A method for detecting impulsive maneuvers of a non-cooperative space object based on support vector machine

Abstract

The application relates to a support vector machine-based space non-cooperative target pulse maneuver detection method, which comprises the following steps: establishing a mapping function of a space target pulse maneuver velocity increment to an orbit element change quantity; establishing a maneuver behavior judgment function according to the mapping function; establishing a space non-cooperative target pulse maneuver identification model according to a maneuver sample set and the maneuver behavior judgment function; obtaining a pulse maneuver identification result according to historical orbit data of the space non-cooperative target and the maneuver identification model; and calculating and reconstructing pulse maneuver accurate information according to the maneuver identification result to obtain a space non-cooperative target pulse maneuver detection result. According to the space non-cooperative target pulse maneuver detection method provided in the examples of the application, the occurrence of the pulse maneuver of the space non-cooperative target can be identified, and detailed pulse maneuver information can be calculated and reconstructed, so that the pulse maneuver of the space non-cooperative target can be quickly and accurately detected from historical orbit data.

Description

Technical Field

[0001] This invention relates to the aerospace field, and in particular to a method for detecting the impulsive maneuvers of non-cooperative targets in space based on support vector machines. Background Technology

[0002] As global research into space technology deepens and space missions become increasingly frequent, competition for space resources intensifies. The number of non-cooperative targets in the space environment, such as spacecraft from other countries and space debris, is constantly increasing, making the space environment increasingly complex. Non-cooperative targets, primarily non-cooperative spacecraft, possess characteristics such as a lack of information sharing, insufficient prior knowledge, and unknown maneuvering behavior, potentially enabling them to reconnoiter, interfere with, or even attack our spacecraft. Therefore, to ensure the safe operation of our spacecraft, pulse maneuver detection of non-cooperative targets in space is a key technology for space situational awareness and a crucial development direction for future space defense technologies, holding significant importance.

[0003] Currently, the mainstream international methods for detecting impulsive maneuvers of non-cooperative space targets include traditional methods primarily based on analytical and numerical methods, and machine learning methods using neural networks. Analytical methods mainly select maneuver-sensitive features, establish a relationship between these features and the maneuver, and detect the maneuver by judging whether the relationship holds true. These methods require manually determining thresholds and often have poor resistance to noise interference. Numerical methods mainly establish dynamic models, solve systems of differential equations, obtain predicted trajectories, and compare them with actual trajectories to detect maneuvers. These methods can calculate specific information about the maneuver (maneuver time, maneuver size, etc.), but due to the need to solve systems of differential equations, they consume a large amount of computational resources and are inefficient. Machine learning methods mainly train neural networks on maneuver sample data, utilizing the strong fitting ability of neural network structures to establish maneuver detection models to achieve the purpose of maneuver detection. These methods improve upon the shortcomings of traditional methods, such as low computational efficiency, the need for manually determining thresholds, and poor resistance to noise interference. However, they often can only identify whether a maneuver has occurred and simple maneuver types, and cannot calculate more accurate and specific maneuver information. Summary of the Invention

[0004] In view of this, this invention proposes a space non-cooperative target pulse maneuver detection method based on support vector machines, targeting space non-cooperative targets with unknown maneuver information. By establishing a mapping function between the space target's pulse maneuver velocity increment and the change in orbital elements, a maneuver behavior judgment function is constructed, and a space non-cooperative target pulse maneuver identification model is established. This model identifies the maneuvers of space non-cooperative targets and, based on the maneuver identification results, calculates the precise information of the pulse maneuvers, thereby achieving space non-cooperative target pulse maneuver detection, including:

[0005] Step S1. Establish a mapping function between the space target's pulse maneuver velocity increment and the changes in orbital elements;

[0006] Step S2. Based on the mapping function of the space target pulse maneuver velocity increment to the change in orbital elements, establish a maneuver behavior judgment function that can reflect the pulse maneuver behavior of non-cooperative space targets;

[0007] Step S3. Based on the space non-cooperative target maneuver sample set and the maneuver behavior judgment function, establish a space non-cooperative target pulse maneuver recognition model;

[0008] Step S4. Perform segmented preprocessing on the historical orbit data of non-cooperative targets in space, and obtain the pulse maneuver recognition result of non-cooperative targets in space based on the segmented orbit data and the space non-cooperative target pulse maneuver recognition model.

[0009] Step S5. Based on the identified pulse maneuver of the non-cooperative target in space, the precise information of the pulse maneuver is calculated and reconstructed to obtain the pulse maneuver detection result of the non-cooperative target in space.

[0010] In the above method, step S1 includes:

[0011]

[0012] Where a is the semi-major axis of the orbit, i is the orbital inclination, and r is the distance between the space target and the Earth's center; e is the orbital eccentricity, E is the angle of anomalous approach, and θ is the true angle of anomalous approach. There are conversion relationships among the three. p is the semi-aperture, p = a(1-e) 2 ), Where Ω is the right ascension of the ascending node, and ω is the argument of perigee. n is the average angular velocity of rotation. Where μ is the Earth's gravitational constant; Δa is the change in the semi-major axis of the space target's orbit, used to describe the change in the size of the space target's orbit; Δe is the change in the eccentricity of the space target's orbit, used to describe the change in the shape of the space target's orbit; Δi is the change in the inclination of the space target's orbit, used to describe the change in the angle between the orbital plane and the Earth's equatorial plane; and ΔΩ is the change in the right ascension of the ascending node of the space target's orbit. The change in the perigee angle of a space target's orbit, Δε, is the change in the geocentric angular distance between the space target and the perigee, used to describe the change in the space target's relative position within its orbit; Δv S Δv T Δv W In the geocentric equatorial coordinate system, the velocity increments are the velocity components in the orbital radial, orbital tangential, and orbital normal directions.

[0013] In the above method, step S2 includes:

[0014] The variance of the orbital eccentricity e within a detection window characterizes the impulsive maneuvering behavior of a non-cooperative target in the orbital plane, and the maneuvering behavior judgment function λ1 is established as follows:

[0015]

[0016] Where k1 and k2 are the amplification coefficients, and the activation function tanh(·) is used to amplify the numerical representation of the difference;

[0017] The variance of the orbital inclination angle i within a detection window is used to characterize the pulse maneuvering behavior of a non-cooperative target perpendicular to the orbital plane. The maneuvering behavior judgment function λ2 is established as follows:

[0018]

[0019] Among them, k3 and k4 are the variation amplification coefficients, and the activation function tanh(·) is used to amplify the numerical representation of the difference.

[0020] In the above method, step S3 includes:

[0021] A single maneuver sample S consists of the orbital data of a space target over a period of time:

[0022] S = {z1, z2, ..., z} n}

[0023] Where z i Discrete orbital data:

[0024] z i ={a i ,e i i i ,Ω i ,ω i ,θ i ,t i ,F},i=1,2,...,n

[0025] Where t i For the time corresponding to this data, a i ,e i i i ,Ω i ,ω i ,θ i t i The orbital Kepler six-element data at time S, where F is the pulse maneuver behavior label for the maneuver sample S, including in-plane pulse maneuvers, perpendicular-plane pulse maneuvers, mixed-direction pulse maneuvers, and no maneuvers, and z1, z2, ..., z under the same maneuver sample S. nHaving the same pulse maneuvering behavior label F;

[0026] Based on the maneuver behavior judgment function, the mapping relationship between pulse maneuvers in different directions and changes in orbital elements is obtained as follows:

[0027] F j =f H-SVM (λ1,λ2,t sj ,t ej )

[0028] Where F j For the j-th maneuver sample S j The pulse maneuvering behavior tag, t sj With t ej Let λ1 and λ2 be the start and end times of the j-th sample, respectively, and let λ1 and λ2 be the judgment functions described in step S2. Based on the judgment function λ1, a first-layer binary classifier is obtained to judge the pulse maneuvering behavior within the orbital plane. Based on the judgment function λ2, a second-layer binary classifier is obtained to judge the pulse maneuvering behavior perpendicular to the orbital plane. The maneuvering sample set is used to train each layer of binary classifiers separately, and the binary classifiers of each layer are combined to establish a spatial non-cooperative target pulse maneuvering recognition model.

[0029] In the above method, step S4 includes:

[0030] Historical orbital data of non-cooperative space targets are preprocessed using a sliding window, dividing them into multiple equal-length orbital data segments. The orbital data within each sliding window is then substituted into the maneuvering behavior judgment functions λ1 and λ2 to obtain the maneuvering behavior judgment function values ​​for each time period corresponding to each sliding window. These function values ​​are then input as feature values ​​into the space non-cooperative target pulse maneuvering recognition model to determine whether a maneuver has occurred within the sliding window, yielding the maneuvering recognition result for this historical orbital data.

[0031] R j ={t sj ,t ej ,F j}

[0032] Among them, R j t represents the motion recognition result of the j-th sliding window. sj With t ej Let F be the start and end times of the track data segment within the j-th sliding window, respectively. j The pulse maneuver type output by the pulse maneuver identification model.

[0033] In the above method, step S5 includes:

[0034] Step S51. Based on the motor recognition result R jThe set of sliding windows that detect the maneuver is obtained, and based on this, a coarse time range for the occurrence of the pulse maneuver is determined, with the start time t. s With termination time t e These are taken as the starting times for the forward and backward integrations, respectively.

[0035] Step S52. Set the integration step size dt and the minimum equivalent velocity increment Δv for which pulse maneuvering is assumed to occur. min And set the loop counter value i = 0;

[0036] Step S53. Let t i =t s +i·dt, from the starting time t of the forward integration s Forward integration to t i Starting from the initial time t of the backward integration e Backward integration to t i The two tracks before and after the maneuver are obtained at t. i The difference in position and velocity at any given time Δr i With Δv i ;

[0037] Step S54. Determine t i Is it greater than t? e If yes, proceed to step S55; otherwise, increment i by 1 and proceed to step S53.

[0038] Step S55. Find |Δr i The minimum point, corresponding to t. i With Δv i This refers to the timing and magnitude of the pulse maneuver.

[0039] Step S56. Compare |Δv i |with Δv min If |Δv i |≥Δv min If the pulse maneuver occurs, the maneuver calculation is considered successful, and the corresponding maneuver time t is output. cal With the size of the maneuver No, then it is assumed that no pulse maneuver occurred;

[0040] The resulting pulse maneuver detection results for non-cooperative targets in space are as follows:

[0041]

[0042] Where t cal For the moment of pulse maneuver, These are the components of the equivalent velocity increment of the pulse maneuver in the three directions of the geocentric inertial coordinate system. Attached Figure Description

[0043] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without any creative effort or labor.

[0044] Figure 1 This is a flowchart of the space non-cooperative target pulse maneuver detection method based on support vector machine provided in the embodiments of the present invention;

[0045] Figure 2 This is a schematic diagram of the H-SVM vector machine construction in an example of the present invention;

[0046] Figure 3 This is a schematic diagram of the track maneuver recognition model in an example of the present invention;

[0047] Figure 4 This is a schematic diagram of the input and output of the pulse maneuver reconstruction algorithm based on forward-backward integration in an example of the present invention;

[0048] Figure 5 This is a schematic diagram of the pulse maneuver reconstruction principle based on forward-backward integration in an example of the present invention;

[0049] Figure 6 This is a flowchart of the pulse maneuver reconstruction algorithm based on forward-backward integration in an example of the present invention;

[0050] Figure 7 This is a schematic diagram of the comparative labels for the maneuvering behavior of non-cooperative targets in space in an example of the present invention;

[0051] Figure 8 This is a schematic diagram of the results of identifying non-cooperative target maneuvers in space based on H-SVM.

[0052] Figure 9 This is a schematic diagram of the simulation results of maneuver reconfiguration considering only position noise.

[0053] Figure 10 This is a schematic diagram of the simulation results of maneuver reconfiguration considering only velocity noise.

[0054] Figure 11 A schematic diagram of the simulation results of maneuver reconfiguration considering both position and velocity noise conditions; Specific Implementation

[0055] To better understand the technical solution of the present invention, the embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0056] It should be understood that the described embodiments are only some embodiments of the present invention, and not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without inventive effort are within the scope of protection of the present invention.

[0057] This invention relates to a method for detecting impulsive maneuvers of non-cooperative targets in space based on support vector machines. Please refer to [link / reference]. Figure 1 This is a flowchart illustrating the spatial non-cooperative target pulse maneuver detection method based on support vector machines provided in this invention, as shown below. Figure 1 As shown, the method includes the following steps:

[0058] Step S1. Establish a mapping function between the space target pulse maneuver velocity increment and the changes in orbital elements.

[0059] Applying an impulse of any magnitude to a space target will cause changes in its six orbital elements. It is stipulated that at the instant the impulse is applied, the space target's position remains unchanged, only its velocity changes. In the geocentric equatorial coordinate system, the velocity increments in the radial (S), tangential (T), and normal (W) directions are defined as Δv. S Δv T and Δv W The relationship is: Δv=Δv W +Δv S +Δv T .

[0060] Velocity increments in different directions will cause different changes in the six elements of the orbit, and the derivation relationship is as follows:

[0061]

[0062] Where a is the semi-major axis of the orbit, i is the orbital inclination, and r is the distance between the space target and the Earth's center; e is the orbital eccentricity, E is the angle of anomalous approach, and θ is the true angle of anomalous approach. There are conversion relationships among the three. p is the semi-aperture, p = a(1-e) 2 ), Where Ω is the right ascension of the ascending node, and ω is the argument of perigee. n is the average angular velocity of rotation. Where μ is the Earth's gravitational constant; Δa is the change in the semi-major axis of the space target's orbit, used to describe the change in the size of the space target's orbit; Δe is the change in the eccentricity of the space target's orbit, used to describe the change in the shape of the space target's orbit; Δi is the change in the inclination of the space target's orbit, used to describe the change in the angle between the orbital plane and the Earth's equatorial plane; and ΔΩ is the change in the right ascension of the ascending node of the space target's orbit. The change in the perigee angle of a space target's orbit, Δε, is the change in the geocentric angular distance between the space target and the perigee, used to describe the change in the space target's relative position within its orbit; Δv S Δv T Δv W In the geocentric equatorial coordinate system, the velocity increments are the velocity components in the orbital radial, orbital tangential, and orbital normal directions.

[0063] Based on the relationship between velocity increment and the six elements of the orbit, it can be seen that a velocity increment perpendicular to the orbital plane will cause changes in the orbital inclination and the right ascension of the ascending node, but will not cause changes in the semi-major axis and eccentricity; a velocity increment within the orbital plane will cause changes in the semi-major axis and eccentricity, but will not cause changes in the orbital inclination and the right ascension of the ascending node.

[0064] Step S2. Based on the mapping function of the space target pulse maneuver velocity increment to the change in orbital elements, establish a maneuver behavior judgment function that can reflect the pulse maneuver behavior of non-cooperative space targets.

[0065] The variance of the orbital eccentricity e within a detection window characterizes the impulsive maneuvering behavior of a non-cooperative target in the orbital plane, and the maneuvering behavior judgment function λ1 is established as follows:

[0066]

[0067] in, Let be the variance of the track eccentricity e within a detection window, and k1 and k2 be the variation amplification coefficients. The activation function tanh(·) is used to amplify the numerical representation of the difference.

[0068] The variance of the orbital inclination angle i within a detection window is used to characterize the pulse maneuvering behavior of a non-cooperative target perpendicular to the orbital plane. The maneuvering behavior judgment function λ2 is established as follows:

[0069]

[0070] in, Let be the variance of the track eccentricity i within a detection window, and k3 and k4 be the variation amplification coefficients. The activation function tanh(·) is used to amplify the numerical representation of the difference.

[0071] Step S3. Based on the space non-cooperative target maneuver sample set and the maneuver behavior judgment function, establish a space non-cooperative target pulse maneuver recognition model based on hierarchical support vector machine (H-SVM).

[0072] Based on historical orbital data sequences of non-cooperative space targets obtained through external measurements, and combined with human experience, maneuver behavior labels are added to obtain a classification training sample set with maneuver behavior judgment labels. A single maneuver sample S consists of the orbital data of the space target over a period of time.

[0073] S = {z1, z2, ..., z} n}

[0074] Where z i Discrete orbital data:

[0075] z i ={a i ,e i i i ,Ω i ,ω i ,θ i ,t i ,F},i=1,2,...,n

[0076] Where t i For the time corresponding to this data, a i ,e i i i ,Ω i ,ω i ,θ i t i The orbital Kepler six-element data at time S, where F is the pulse maneuver behavior label for the maneuver sample S, including in-plane pulse maneuvers, perpendicular-plane pulse maneuvers, mixed-direction pulse maneuvers, and no maneuvers, and z1, z2, ..., z under the same maneuver sample S. n They have the same pulse maneuvering behavior label F.

[0077] A hierarchical support vector machine (H-SVM) is constructed using the standard support vector machine (C-SVM) binary classifier as the basic unit. Based on the maneuver behavior judgment function, the mapping relationship between pulse maneuvers in different directions and changes in orbital elements is obtained as follows:

[0078] F j =f H-SVM (λ1,λ2,t sj ,t ej )

[0079] Where F j For the j-th maneuver sample S j The pulse maneuvering behavior tag, t sj With t ej These represent the start and end times of the j-th sample, respectively; λ1 and λ2 are the judgment functions described in step S2. Based on the judgment function λ1, a first-layer classifier of the hierarchical support vector machine is obtained to judge the impulsive maneuvering behavior within the orbital plane; based on the judgment function λ2, a second-layer classifier of the hierarchical support vector machine is obtained to judge the impulsive maneuvering behavior perpendicular to the orbital plane; the C-SVM binary classifiers of each layer are trained using the maneuvering sample set, and the binary classifiers of each layer are combined (e.g., ...). Figure 2(As shown) Establish a space non-cooperative target pulse maneuver identification model based on H-SVM.

[0080] Step S4. Use a sliding window to segment and preprocess the historical orbit data of non-cooperative targets in space. Based on the segmented orbit data and the space non-cooperative target pulse maneuver recognition model, obtain the space non-cooperative target pulse maneuver recognition result.

[0081] like Figure 3 As shown, based on historical orbit data obtained from external measurements, a sliding window is used for segmented preprocessing, dividing the data into multiple equal-length segments. The orbit data within each sliding window is substituted into the maneuver behavior judgment functions λ1 and λ2 to obtain the maneuver behavior judgment function values ​​for the corresponding time periods of each sliding window. These function values ​​are then input as feature values ​​into the H-SVM-based space non-cooperative target pulse maneuver recognition model to determine whether a maneuver has occurred within the sliding window, resulting in the maneuver recognition result for this historical orbit data:

[0082] R j ={t sj ,t ej ,F j}

[0083] Among them, R j t represents the motion recognition result of the j-th sliding window. sj With t ej Let F be the start and end times of the track data segment within the j-th sliding window, respectively. j The pulse maneuver type output by the pulse maneuver identification model.

[0084] Step S5. Based on the identified pulse maneuver of the non-cooperative target in space, a pulse maneuver reconstruction algorithm based on forward-backward integration is used to solve and reconstruct the accurate pulse maneuver information to obtain the pulse maneuver detection result of the non-cooperative target in space.

[0085] like Figure 4 As shown, the identification result R of the non-cooperative space target pulse maneuver is used. j The historical orbit data corresponding to each window is used as the basis for a pulse maneuver reconstruction algorithm based on forward-backward integration (the principle is as follows). Figure 5 The input (as shown) is used to obtain the pulse maneuver detection results for non-cooperative spacecraft through calculation. Figure 6 As shown, the pulse maneuver solution is performed according to the following procedure:

[0086] Step S51. Based on the motor recognition result R j The set of sliding windows that detect the maneuver is obtained, and based on this, a coarse time range for the occurrence of the pulse maneuver is determined, with the start time t. s With termination time te These are taken as the starting times for the forward and backward integrations, respectively.

[0087] Step S52. Set the integration step size dt and the minimum equivalent velocity increment Δv for which pulse maneuvering is assumed to occur. min And set the loop counter value i = 0;

[0088] Step S53. Let t i =t s +i·dt, from the starting time t of the forward integration s Forward integration to t i Starting from the initial time t of the backward integration e Backward integration to t i The two tracks before and after the maneuver are obtained at t. i The difference in position and velocity at any given time Δr i With Δv i ;

[0089] Step S54. Determine t i Is it greater than t? e If yes, proceed to step S55; otherwise, increment i by 1 and proceed to step S53.

[0090] Step S55. Find |Δr i The minimum point, corresponding to t. i With Δv i This refers to the timing and magnitude of the pulse maneuver.

[0091] Step S56. Compare |Δv i |with Δv min If |Δv i |≥Δv min If the pulse maneuver occurs, the maneuver calculation is considered successful, and the corresponding maneuver time t is output. cal With the size of the maneuver If not, it is assumed that no pulse maneuver occurred.

[0092] The resulting pulse maneuver detection results for non-cooperative targets in space are as follows:

[0093]

[0094] Where t cal For the moment of pulse maneuver, These are the components of the equivalent velocity increment of the pulse maneuver in the three directions of the geocentric inertial coordinate system.

[0095] Based on the method provided in the embodiments of the present invention, a simulation of a space non-cooperative target pulse maneuver detection method based on support vector machine was carried out, and a simulation experiment was conducted on the recognition, calculation and reconstruction of space non-cooperative target maneuvers.

[0096] Specifically, in this example, the amplification factor k1 is defined as 1 × 10. 35 k2 = 10, k3 = 1 × 10 40 If k4 = 10, then the judgment function for the spatial non-cooperative target pulse maneuvering behavior is:

[0097]

[0098]

[0099] Based on the orbital dynamics model with J2J3J4 terms of Earth non-spherical perturbation, the random initial Kepler elements and random pulse maneuver size ranges of the space non-cooperative target are set as shown in Table 1. The orbit is recursively calculated at 10-second intervals for a total of 10 days. Random pulse maneuvers of random size and direction are added at random times from day 3 to day 7. Noise is added to each orbital position and velocity data. Zero-mean Gaussian white noise with σ = 200m is added to the position components in all three directions, and zero-mean Gaussian white noise with σ = 5cm / s is added to the velocity components in all three directions. This generates time-series data of space non-cooperative target orbital elements with mixed navigation errors and behavioral labels, which serves as the training dataset for the H-SVM maneuver recognition model. 450 test datasets are generated using the same method. The number of occurrences of each type of maneuver is shown in Table 2.

[0100] Table 1 Initial Kepler Six Elements and Maneuver Range of Non-Cooperative Targets in Space

[0101]

[0102] Table 2 Dataset for Recognizing Non-Cooperative Target Maneuvers in Space

[0103]

[0104] Based on the maneuver behavior judgment function in step S2, a space non-cooperative target pulse maneuver recognition model is trained, and the model is tested using test set data. The results are as follows: Figures 7-8 As shown, Figure 7 For motor behavior comparison labels, Figure 8 The results of maneuver recognition using the trained model on the test set data show that the recognition accuracy is as high as 98.7% for 450 test samples in the test set. There are only 1.3% of test samples with missed detection (i.e., actual maneuvering but not recognized) and no false detection (i.e., no actual maneuvering but the recognition result is that maneuvering has occurred). This proves the effectiveness of the pulse maneuver recognition model based on H-SVM in step S3.

[0105] To verify the maneuver reconfiguration algorithm in step S5, the following three simulation experiments are set up:

[0106] (1) Only consider the orbit determination error of the position of the non-cooperative target in space.

[0107] ① The initial Kepler six roots of the non-cooperative space target are randomly generated, and the range is shown in Table 1;

[0108] ② Zero-mean Gaussian white noise with standard deviations of 50 / 100 / 200 / 500m was added to the position components in the three directions respectively (according to the 3σ principle, that is, the position orbit determination accuracy is 259.8 / 519.6 / 1039.2 / 2598.1m respectively) to serve as orbit determination data for non-cooperative space targets;

[0109] ③ The interval for determining the orbit of non-cooperative targets in space is 1 day (i.e., one target orbit data with mixed noise per day, and intermediate data are generated using an orbit prediction model with an interval of 10 seconds).

[0110] ④ Add pulse maneuvers in random directions at random times between day 3 and day 7, with the equivalent velocity increments of the pulse maneuvers being 1 / 5 / 10 / 30 / 50 m / s, respectively;

[0111] ⑤ Conduct 1000 simulations, with each simulation lasting 10 days.

[0112] (2) Only consider the orbit determination error of the velocity of non-cooperative targets in space.

[0113] ① The initial Kepler six roots of the non-cooperative space target are randomly generated, and the range is shown in Table 1;

[0114] ② Zero-mean Gaussian white noise with standard deviations of 1 / 2 / 5 / 10 cm / s was added to the velocity components in the three directions respectively (according to the 3σ principle, that is, the velocity orbit determination accuracy is 5.2 / 10.4 / 26.0 / 52.0 cm / s respectively) to serve as orbit determination data for non-cooperative space targets;

[0115] ③ The interval for determining the orbit of non-cooperative targets in space is 1 day;

[0116] ④ Add pulse maneuvers in random directions at random times between day 3 and day 7, with the equivalent velocity increments of the pulse maneuvers being 1 / 5 / 10 / 30 / 50 m / s, respectively;

[0117] ⑤ Conduct 1000 simulations, with each simulation lasting 10 days.

[0118] (3) Simultaneously consider the orbit determination error of the position and velocity of the non-cooperative target in space.

[0119] ① The initial Kepler six roots of the non-cooperative space target are randomly generated, and the range is shown in Table 1;

[0120] ② Zero-mean Gaussian white noise with standard deviations of 50 / 100 / 200 / 500m was added to the position components in the three directions, and zero-mean Gaussian white noise with standard deviations of 1 / 2 / 5 / 10cm / s was added to the velocity components in the three directions, as orbit determination data for non-cooperative space targets.

[0121] ③ The interval for determining the orbit of non-cooperative targets in space is 1 day;

[0122] ④ Add pulse maneuvers in random directions at random times between day 3 and day 7, with the equivalent velocity increments of the pulse maneuvers being 1 / 5 / 10 / 30 / 50 m / s, respectively;

[0123] ⑤ Conduct 1000 simulations, with each simulation lasting 10 days.

[0124] The simulation results of experiment (1) are as follows: Figure 9 As shown in Figures a), b), and c), the curves represent the success rate of maneuver detection, the average calculation error of maneuver size, and the average calculation error of maneuver timing, respectively. From the curves in the figures, it can be seen that, considering only the position and orbit determination error, when the orbit determination accuracy is high (σ≤100m), the success rate of calculating pulse maneuvers above 30m / s is close to 100% (with the maneuver timing calculation error less than 3600s and the maneuver size calculation error less than 30% as the success condition), the average calculation error of maneuver size is less than 2%, and the average calculation error of maneuver timing is less than 100s; the simulation results of experiment (2) are as follows: Figure 10 As shown in the figure. From the curve in the figure, it can be seen that, under the condition of considering only the velocity orbit determination error, when the orbit determination accuracy is high (σ≤2cm / s), the success rate of solving pulse maneuvers above 30m / s is higher than 95%, the average solution error of maneuver size is less than 3%, and the average solution error of maneuver timing is less than 135s; the simulation results of experiment (3) are as follows. Figure 11 As shown in the figure, the curves indicate that, considering both position and velocity errors in orbit determination, with high orbit determination accuracy (σ≤100m, 2cm / s), the success rate for calculating pulse maneuvers above 30m / s is higher than 93%, the average calculation error for maneuver size is less than 3%, and the average calculation error for maneuver timing is less than 150s. Even with very poor orbit determination accuracy (σ≥200m, 5cm / s), the success rate for calculating pulse maneuvers at 50m / s still remains above 80%, the average calculation error for maneuver size is less than 10%, and the average calculation error for maneuver timing is less than 400s.

[0125] This concludes the entire content of the space non-cooperative target pulse maneuver detection method in this example.

[0126] The embodiments of the present invention have the following beneficial effects:

[0127] This paper combines the advantages of machine learning and traditional methods for maneuver detection. A hierarchical support vector machine (Hierarchical Support Vector Machine) is used to establish a maneuver recognition model. This model performs preliminary screening of impulsive maneuvers in historical target orbit data. A impulsive maneuver reconstruction algorithm based on forward-backward integration is then designed to further detect time periods identified by the maneuver recognition model as potentially indicative of impulsive maneuvers. This allows for the calculation of more accurate details such as the timing and magnitude of the impulsive maneuvers, demonstrating good detection accuracy and robustness. According to the technical solution provided in this embodiment, effective detection of impulsive maneuvers of non-cooperative space targets can be achieved, allowing for the acquisition of their maneuvering information and thus assisting our spacecraft in avoiding potential risks from non-cooperative spacecraft.

[0128] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

[0129] The contents not described in detail in this specification are common knowledge to those skilled in the art.

Claims

1. A method for detecting impulsive maneuvers of non-cooperative targets in space based on support vector machines, characterized in that, The method includes: Step S1. Establish a mapping function between the space target's pulse maneuver velocity increment and the changes in orbital elements; Step S2. Based on the mapping function of the space target's pulse maneuver velocity increment to the change in orbital elements, establish a maneuver behavior judgment function that can reflect the pulse maneuver behavior of a non-cooperative space target, wherein the orbital eccentricity within a detection window is used as the criterion. The variance characterizes the impulsive maneuvering behavior of a non-cooperative target in the orbital plane, and a maneuvering behavior judgment function is established. for: in, , To vary the amplification factor, an activation function is used. Expanding the differences in numerical representation; Track inclination angle within a detection window The variance characterizes the pulse maneuvering behavior of non-cooperative targets in space perpendicular to the orbital plane, and a maneuvering behavior judgment function is established. for: in, , To vary the amplification factor, an activation function is used. Expanding the differences in numerical representation; Step S3. Based on the space non-cooperative target maneuver sample set and the maneuver behavior judgment function, establish a space non-cooperative target pulse maneuver recognition model; Step S4. Perform segmented preprocessing on the historical orbit data of non-cooperative targets in space, and obtain the pulse maneuver recognition result of non-cooperative targets in space based on the segmented orbit data and the space non-cooperative target pulse maneuver recognition model. Step S5. Based on the identified pulse maneuver of the non-cooperative target in space, the precise information of the pulse maneuver is calculated and reconstructed to obtain the pulse maneuver detection result of the non-cooperative target in space.

2. The method according to claim 1, characterized in that, Step S1 includes: in, For the semi-major axis of the track, For the track inclination angle, The distance between the spatial target and the Earth's center; Here, E is the orbital eccentricity, and E is the asymmetry angle. For true near point angles, there are conversion relationships among the three. ; semi-pipe , ,in Right ascension of the ascending node, The perigee argument, ; average rotational angular velocity ,in It is the Earth's gravitational constant; This is the change in the semi-major axis of a space target's orbit, used to describe the magnitude change in the orbit of a space target. This refers to the change in the eccentricity of a space target's orbit, used to describe the shape changes of the space target's orbit. The change in orbital inclination of a space target describes the change in the angle between the orbital plane and the Earth's equatorial plane. The change in right ascension of the ascending node of a space target's orbit; The change in the perigee angle of the orbit of a space target. This is the change in geocentric angular distance between a space target and its perigee, used to describe the change in the relative position of the space target in its orbit. , , In the geocentric equatorial coordinate system, the velocity increments are the velocity components in the orbital radial, orbital tangential, and orbital normal directions.

3. The method according to claim 1, characterized in that, Step S3 includes: Single maneuver sample It consists of orbital data of a space target over a period of time: in Discrete orbital data: in This is the time corresponding to the data. They are respectively Kepler six-element data of orbit at any given time. For this motor sample The pulse maneuver behavior labels include in-plane pulse maneuvers, vertical-plane pulse maneuvers, mixed-direction pulse maneuvers, and no maneuvers, for the same maneuver sample. Below Having the same pulse maneuvering behavior label ; Based on the maneuver behavior judgment function, the mapping relationship between pulse maneuvers in different directions and changes in orbital elements is obtained as follows: in For the first One motor sample Pulse maneuvering behavior tags, and The first The start and end times of each sample. , The judgment function described in step S2; based on the judgment function The first-layer binary classifier is obtained to determine the pulse maneuvering behavior within the orbital plane; based on the judgment function... A second-layer binary classifier is obtained to determine the pulse maneuvering behavior on the vertical track surface; the binary classifiers of each layer are trained using the maneuvering sample set, and the binary classifiers of each layer are combined to establish a spatial non-cooperative target pulse maneuvering recognition model.

4. The method according to claim 1, characterized in that, Step S4 includes: Historical orbital data of non-cooperative space targets are preprocessed using a sliding window, dividing them into multiple equal-length orbital data segments; the orbital data within each sliding window is then substituted into the maneuver behavior judgment function. , In the process, the maneuver behavior judgment function value for each time period corresponding to each sliding window is obtained; the function value is used as a feature value and input into the space non-cooperative target pulse maneuver recognition model to determine whether a maneuver occurs within the sliding window, and the maneuver recognition result for the historical trajectory data is obtained as follows: in, Indicates the first The motion recognition results of a sliding window and The first The start and end times of the track data segment within each sliding window. The pulse maneuver type output by the pulse maneuver identification model.

5. The method according to claim 1, characterized in that, Step S5 includes: Step S51. Based on the motor recognition result The set of sliding windows that detect the maneuver is obtained, and based on this, a rough time range and start time of the pulse maneuver are determined. With end time These are taken as the starting times for the forward and backward integrations, respectively. Step S52. Set the integration step size The magnitude of the minimum equivalent velocity increment that is considered to cause pulse maneuvering and make the loop count value ; Step S53. Let From the start time of forward integration Forward integration to Starting from the beginning of the backward integration Backward integration to The two tracks before and after the maneuver were obtained. Difference in position and velocity at time and ; Step S54. Determine Is it greater than If yes, then proceed to step S55; otherwise, then... Execute S53; Step S55. Search Minimum point, the point corresponding to and This refers to the timing and magnitude of the pulse maneuver. Step S56. Comparison and ,like If the pulse maneuver occurs, the maneuver calculation is considered successful, and the corresponding maneuver time is output. With the size of the maneuver If not, then it is assumed that no pulse maneuver occurred; The resulting pulse maneuver detection results for non-cooperative targets in space are as follows: in For the moment of pulse maneuver, , , These are the components of the equivalent velocity increment of the pulse maneuver in the three directions of the geocentric inertial coordinate system.

Images