An intelligent identification method for orbit types of only angularly lunar space lagrange points

Abstract

The application discloses a kind of only angle measurement lunar space Lagrange point orbit type intelligent identification method, belong to space technology field.The application establishes only angle measurement observation model, sets only angle measurement observation error covariance matrix;Establish the motion model of space Lagrange point orbit spacecraft in lunar space;According to the number of space Lagrange point orbit in lunar space, randomly generate space Lagrange point orbit in lunar space and recursion length, generate only angle measurement observation vector;The sample point required for training orbit type identification deep neural network is generated;Based on deep neural network, the fast and efficient identification of space Lagrange point orbit in lunar space is realized, and then the initial orbit determination and refined orbit estimation of space Lagrange point orbit in lunar space are realized.The application has the advantages of not depending on physical model, high identification accuracy, small amount of calculation, high efficiency and good robustness.The application is beneficial to improve the space situation awareness capability in lunar space, improve the determination accuracy and calculation speed of space Lagrange point orbit in lunar space.

Description

Technical Field

[0001] This invention relates to an intelligent identification method for the type of Lagrange point orbit in Earth-Moon space, belonging to the field of space technology. Background Technology

[0002] The Lagrange point orbits of the Earth-Moon three-body system possess exceptional stability, making them invaluable for deep-space exploration, Sun-Earth-Moon observation, and Earth-Moon space communication. Precise orbit determination of these orbits is a prerequisite for the successful implementation of such missions. Due to the numerous types and complex forms of Earth-Moon Lagrange point orbits, their classification needs to be identified before precise orbit determination. Existing physical model-driven orbit type identification and classification methods rely on accurate mathematical models and orbit measurement information, resulting in low accuracy and high computational cost under ultra-short arc observation conditions. Therefore, it is necessary to develop data-driven identification methods for Earth-Moon Lagrange point orbit type identification, using deep neural networks to achieve rapid and accurate identification of orbit types. Summary of the Invention

[0003] The primary objective of this invention is to provide an intelligent identification method for the type of Earth-Moon Lagrange point orbits using only angle measurement. Based on a deep neural network, this method achieves rapid and efficient identification of Earth-Moon Lagrange point orbit types, thereby enabling initial orbit determination and refined orbit estimation. This invention offers advantages such as independence from physical models, high identification accuracy, low computational cost, high efficiency, and good robustness. This invention is beneficial for enhancing situational awareness in Earth-Moon space, improving the accuracy and speed of Earth-Moon Lagrange point orbit determination, and enhancing the on-orbit servicing and operational capabilities of spacecraft orbiting Earth-Moon Lagrange point orbits.

[0004] The objective of this invention is achieved through the following technical solutions.

[0005] This invention discloses an intelligent identification method for the type of Earth-Moon Lagrange point orbits using only angle measurement. The method sets the initial time, the position and state of the ground station, the number of Earth-Moon Lagrange point orbit types, the number of samples for each type, the upper limit and lower limit of the orbit recursion duration, and the number of observations. An angle-only observation model is established, and the angle-only observation error covariance matrix is ​​defined. A spacecraft motion model for Earth-Moon Lagrange point orbits is established. Earth-Moon Lagrange point orbits and recursion durations are randomly generated based on the number of Earth-Moon Lagrange point orbit types, generating angle-only observation vectors. Each orbit is labeled, and the label represents the orbit type. Input parameters corresponding to each orbit are generated, and sample points are generated for training a deep neural network for orbit type identification. Cross-entropy is used as the loss function, and the parameter weights of the deep neural network for orbit type identification are calculated through iterative updates to obtain an offline-trained deep neural network for orbit type identification. The continuous angle-only information obtained from actual measurements is input into a trained deep neural network for orbit type identification. The output of the trained deep neural network corresponds to the Earth-Moon space Lagrange point orbit type. Based on the obtained Earth-Moon space Lagrange point orbit type, further orbit matching and orbit determination are performed. The Kalman filter algorithm or the least quadratic algorithm is used to perform refined orbit estimation of the Earth-Moon space Lagrange point orbit, thereby enabling rapid perception of the state of targets in Earth-Moon space, improving the situational awareness capability in Earth-Moon space, and improving the accuracy and calculation speed of Earth-Moon space Lagrange point orbit determination. This is beneficial to improving the on-orbit service and on-orbit operation capabilities of spacecraft in Earth-Moon space Lagrange point orbits.

[0006] This invention discloses an intelligent identification method for the orbit type of the Earth-Moon Lagrange point in space using only angle measurement, comprising the following steps:

[0007] Step 1: Set the initial time t0, the position and status of the ground station, the number of Earth-Moon space Lagrange point orbit types M, the number of samples for each type of orbit N, and the upper limit of the orbit recursion time T. max Lower limit of orbit recursion duration T min With the number of observations n O An angle-only observation model was established, and the angle-only observation error covariance matrix was set.

[0008] Set the initial time t0. Set the position status of the ground station, including longitude L, latitude B, and altitude H. Set the number of Earth-Moon space Lagrange point orbit types M, the number of samples N for each type of orbit, and the upper limit T for orbit recursion. max Lower limit of orbit recursion duration T min With the number of observations n O The angle-only observation model is established as follows:

[0009]

[0010] in Indicates the vector observed only for angle measurement. For observation models that only measure angles, This indicates noise from angle-only observations. and Let represent the position vectors of the spacecraft and the station in the geocentric inertial frame, respectively. The covariance matrix of the angle-only observation error is defined.

[0011] Step 2: Establish a motion model of the spacecraft orbiting at the Earth-Moon Lagrange point.

[0012] For a spacecraft orbiting at a Lagrange point in the Earth-Moon space system, its motion is mainly affected by the gravitational perturbations of the Earth, the Moon, and the Sun. Other perturbations are negligible in magnitude compared to the influence of these three gravitational forces. Therefore, the motion model of the spacecraft orbiting at a Lagrange point in the Earth-Moon space system is established as follows:

[0013]

[0014] in μ represents the velocity vector of a spacecraft in a geocentric inertial frame. e μ m and μ s These represent the gravitational constants of the Earth, Moon, and Sun, respectively. and This represents the position vectors of the Moon and the Sun in the Earth-centric inertial frame.

[0015] Step 3: Based on the number of Earth-Moon space Lagrange point orbit types M, randomly generate Earth-Moon space Lagrange point orbits and recursive duration T, and generate angle-only observation vectors.

[0016] For each of the M types of Earth-Moon Lagrange point orbits, N orbits are randomly generated according to amplitude, resulting in a total of MN orbits. For each orbit, the recursion duration T is at the upper limit T of the orbit recursion duration. max Lower limit of orbit recursion duration T min The selection is randomized between [0, T]. Simultaneously, n is randomly generated within the time interval [0, T]. O There are 1 observation time points, denoted as 0≤t1<...<t nO ≤T. For each time t i (1≤i≤n O ), calculate the corresponding angle measurement observation vector according to equation (1), where the observation noise Based on the covariance matrix of the angle-only observation error Randomly generated. The recursive duration T and the corresponding observation time t are generated randomly. iThis improves the randomness of orbits and observations in the sample data, thereby enhancing the applicability and robustness of the deep neural network for orbit type identification trained on it.

[0017] Step 4: Label each track with a tag y, which is the track type. Generate the input parameter X corresponding to each track and generate the sample points S = {X→y} required to train the deep neural network for track type identification.

[0018] For the k-th orbital, the label Defined as:

[0019]

[0020] The superscript i represents the i-th element of label y.

[0021] For each orbit, n O The secondary line-of-sight angle measurement, the corresponding time of each measurement, and the position vector of the station in the geocentric inertial frame all contain information used to identify the orbit type of the Earth-Moon Lagrange point. Therefore, the input parameters for the sample are:

[0022]

[0023] Based on equations (3) and (4), the samples required for training the deep neural network for track type identification are as follows:

[0024] S={X→y} (5)

[0025] Step 5: Based on the sample points S = {X→y} obtained in Step 4, use cross-entropy as the loss function and calculate the parameter weights of the orbit type identification deep neural network through iterative updates to obtain the offline trained orbit type identification deep neural network.

[0026] Neural network training is a regression problem; therefore, the cross-entropy between the network output value and the expected value is used as the metric to evaluate the predictive performance of the deep neural network model during training. This metric is the loss function. For a given set of M orbital types and MN sample points, the expression for the cross-entropy loss function is as follows:

[0027]

[0028] Where y ic The sign function is p, which takes the value 1 if the true class of sample i is equal to c, and 0 otherwise. icLet be the predicted probability that sample i belongs to category c. The closer the cross-entropy loss function is to 0, the better the performance of the orbital type identification deep neural network. Substitute the sample points calculated in step 4 into the cross-entropy loss function to calculate the current loss function value, randomly generate the weights of the orbital type identification deep neural network, continuously update the weights based on the current loss function value, and recalculate the cross-entropy loss function value.

[0029] Step 6: Input the continuous angle-only information obtained from the actual measurement into the orbit type identification deep neural network trained in Step 5. The output of the trained orbit type identification deep neural network corresponds to the orbit type of the Earth-Moon space Lagrange point.

[0030] The continuous angle-only information obtained from actual measurements is input into the trained orbit type identification deep neural network obtained in step 5. The output of the trained orbit type identification deep neural network is defined as... The trained deep neural network for track type recognition outputs The superscript corresponding to the largest element is the track type.

[0031] It also includes step 7, which, based on the obtained Earth-Moon space Lagrange point orbit type, further performs orbit matching and orbit determination, and uses the Kalman filter algorithm or the least quadratic algorithm to perform refined orbit estimation of the Earth-Moon space Lagrange point orbit, thereby improving the situational awareness capability of Earth-Moon space, improving the accuracy and calculation speed of Earth-Moon space Lagrange point orbit determination, and helping to improve the on-orbit service and on-orbit operation capability of spacecraft in Earth-Moon space Lagrange point orbit.

[0032] Beneficial effects:

[0033] 1. This invention discloses an intelligent identification method for the type of Earth-Moon Lagrange point orbits using only angle measurement. The method sets the initial time, the position and state of the ground station, the number of Earth-Moon Lagrange point orbit types, the number of samples for each type, the upper limit and lower limit of the orbit recursion duration, and the number of observations. An angle-only observation model is established, and the angle-only observation error covariance matrix is ​​defined. A spacecraft motion model for Earth-Moon Lagrange point orbits is established. Earth-Moon Lagrange point orbits and recursion durations are randomly generated based on the number of Earth-Moon Lagrange point orbit types, generating angle-only observation vectors. Each orbit is labeled, and the label represents the orbit type. Input parameters corresponding to each orbit are generated, and sample points are generated for training a deep neural network for orbit type identification. Cross-entropy is used as the loss function, and the parameter weights of the deep neural network for orbit type identification are calculated through iterative updates to obtain an offline-trained deep neural network for orbit type identification. The continuous angle-only information obtained from actual measurements is input into a trained deep neural network for orbit type identification. The output of the trained deep neural network corresponds to the orbit type at the Earth-Moon Lagrange point. Based on the obtained Earth-Moon Lagrange point orbit type, orbit matching and orbit determination are further performed. The Kalman filter algorithm or the least quadratic algorithm is used to perform refined orbit estimation of the Earth-Moon Lagrange point orbit, thereby achieving rapid perception of the target status in Earth-Moon space.

[0034] 2. The present invention discloses an intelligent identification method for the orbit type of the Earth-Moon Lagrange point in space using only angle measurement. By training a deep neural network for orbit type identification, it achieves efficient and rapid identification of the orbit type of the Earth-Moon Lagrange point, which is beneficial to improving the orbit determination accuracy of the Earth-Moon Lagrange point orbit.

[0035] 3. The present invention discloses an intelligent identification method for orbit types at Lagrange points in Earth-Moon space that only measures angles. By randomly generating orbit recursion duration and observation time, it is beneficial to improve the randomness of orbits and observations in sample data, and to enhance the applicability and robustness of the trained orbit type identification deep neural network.

[0036] 4. The intelligent identification method for the type of Earth-Moon Lagrange point orbit in space disclosed in this invention, based on the above three beneficial effects, has the advantages of not relying on a physical model, high identification accuracy, low computational load, high efficiency, and good robustness. This invention is beneficial for improving the situational awareness capability in Earth-Moon space, improving the accuracy and calculation speed of Earth-Moon Lagrange point orbit determination, and improving the on-orbit service and on-orbit operation capabilities of spacecraft in Earth-Moon Lagrange point orbits.

[0037] 5. The present invention discloses an intelligent identification method for the type of Earth-Moon space Lagrange point orbit that only measures angles. By replacing the Earth-Moon space Lagrange point orbit spacecraft dynamics with the Sun-Earth space Lagrange point orbit dynamics, it extends the application to the Sun-Earth space orbit determination problem and solves related technical problems in the field of spacecraft. Attached Figure Description

[0038] Figure 1 This is a flowchart of an intelligent identification method for the orbit type of a Lagrange point in Earth-Moon space using only angle measurement, as disclosed in this invention.

[0039] Figure 2 The trajectory of the Halo orbit at point L2 in the Earth-Moon space in the lunar inertial frame.

[0040] Figure 3 Let Halo orbit at point L2 in Earth-Moon space be the trajectory of the Earth-Moon rotational system. Detailed Implementation

[0041] To better illustrate the purpose and advantages of the present invention, the invention will be further described below in conjunction with the accompanying drawings and embodiments.

[0042] This example uses the Halo orbit and Lyapunov orbit at the L2 point in Earth-Moon space as examples to identify the type of Lagrange point orbit in Earth-Moon space using the method proposed in this invention.

[0043] like Figure 1 As shown in the figure, this example discloses an intelligent identification method for the orbit type of the Earth-Moon Lagrange point in space using only angle measurement. The specific implementation steps are as follows:

[0044] Step 1: Set the initial time t0, the position and status of the ground station, the number of Earth-Moon space Lagrange point orbit types M, the number of samples for each type of orbit N, and the upper limit of the orbit recursion time T. max Lower limit of orbit recursion duration T min With the number of observations n O An angle-only observation model was established, and the angle-only observation error covariance matrix was set.

[0045] The initial time is set to t0 = 0, corresponding to January 1, 2020. The position status of the ground station is set, including longitude L = -86.21, latitude B = 30.57, and altitude H = 34.7m. The number of Earth-Moon space Lagrange point orbit types is set to M = 2, the number of samples for each type of orbit is set to N = 100, and the upper limit of the orbit recursion time T is set. max =6 hours, lower limit of orbital recursion duration T min = 1 hour and the number of observations n O =10, establish an angle-only observation model, and set the angle-only observation error covariance matrix as follows:

[0046]

[0047] Step 2: Establish a motion model of the spacecraft orbiting at the Earth-Moon Lagrange point.

[0048] Establish a motion model for a spacecraft orbiting at a Lagrange point in the Earth-Moon space.

[0049] Step 3: Randomly generate the Earth-Moon space Lagrange point orbit and recursive duration based on the number of Earth-Moon space Lagrange point orbit types, and generate the angle-only observation vector.

[0050] Step 4: Label each track with a tag, which is the track type. Generate the input parameters corresponding to each track and generate the sample points needed to train the deep neural network for track type recognition.

[0051] Step 5: Based on the sample points obtained in Step 4, use cross-entropy as the loss function and calculate the parameter weights of the orbit type identification deep neural network through iterative updates to obtain the offline trained orbit type identification deep neural network.

[0052] Step 6: Input the continuous angle-only information obtained from the actual measurement into the orbit type identification deep neural network trained in Step 5. The output of the trained orbit type identification deep neural network corresponds to the orbit type of the Earth-Moon space Lagrange point.

[0053] Taking a Halo orbit at the Earth-Moon L2 point as an example, the state of this Halo orbit in the lunar inertial frame on January 1, 2020, is shown in Table 1 below. The corresponding orbit is as follows: Figure 2 and Figure 3 As shown.

[0054] Table 1 Initial values ​​of the Halo orbit in the inertial frame of gravity in January.

[0055] x y z Location / km 13783.2014276077 25401.5797340448 -70574.4430685413 Speed ​​(km / s) -0.02101744926256 -0.08174887938430 -0.02451279768752

[0056] The continuous angle-only information obtained from the actual measurement is input into the neural network obtained in step 5. The output of the neural network is:

[0057]

[0058] This corresponds to the L2 point Halo orbit, and the type identification is correct.

[0059] Step 7: Based on the obtained Earth-Moon space Lagrange point orbit type, orbit matching and orbit determination can be further performed. The Kalman filter algorithm or the least quadratic algorithm is used to perform refined orbit estimation of the Earth-Moon space Lagrange point orbit, thereby realizing rapid perception of the target status in Earth-Moon space and providing an information basis for subsequent close-range guidance, game control and other tasks.

[0060] The above detailed description further illustrates the purpose, technical solution, and beneficial effects of the invention. It should be understood that the above description is only a specific embodiment of the present invention and is not intended to limit the scope of protection of 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.

Claims

1. A method for only measuring the angle of the space Lagrange point orbit type intelligent identification, characterized in that: Comprising the following steps, Step 1: set initial time t0, position state of ground station, number of orbit types of Earth-Moon space Lagrange point M, number of samples N of each type of orbit, upper limit T of orbit recursion length max , lower limit T of orbit recursion length min , and observation times n O , establish an angle-only observation model, and set the angle-only observation error covariance matrix; Step 1 is implemented by, Setting an initial time t0; setting a position state of the ground station, the position state including a longitude L, a latitude B and a height H; setting a number of types of lunar-space Lagrange point orbits M, a number of samples N of each type of orbit, an upper limit T of an orbit recursion length max , a lower limit T of the orbit recursion length min and a number of observations n O , a model of only angle observation is established as follows: where denotes the angle-only observation vector, is the angle-only observation model, denotes the angle-only observation noise, and denote the position vectors of the spacecraft and the station in the geocentric inertial frame, respectively; and the angle-only observation error covariance matrix is set as Step 2: Establishing a spacecraft motion model of the Earth-Moon space Lagrange point orbit; Step 2 is implemented by, For the spacecraft of the Earth-Moon space Lagrange point orbit, its motion is mainly affected by the gravitational perturbation of the Earth, the Moon and the Sun, and other perturbation forces can be ignored compared to the three gravitational effects, so the spacecraft motion model of the Earth-Moon space Lagrange point orbit is established as follows: where denotes the velocity vector of the spacecraft in the geocentric inertial frame, μ e , μ m , and μ s denote the gravitational constants of the Earth, Moon, and Sun, respectively, and denote the position vectors of the Moon and Sun in the geocentric inertial frame; Step 3: Randomly generating the Earth-Moon space Lagrange point orbit and the recursion length T according to the number M of the Earth-Moon space Lagrange point orbit type, and generating the only angle observation vector; Step 4: Labeling the label y of each orbit, which is the orbit type, generating the input parameter X corresponding to each orbit, and generating the sample point S = {X→y} required for training the orbit type recognition deep neural network; Step 5: Based on the sample point S = {X→y} obtained in step 4, taking cross entropy as the loss function, the orbit type recognition deep neural network parameter weight is calculated by iterative updating to obtain the offline trained orbit type recognition deep neural network; Step 6: Input the actual measured continuous angle-only information into the orbit type recognition deep neural network trained in step 5, and the output of the trained orbit type recognition deep neural network corresponds to the Earth-Moon space Lagrange point orbit type.

2. The method of claim 1, wherein the method is a method of only angularly identifying the LEO orbit type of the LEO satellite. Further comprising step 7, based on the obtained Earth-Moon space Lagrange point orbit type, further performing orbit matching and orbit determination, using Kalman filtering algorithm or least square algorithm to perform fine orbit estimation of the Earth-Moon space Lagrange point orbit, improving the Earth-Moon space situation awareness capability, improving the Earth-Moon space Lagrange point orbit determination accuracy and calculation speed, which is beneficial to improve the on-orbit service and on-orbit operation capability of the Earth-Moon space Lagrange point orbit spacecraft.

3. The method of claim 2, wherein the method is a method of only angularly identifying the LEO orbit type of the LEO satellite. Step 3 is implemented by, For each of the M classes of the Earth-Moon space Lagrange points, N orbits are randomly generated according to the amplitude, and a total of MN orbits are generated; for each orbit, the recursion time T is randomly selected between the upper limit T max and the lower limit T min of the orbit recursion time; at the same time, n O observation times are randomly generated in the time period [0, T], denoted as For each time t i (1≤i≤n O ), the corresponding angle measurement vector is calculated according to formula (1), where Observation noise According to the angle-only observation error covariance matrix Randomly generated; through random generation recursive duration T and the corresponding observation time t i , improve the randomness of the orbit and observation in the sample data, and improve the applicability and robustness of the orbit type recognition deep neural network obtained by training.

4. The method of claim 3, wherein the method is a method of only angularly identifying the LEO orbit type of the LEO satellite. Step 4 is implemented by, For the kth class of orbits, the label is defined as: Wherein the superscript i represents the i-th element of the label y; For each orbit, n O The instantaneous time of each measurement and the position vector of the station in the geocentric inertial system contain information for identifying the type of orbit of the Earth-Moon spatial Lagrange point, so the input parameters of the sample are: According to formula (3) and (4), the sample required for training the orbit type recognition deep neural network is constructed as follows: S = {X→y} (5).

5. The method of claim 4, wherein the method is a method of only angularly identifying the LEO orbit type of the LEO satellite. Step 5 is implemented by, Neural network training is a regression problem, so the index for evaluating the prediction performance of the deep neural network model in the training process is selected as the cross entropy of the network output value and the expected value, which is the loss function; for the given M orbit types and MN sample points, the expression of the cross entropy loss function is as follows: where y ic is a sign function that takes 1 if the true class of sample i equals c and 0 otherwise, p ic is the predicted probability that sample i belongs to class c; the closer the cross-entropy loss function is to 0, the better the performance of the orbit type recognition deep neural network; the sample points calculated in step 4 are substituted into the cross-entropy loss function to calculate the current loss function value, the values of the orbit type recognition deep neural network weights are randomly generated, and the weights are constantly updated according to the current loss function value and the cross-entropy loss function value is recalculated.

6. The method of claim 5, wherein the method is a method of only angularly identifying the LEO orbit type of the LEO satellite. Step 6 is implemented by, The actual measured continuous angle-only information is input into the trained track type recognition deep neural network obtained in step 5, and the output of the trained track type recognition deep neural network is defined as The maximum element of the trained track type recognition deep neural network output The superscript corresponding to the maximum element is the track type.

Images