An intelligent detection method for abnormal spacecraft behavior

By establishing an intelligent detection method for spacecraft behavior anomalies based on feature dimensionality enhancement, and utilizing a convolutional neural network structure to detect spacecraft orbital anomalies, the problem of insufficient real-time performance and confidence level in existing technologies is solved, and efficient orbital anomaly detection is achieved.

CN113780332BActive Publication Date: 2024-11-19CHINESE PEOPLES LIBERATION ARMY UNIT 63768
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202110638651.3
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2021-06-08
Publication Date
2024-11-19
Estimated Expiration
2041-06-08

AI Technical Summary

Technical Problem

Existing technologies are insufficient for real-time, high-confidence detection of spacecraft orbital anomalies, and are also insufficient for revealing the physical changes in a spacecraft's orbit during its operation in orbit.

Method used

A spacecraft behavior anomaly detection method based on feature dimensionality enhancement is established. A high-dimensional representation matrix is ​​generated by describing orbital parameters, and a convolutional neural network structure that integrates multi-dimensional feature classification and detection is designed to automatically learn orbital anomaly behavior features.

Benefits of technology

It improved the success rate of orbital anomaly detection from 85% to 98.5%, and is effectively used for detecting anomalies in spacecraft orbital behavior.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN113780332B_ABST
    Figure CN113780332B_ABST
Patent Text Reader

Abstract

The present invention provides an intelligent detection method for spacecraft orbit anomalies, which has quasi-real-time and high confidence capabilities and can be effectively used to judge the orbital behavior of spacecraft. It includes: extracting the time series characteristics of spacecraft behavior based on the target orbital element database and generating a high-dimensional representation matrix; designing a convolutional neural network structure that integrates multi-dimensional feature classification detection; automatically learning the characteristics of orbital abnormality behavior to detect whether the spacecraft orbit is abnormal. The recognition rate, accuracy and real-time indicators of this method meet the requirements of engineering applications and can provide effective support for the long-term safe and stable operation of spacecraft.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention belongs to the field of data recognition technology, and relates to the rapid detection and identification of abnormal spacecraft behavior. It can be used for timely re-tracking of spacecraft and analysis of abnormal factors, and provide effective support for the long-term safe and stable operation of spacecraft. Background Art

[0002] Spacecraft orbit anomalies usually refer to changes in orbital elements caused by abnormal perturbations. There are three reasons for orbit anomalies: one is human-controlled maneuvers, such as correcting the orbit to overcome perturbations; the second is orbit changes caused by non-human control, such as collisions with space objects, explosions and disintegration of spacecraft; and the third is orbit anomalies caused by the space environment.

[0003] It is of great significance to timely and accurately detect and obtain and analyze the factors that may cause the abnormality of the spacecraft with orbital anomalies, and to re-track the target after the orbit change, which will provide an effective basis for the safe and stable operation of the spacecraft. Therefore, the detection of spacecraft orbital anomalies has also attracted widespread attention from scholars at home and abroad, and a variety of detection methods have been proposed in recent years. For example, Patera et al. of Aerospace Company of the United States applied the moving window curve fitting technology to space event detection. This method can remove the changes caused by orbital perturbations and noise, respond to non-periodic changes in parameter values, and achieve faster detection of orbital anomalies; Zhang Zhenjun et al. of Beijing University of Aeronautics and Astronautics proposed a method based on wavelet analysis, which can provide a better step signal recognition method by extracting characteristic signals from noise; Yang Xu et al. of the National Astronomical Observatory of the Chinese Academy of Sciences used a comprehensive criterion method of three parameters: orbital semi-major axis, absolute value of semi-major axis change, and discrete ratio of orbital average motion, which can effectively reduce the misjudgment rate. Huang et al. used the Empirical Mode Decomposition (EMD) method to decompose the non-stationary signal into a finite number of intrinsic mode functions of different characteristic scales, and adaptively obtain abnormal point and time information from the data. The above detection method is basically to process the acquired spacecraft orbit data, and to judge the orbit anomaly by analyzing and extracting its characteristic quantities, which is difficult to reveal the physical change law of the orbit when the spacecraft is in orbit. Summary of the invention

[0004] In view of the shortcomings of current research methods and the need for quasi-real-time and high-confidence judgment of spacecraft abnormal approach symptoms, the purpose of the present invention is to propose an intelligent detection method for spacecraft behavior anomalies based on feature dimensionality upgrade, and to establish a unified maneuver detection model by describing orbital parameters to detect whether the spacecraft orbit is abnormal. The model extracts the time series characteristics of spacecraft behavior based on the target orbital element database and generates a high-dimensional representation matrix. It designs a convolutional neural network structure that integrates multi-dimensional feature classification detection and can automatically learn the characteristics of orbital abnormal behavior. The orbital anomaly detection success rate of the intelligent detection method of the present invention on the self-established test set is increased to 98.5% compared with 85% of the Mahalanobis distance method, and can be effectively used for the detection of spacecraft orbital behavior anomalies.

[0005] In order to achieve the above object, the present invention adopts the following technical solution to achieve it.

[0006] An intelligent detection method for abnormal spacecraft behavior comprises the following steps:

[0007] Step 1: Establish a unified detection model for in-plane maneuvers and out-of-plane maneuvers;

[0008] Step 2, selecting characteristic orbital parameters for spacecraft orbit anomaly monitoring;

[0009] Step 3: Predict t k The orbit of time (r,v) k and the characteristic orbital parameter σ k ;

[0010] Step 4, obtaining the deviation sequence of characteristic orbital parameters;

[0011] Step 5, using 9 consecutive deviation sequences to form a deviation matrix;

[0012] Step 6, designing an intelligent detection network for abnormal orbital behavior of spacecraft;

[0013] Step 7: Classify and detect the two orbital behaviors of the spacecraft.

[0014] The characteristics and further improvements of the present invention are:

[0015] (1) Establish a unified detection model for in-plane and out-of-plane maneuvers

[0016] (1a) The main characterization parameters of the in-plane maneuver are the orbital semi-major axis and eccentricity;

[0017] (1b) The main characterization parameters of the out-of-plane maneuver are selected as orbital inclination and right ascension of ascending node;

[0018] (2) Selecting and calculating characteristic orbital parameters for spacecraft orbit anomaly detection

[0019] (2a) Read the epoch time as tk-1 Two-row track element (TLE) data σ k-1 , as orbit characteristic parameters; in the characteristic parameters, a is the semi-major axis, e is the eccentricity, i is the inclination, Ω is the right ascension of the ascending node, ω is the argument of perigee, and M is the mean anomaly;

[0020] (2b) The orbital energy ε, orbital angular momentum H and average angular velocity n are calculated from the selected orbital parameters as the other three orbital characteristic parameters. The specific calculation method is as follows:

[0021]

[0022] H=r×v

[0023]

[0024] (3) Forecast t k The orbit of time (r,v) k and the characteristic orbital parameter σ k

[0025] (3a) Read the epoch time as t k-1 Two-row track element (TLE) data σ k-1 , as the orbital characteristic parameter.

[0026] (3b) The long-term forecast of the SGP4 / SDP4 orbit prediction model is used to obtain t k The expected value of the long-term change of the orbital elements at time k / k-1 Among them, the parameters predicted by the model include σ=(t,n,e,i,ω,Ω,M,B * ), the subscript “k / (k-1)” indicates the time from epoch t k-1 The predicted time t k Characteristic orbital parameters and trajectories.

[0027] (4) Obtaining the deviation sequence of characteristic orbital parameters

[0028] (4a) Calculate the difference between the cataloged value of the characteristic orbital element at the same time and the predicted value in step (3);

[0029] (4b) Obtain the deviation sequence {t k ,Δε k ,Δe k ,Δi k ,ΔΩ k ,Δ|H| k ,Δn k ,Δ|R| k ,Δ|T| k ,Δ|N| k k=1,2,3,…,N

[0030] The specific method is as follows:

[0031] Δε k =ε k -ε k / (k-1) ,k=1,2,3,…,N

[0032] Δe k =e k -e k / (k-1) ,k=1,2,3,…,N

[0033] Δi k =i k -i k / (k-1) ,k=1,2,3,…,N

[0034] ΔΩ k =Ω k -Ω k / (k-1) ,k=1,2,3,…,N

[0035] Δ|H| k =|H| k -|H| k / (k-1) ,k=1,2,3,…,N

[0036] Δn k =n k -n k / (k-1) ,k=1,2,3,…,N

[0037] Δ|R| k =|R| k -|R| k / (k-1) ,k=1,2,3,…,N

[0038] Δ|T| k =|T| k -|T| k / (k-1) ,k=1,2,3,…,N

[0039] Δ|N| k =|N| k -|N| k / (k-1) ,k=1,2,3,…,N

[0040] (5) Using 9 consecutive deviation sequences to form a deviation matrix;

[0041] (5a) Read two lines of orbital element (TLE) data for nine consecutive times from the same spacecraft and calculate the deviation characteristic sequence;

[0042] (5b) Generate a feature parameter deviation matrix, and normalize the vector to transform it into a two-dimensional feature parameter deviation matrix as follows:

[0043]

[0044] (6) Design of an intelligent detection network for abnormal orbital behavior of spacecraft

[0045] (6a) The intelligent detection network used in the present invention includes 1 input layer, 1 track feature calculation layer, 1 track feature residual calculation layer, 1 feature dimension enhancement layer, 2 DCN layers (including 2 convolutional layers, 2 relu activation function layers, 1 Dropout layer and 1 maximum pooling layer), 1 flat layer, 1 hidden layer and 1 output layer (Softmax regression layer).

[0046] (6b) The specific settings of the intelligent detection network are as follows: the input layer is the orbital elements obtained from the spacecraft catalog; the orbital feature calculation layer is to calculate the feature parameter vector corresponding to the current orbit and the most recent 8 orbital elements in history; the orbital feature residual calculation layer is to calculate the adjacent change residual vector based on the orbital behavior feature vector obtained by the orbital feature calculation layer; the feature dimension upgrade layer is to upgrade the spacecraft orbital behavior feature residual vector into a two-dimensional feature parameter deviation matrix after normalization transformation; the convolution kernel size of DCN1 layer 1 is 3×3, the number of convolution kernels becomes 32, the convolution kernel size of convolution layer 2 is 3×3, the number of convolution kernels becomes 64, and the sampling size of pooling layer 1 is 2×2 non-overlapping sampling; the convolution kernel size of DCN2 layer 1 is 1×1, the number of convolution kernels becomes 128, the convolution kernel size of convolution layer 2 is 1×1, the number of convolution kernels becomes 256, and the sampling size of pooling layer 2 is 2×2 non-overlapping sampling. The number of neurons in the flat layer is set to 1024, and the number of neurons in the hidden layer is set to 1000 to classify the features output by the output layer.

[0047] (7) Classification and detection of two types of spacecraft orbital behaviors

[0048] (7a) Using the intelligent detection network designed in step 6, classify the features output by the output layer;

[0049] (7b) The spacecraft orbital behavior state is classified into two detection results, namely 0 (spacecraft orbital behavior is normal) and 1 (spacecraft orbital behavior is abnormal). BRIEF DESCRIPTION OF THE DRAWINGS

[0050] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0051] Figure 1It is a flow chart of the steps of the present invention;

[0052] Figure 2 It is a flow chart of intelligent detection of abnormal orbital behavior of spacecraft;

[0053] Figure 3 is a curve diagram of changes in typical orbital parameters (including semi-major axis a, inclination i, eccentricity e, and translation n) of a spacecraft according to an embodiment of the present invention;

[0054] Figure 4 is a curve diagram of the variation of residual parameters (including U-direction component, V-direction component, W-direction component, and R-direction component) in the UVW coordinate system of the spacecraft according to an embodiment of the present invention;

[0055] Figure 5 is a characteristic parameter vector of a certain orbit segment extracted by the spacecraft in an embodiment of the present invention;

[0056] Figure 6 is an orbital characteristic parameter matrix extracted from TLE data updated nine times continuously by the spacecraft in the embodiment of the present invention;

[0057] Figure 7 It is a schematic diagram of a behavior anomaly detection network designed by the present invention;

[0058] Figure 8 This is a typical sample example of abnormal spacecraft orbital behavior;

[0059] Fig. 9 This is an example diagram of a typical sample of normal spacecraft orbital behavior;

[0060] Fig.10 It is the intelligent error convergence curve diagram of spacecraft orbital behavior anomaly detection;

[0061] Fig.11 It is a curve chart of the accuracy of intelligent detection of abnormal orbital behavior of spacecraft;

[0062] Fig.12 This is a comparison chart of the results of using intelligent anomaly detection and Mahalanobis distance to detect abnormal spacecraft behavior;

[0063] Fig.13 A sample set table for detecting abnormal track behavior;

[0064] Fig.14 Table for setting hyperparameters for spacecraft behavior anomaly detection network;

[0065] Fig.15 Confusion matrix table for detecting anomalies in spacecraft orbital behavior;

[0066] Fig.16 Confusion matrix table for intelligent detection of track behavior anomalies in the test sample set;

[0067] Fig.17 Confusion matrix table of the Mahalanobis distance detection method for track behavior anomalies in the test sample set. DETAILED DESCRIPTION

[0068] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0069] Reference Figure 2 The specific implementation steps of the present invention are as follows:

[0070] (1) Read the continuous TLE data of the spacecraft and extract orbital parameters including epoch time, eccentricity, inclination, right ascension of ascending node, etc. σ=(t,n,e,i,ω,Ω,M,B * ), and calculate orbital energy, orbital angular momentum, Kepler translation, and spacecraft R, T, and N direction vectors as characteristic orbital parameters for anomaly detection.

[0071] (2) Using t k-1 The TLE orbital parameter σ at time k-1 , orbit prediction is done by SGP4 / SDP4 model, and t is calculated k Time track (r,v) k / k-1 and the characteristic orbital parameter σ k / k-1 ;

[0072] (3) Subtract the cataloged characteristic orbit parameters from the predicted characteristic parameters to obtain the deviation characteristic sequence {t k ,Δε k ,Δe k ,Δi k ,ΔΩ k ,Δ|H| k ,Δn k ,Δ|R| k ,Δ|T| k ,Δ|N| k}k=1,2,3,…,N, each item in the characteristic sequence is calculated by the following formula:

[0073] Δε k =ε k -ε k / (k-1) ,k=1,2,3,…,N

[0074] Δe k =e k -e k / (k-1) ,k=1,2,3,…,N

[0075] Δi k =i k -i k / (k-1) ,k=1,2,3,…,N

[0076] ΔΩ k =Ω k -Ω k / (k-1) ,k=1,2,3,…,N

[0077] Δ|H| k =|H| k -|H| k / (k-1) ,k=1,2,3,…,N

[0078] Δn k =n k -n k / (k-1) ,k=1,2,3,…,N

[0079] Δ|R| k =|R| k -|R| k / (k-1) ,k=1,2,3,…,N

[0080] Δ|T| k =|T| k -|T| k / (k-1) ,k=1,2,3,…,N

[0081] Δ|N| k =|N| k -|N| k / (k-1) ,k=1,2,3,…,N

[0082] The characteristic parameters and residual error curves of each component of the characteristic sequence of typical spacecraft are shown in the attached Figure 3 and attached Figure 4 As shown;

[0083] (4) Generate the deviation feature vector based on the result calculated in (3), as shown in the following figure: Figure 5 The orbital characteristic parameters are generated by extracting TLE with 9 consecutive updates, as shown in the attached figure. Figure 6 As shown;

[0084] (5) The intelligent detection network designed by the present invention is used to perform abnormal detection and intelligent judgment on the dimension-raising spectrum. The abnormal intelligent detection network includes 1 input layer, 1 track feature calculation layer, 1 track feature residual calculation layer, 1 feature dimension-raising layer, 2 DCN layers (including 2 convolution layers, 2 relu activation function layers, 1 Dropout layer and 1 maximum pooling layer), 1 flat layer, 1 hidden layer and 1 output layer (Softmax regression layer), as shown in the attached figure. Figure 7 shown.

[0085] The effect of the present invention is further illustrated by the following test:

[0086] Experiment 1: Simulation image generation and operation speed comparison:

[0087] 1. Experimental scenario:

[0088] In this experiment, 200 spacecraft data released by spacetrack with continuous TLE updates within 5 years were selected to mark the orbital characteristic parameters corresponding to abnormal and normal orbital behaviors of spacecraft to generate sample sets. Among them, the number of orbital behavior abnormal samples in the training set is 866, and the number of orbital stable samples is 1360; the number of orbital stable samples in the test set is 445, and the number of orbital maneuver samples is 187, see the attached Fig.13 Typical orbital behavior abnormal and normal matrix samples are shown in the attached Figure 8 and attached Fig. 9 shown.

[0089] 2. Simulation content:

[0090] In this experiment, the proposed intelligent detection network is used to train the data set, that is, the training data set samples are 2226 and the test data set samples are 632. Both the training set and the test set are randomly generated. The hyperparameter settings of the optimizer, ReLU slope, batch size, number of iterations, and weight initialization in the intelligent detection network are shown in the attached Fig.14 shown.

[0091] 3. Analysis of simulation results:

[0092] In order to accurately analyze and evaluate the detection algorithm used in the present invention, three indicators, namely accuracy, recall rate and false alarm rate, are introduced in the experiment. The specific calculation method is as follows:

[0093]

[0094]

[0095]

[0096] Among them, T is the number of orbital elements in the sample that contain abnormal spacecraft orbital behavior; F is the number of orbital elements in the sample that contain normal spacecraft orbital behavior. TT and FT represent the number of orbital elements in the sample with abnormal spacecraft orbital behavior, the number of orbital elements identified as abnormal spacecraft orbital behavior and the number of orbital elements in the sample with normal spacecraft orbital behavior, and the number of arc segments identified as attitude tumbling and attitude stability. In addition, the detection confusion matrix is ​​defined based on the above data, see Appendix Fig.15 shown.

[0097] The intelligent detection network proposed in this invention is used to test realistic samples. The accuracy convergence curve and error convergence curve of typical sample sets are shown in the attached figure. Fig.10 and attached Fig.11 As shown in the figure, the algorithm converges to a stable state quickly after about 25 generations. The confusion matrix results show that there are only 9 samples with incorrect recognition among the 632 test samples. See the attached Fig.16 shown.

[0098] In order to verify the classification and recognition effect, this experiment also used the Mahalanobis distance detection method to perform classification detection on the same data set as a comparison result. Fig.17 The detection confusion matrix using the Mahalanobis distance method is shown in the figure. Fig.12 The comparison results of the accuracy, recall rate and false alarm rate of the two methods are shown in FIG. The results show that the intelligent detection method used in the present invention is significantly better than the Mahalanobis distance detection method.

[0099] The computer used in the present invention uses an Intel Core (TM) i5-3470 CPU 3.20 GHz processor.

[0100] The above is only a specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art can easily think of changes or substitutions within the technical scope disclosed by the present invention, which should be included in the protection scope of the present invention. Therefore, the protection scope of the present invention should be based on the protection scope of the claims.

Claims

1. An intelligent detection method for abnormal spacecraft behavior, characterized in that: The intelligent detection method of abnormal spacecraft behavior includes the following steps: Step 1: Establish a unified detection model for in-plane maneuvers and out-of-plane maneuvers; Step 2, selecting orbital characteristic parameters for spacecraft orbit anomaly monitoring; Step 3: Predict t k The orbit of time (r,v) k and orbital characteristic parameter σ k ; Step 4, obtaining the deviation sequence of orbital characteristic parameters; Step 5, using 9 consecutive deviation sequences to form a deviation matrix; Step 6, designing an intelligent detection network for abnormal orbital behavior of spacecraft; Step 7, classify and detect the two orbital behaviors of the spacecraft; The step 2 includes the following sub-steps: (3a) Read the epoch time as t k-1 Two rows of orbital element TLE data σ k-1 , as orbital characteristic parameters; Among the characteristic parameters, a is the semi-major axis, e is the eccentricity, i is the inclination, Ω is the right ascension of the ascending node, ω is the argument of perigee, and M is the mean anomaly; (3b) The orbital energy ε, orbital angular momentum H and average angular velocity n are calculated from the selected orbital parameters as the other three orbital characteristic parameters. The specific calculation method is as follows: H=r×v The step 3 includes the following sub-steps: (4a) Read the epoch time as t k-1 Two rows of orbital element TLE data σ k-1 , as orbital characteristic parameters; (4b) The long-term forecast of the SGP4 / SDP4 orbit prediction model is used to obtain t k The expected value of the long-term change of the orbital elements at time k / k-1 , where the model prediction parameters include σ=(t,n,e,i,ω,Ω,M,B * ), the subscript "k / (k-1)" indicates the time from epoch t k-1 The predicted time t k Orbital characteristic parameters and trajectories; The step 4 includes the following sub-steps: (5a) Calculate the difference between the cataloged value of the orbital characteristic element at the same time and the predicted value in step (3); (5b) Obtaining the deviation sequence {t k ,No k ,No k ,No k ,LOOK k ,D|H| k ,Δn k ,Δ|R| k ,Δ|T| k ,D|N| k },k=1,2,3,···,N specific method as follows: No k =e k -e k / (k-1) , Δe k =e k -e k / (k-1) Δi k =i k -i k / (k-1) DO k =Oh k -Oh k / (k-1) Δ|H| k =|H| k -|H| k / (k-1) Δn k =n k -n k / (k-1) Δ|R| k =|R| k -|R| k / (k-1) Δ|T| k =|T| k -|T| k / (k-1) D|N| k =|N| k -|N| k / (k-1) ; The step 5 comprises the following sub-steps: (6a) Read two lines of orbital element TLE data for nine consecutive times from the same spacecraft and calculate the deviation characteristic sequence; (6b) Generate a feature parameter deviation matrix and normalize the vector to upgrade it to a two-dimensional feature parameter deviation matrix:

2. The intelligent detection method for abnormal spacecraft behavior according to claim 1, characterized in that: The step 1 comprises the following sub-steps: (2a) The main characterization parameters of the in-plane maneuver are the orbital semi-major axis and eccentricity; (2b) The main characterization parameters of the out-of-plane maneuver are orbital inclination and right ascension of the ascending node.

3. The intelligent detection method for abnormal spacecraft behavior according to claim 1, characterized in that: The step 1 comprises the following sub-steps: (7a) The intelligent detection network used includes 1 input layer, 1 track feature calculation layer, 1 track feature residual calculation layer, 1 feature dimension enhancement layer, 2 DCN layers, 1 flat layer, 1 hidden layer and 1 output layer Softmax regression layer, wherein the DCN layer includes 2 convolutional layers, 2 relu activation function layers, 1 Dropout layer and 1 maximum pooling layer; (7b) The specific design method is as follows: the input layer is the orbital elements obtained from the spacecraft catalog; the orbital feature calculation layer is to calculate the characteristic parameter vector corresponding to the current orbit and the orbital elements of the most recent 8 times in history; the orbital feature residual calculation layer is to calculate the adjacent change residual vector based on the orbital behavior feature vector obtained by the orbital feature calculation layer; the feature dimension upgrade layer is to upgrade the spacecraft orbital behavior feature residual vector to a two-dimensional feature parameter deviation matrix after normalization transformation; the convolution kernel size of the DCN1 layer 1 is 3×3, the number of convolution kernels becomes 32, and the convolution kernel size of the convolution layer 2 is 3×3. , the number of convolution kernels is changed to 64, and the sampling size of pooling layer 1 is 2×2 non-overlapping sampling; the convolution kernel size of DCN2 layer 1 is 1×1, the number of convolution kernels is changed to 128, the convolution kernel size of convolution layer 2 is 1×1, the number of convolution kernels is changed to 256, and the sampling size of pooling layer 2 is 2×2 non-overlapping sampling; the number of neurons in the flat layer is set to 1024, and the number of neurons in the hidden layer is set to 1000. The features output by the output layer are classified, and the two orbital behavior state classification detection results are obtained: 0 spacecraft orbital behavior is normal and 1 spacecraft orbital behavior is abnormal.

4. The intelligent detection method for abnormal spacecraft behavior according to claim 1, characterized in that: The step 7 comprises the following sub-steps: (8a) Using the intelligent detection network designed in step 6, classify the features output by the output layer; (8b) The spacecraft orbital behavior status is classified into two detection results, namely, 0: normal spacecraft orbital behavior and 1: abnormal spacecraft orbital behavior.