, predicted closest velocity, relative orbital semi-major axis, relative orbital eccentricity, relative orbital inclination, ellipse fitting error for XY data, and trigonometric function fitting error for YZ data.

[0069] In this embodiment, the close-in reconnaissance data in S1 is processed. This data contains 60 satellite orbital parameter data points. The input is 60 satellite orbital parameter data, and the output is the average value of the possible feature values ​​calculated using the 60 orbital parameters, which is used as the orbital parameter feature of this data.

[0070] S3. Establish a mapping relationship between the feature database and the target intent using a supervised classification algorithm, optimize feature selection using the particle swarm optimization algorithm, extract key features for target intent recognition, and store the key feature data in chronological order to obtain a key feature database, including the following:

[0071] The random forest algorithm is used to establish a mapping relationship between possible features and target intent, resulting in a first decision tree random forest model. The first decision tree random forest model includes multiple first decision forests. The input of the first decision tree forest is the possible features, and the output is the target intent corresponding to the possible features. Then, the evaluation criterion is the prediction performance evaluation function value of each intent. The particle swarm optimization algorithm is used to optimize the key feature combination and the number of random forest decision trees in the training of the random forest model. The feature combination with the best classification effect is used as the key feature for target intent recognition.

[0072] In this embodiment, by classifying the target intent based on all possible feature data obtained in step (2), the mapping relationship between the range of feature parameter values ​​and the orbital behavior intent can be obtained. The Gini coefficient is selected as the decision tree splitting criterion. The formula for calculating the Gini coefficient is:

[0073]

[0074] In the above formula, K represents the number of data types in the sample dataset D, and p k This represents the probability of the k-th data type appearing. In practical applications, the frequency of the k-th data type is generally used. Replace p k ,|C k | indicates the frequency of occurrence of the k-th data type.

[0075] First, the feature library obtained in step S2 is sampled using Bootstrap to obtain multiple training sets. Each training set can be used to train a decision tree. The construction of the decision tree uses a recursive method to select the optimal feature of each node to partition the training data samples. The specific steps are as follows:

[0076] Constructing the root node: Initially, all data samples are located at the root node.

[0077] Determine the optimal features of a node: Split the sample data within the root node into multiple optimal child nodes based on the Gini coefficient. If the classification result of a child node is less than the splitting threshold of the Gini coefficient, then the node is a leaf node and the splitting stops; otherwise, continue splitting. The classification threshold of the Gini coefficient is given by expert experience, and an appropriate threshold is set manually.

[0078] Recursive training and pruning: A recursive algorithm is used to classify the sample data of the root node until all nodes meet the evaluation criteria based on the Gini coefficient. To avoid the decision tree overfitting the data, the decision tree is pruned during or after training.

[0079] Once all training sets have been trained, the resulting decision trees can be combined to obtain a random forest classifier. The trained random forest model is then tested using feature data from the test set to evaluate its prediction accuracy.

[0080] Bootstrap abstract sampling is a method for generating decision trees by randomly selecting a subset of samples from the training set. The algorithm is trained in multiple rounds, with each round's training set consisting of m training samples randomly selected from the initial training set. A given initial training sample may appear multiple times or not at all in a particular round's training set. After training, a sequence of prediction functions is obtained. The final prediction function H uses a voting method for classification problems and a simple averaging method for regression problems to discriminate new samples.

[0081] Using the performance evaluation function value of each intention prediction as the evaluation criterion, the particle swarm optimization algorithm is used to optimize the key feature combination and the number of random forest decision trees during random forest model training, including the following:

[0082] Let the set of independent variables for particle swarm optimization be x = (x1, x2, ..., xn). n ,x tree ), where n is the number of possible features, x i ,i = 1,2,…,n represents whether to select the i-th feature as the key feature, with a value of 0 or 1, x tree The number of decision trees in the random forest is denoted by , with values ​​ranging from (0, 1000). The feature selected from the optimal combination of independent variables that maximizes the intention prediction performance evaluation function value is used as the key feature for target intention recognition.

[0083] The intention prediction effectiveness evaluation function is as follows:

[0084] Value=α1V1+α2V2+α3V3+α4V4+α5V5+α6L+α7n

[0085] In the above formula, α i Let i = 1, 2, ..., 7 be the weights of the variables, satisfying... V1 represents the accuracy rate of identifying close-range reconnaissance intent; V2 represents the accuracy rate of identifying kinetic energy interception intent; V3 represents the accuracy rate of identifying rendezvous and acquisition intent; and V4 represents the accuracy rate of identifying hovering jamming intent. Let L be the average accuracy rate for recognizing four different intentions, and L be the ratio of the number of training features to the total number of features. Normalized values ​​of decision trees for training a random forest model.

[0086] Figure 11 This relates the model's prediction accuracy to the number of trees in the random forest when all features are selected as classification features. Figure 12 This relates the model prediction accuracy to the number of trees in the random forest when the average relative distance, the rate of change of the relative distance magnitude, the relative orbital eccentricity, the predicted closest distance, the predicted closest velocity, and the YZ trigonometric function fitting error are selected as classification features. As can be seen from the figure, selecting... Figure 12 The middle feature can effectively improve the classification accuracy, that is, the feature is the key feature for classification.

[0087] S4. A spatial target intent database is constructed by establishing a mapping relationship between key feature databases and target intents through supervised classification algorithms. This mapping relationship includes the following:

[0088] A second random forest algorithm is used to establish a mapping relationship between target key feature data and target intent, constructing a spatial target intent database. This database is a second decision tree random forest model trained using the key feature combination that maximizes the intent prediction performance evaluation function and the number of decision trees. The second decision tree random forest model comprises multiple second decision tree forests. The input to each second decision tree forest is the key features of the spatial target, and the output is the spatial target intent corresponding to those key features.

[0089] In this embodiment, step S4 is the same as step S3 in terms of specific methods, but the classification features of step S4 are the output data of step S3. After random forest classification, the spatial target intent library can be obtained.

[0090] S5. Real-time warning information intent recognition is achieved through a spatial target intent database, including the following:

[0091] The system acquires real-time early warning information of space targets, extracts key features required for the space target intent database, inputs the key features into the space target intent database for data processing, and outputs the orbital behavior intent corresponding to the key features.

[0092] In this embodiment, when real-time warning information is obtained, the warning information is processed through wavelet transform and data conversion, and then input into the trained intent library to output the spatial target intent.

[0093] In the above method, both the first decision tree forest and the second decision tree forest include multiple decision trees. Each decision tree includes: a root node at the top level of the decision tree, several internal nodes at the middle level of the decision tree, and several leaf nodes at the bottom level of the decision tree. The root node can be connected to the internal nodes through branches or to the leaf nodes through branches. The root node serves as the input to the decision tree, and the leaf nodes serve as the output of the decision tree.

Claims

1. A method for rapid identification of spatial target intent based on early warning information, characterized in that, Includes the following steps: S1. Extract historical orbital data of space targets using historical early warning information and label the target's historical true intentions to obtain the original library construction data; S2. Analyze and extract possible features from the orbital data. Obtain possible feature values ​​by performing coordinate transformation and calculation on the original database construction data. Use these possible feature values ​​as feature database construction data. Store the feature database construction data in chronological order to obtain a feature database. The possible features include: average relative distance, average relative velocity, rate of change of relative distance modulus, rate of change of relative velocity modulus, predicted closest distance, predicted closest velocity, orbital semi-major axis, orbital eccentricity, orbital inclination, relative orbital semi-major axis, relative orbital eccentricity, relative orbital inclination, ellipse fitting error for XY data, and trigonometric function fitting error for YZ data. S3. Establish a mapping relationship between the feature database and the target intent through a supervised classification algorithm, optimize feature selection using the particle swarm optimization algorithm, extract key features for target intent recognition, and store the key feature data in chronological order to obtain a key feature database; including the following: establish a mapping relationship between possible features and target intent using the random forest algorithm to obtain a first decision tree random forest model. The first decision tree random forest model includes multiple first decision forests. The input of the first decision tree forest is the possible features, and the output is the target intent corresponding to the possible features. Then, using the evaluation function value of each intent prediction performance as the evaluation standard, the particle swarm optimization algorithm is used to optimize the key feature combination and the number of random forest decision trees in the training of the random forest model, and the feature combination with the best classification effect is used as the key feature for target intent recognition; The evaluation criterion for using the prediction performance assessment function value of each intention is to optimize the key feature combination and the number of decision trees in the random forest model training using the particle swarm optimization algorithm, including the following: Let the set of independent variables for particle swarm optimization be . ,in n The number of possible features, To choose whether to select the first i Each feature is used as a key feature, taking a value of 0 or 1. The number of decision trees in the random forest, with values ​​ranging from 1 to 2. The features selected from the optimal combination of independent variables that maximizes the value of the intent prediction effectiveness evaluation function are used as the key features for target intent recognition. The intention prediction effectiveness evaluation function is as follows: In the above formula, Let the weights of the variables satisfy... , To improve the accuracy of identifying close-range reconnaissance intentions, To improve the accuracy of kinetic energy interception intent recognition To improve the accuracy of intersection capture intent recognition, To improve the accuracy of hovering interference intent recognition The average accuracy rate for identifying the four intents. The ratio of the number of classification features to the total number of features is used to train the model. Normalized values ​​for the decision trees used in the random forest model; S4. Construct a spatial target intent library by establishing a mapping relationship between a key feature database and target intent through a supervised classification algorithm, thereby establishing a mapping relationship between key features and target intent; including the following: secondarily using the random forest algorithm to establish a mapping relationship between target key feature data and target intent, constructing a spatial target intent library, wherein the spatial target intent library is a second decision tree random forest model trained using the key feature combination when the intent prediction performance evaluation function value is maximized and the number of decision trees, the second decision tree random forest model includes multiple second decision tree forests, the input of the second decision tree forest is the key features of the spatial target, and the output is the spatial target intent corresponding to the key features; S5. Real-time warning information intent recognition is achieved through a spatial target intent database.

2. The method for rapid identification of spatial target intent based on early warning information as described in claim 1, characterized in that, The historical early warning information includes orbital parameters and intentions, including kinetic interception, close-range reconnaissance, rendezvous and docking, and normal operation.

3. The method for rapid identification of spatial target intent based on early warning information as described in claim 1, characterized in that, Step S1 includes the following steps: S1-1. Obtain historical early warning information of space targets, and solve the historical early warning information using the SGP4 model / SDP4 model to obtain the solved historical early warning information; S1-2. The calculated historical early warning information is then preprocessed to obtain preprocessed historical early warning information. The complete orbital parameters, orbital behavior sequence, and target intent corresponding to the complete orbital behavior sequence of the space target are recorded as the original library construction data.

4. The method for rapid identification of spatial target intent based on early warning information as described in claim 3, wherein the preprocessing includes: Wavelet denoising and data transformation.

5. The method for rapid identification of spatial target intent based on early warning information as described in claim 1, characterized in that, Step S5 includes the following: Acquire real-time early warning information of space targets, extract key features required for the space target intent database, input the key features into the space target intent database for data processing, and output the orbital behavior intent corresponding to the key features.

Images