A neural network-based spacecraft relative motion configuration identification system and method

Abstract

This invention discloses a spacecraft relative motion configuration recognition system and method based on neural networks, belonging to the field of aerospace technology. The invention presents a method for recognizing the relative motion configuration of non-cooperative spacecraft, proposing a neural network-based approach. By constructing a spacecraft relative motion configuration dataset, relative motion configuration recognition can be achieved even when domain expert knowledge is insufficient. Through neural network models, the relative motion configuration of target spacecraft is identified, enabling the understanding of the relative motion configuration of non-cooperative spacecraft, thereby providing information-level support for space safety early warning and avoidance.

Description

Technical Field

[0001] This invention belongs to the field of aerospace technology, specifically relating to a spacecraft relative motion configuration recognition system and method based on neural networks. Background Technology

[0002] Currently, there are tens of thousands of trackable man-made objects orbiting the Earth. Of these, only about two thousand are operational, while the others are malfunctioning or damaged space debris from spacecraft disintegration. Incidents of debris impacts causing damage to spacecraft are increasingly common. In these events, malfunctioning spacecraft, space debris, and unknown spacecraft (those whose identity cannot be identified or whose information is unknown) are collectively referred to as spacecraft. The existence of these non-cooperative target spacecraft poses a serious threat to the safety of space activities. Current methods for predicting spacecraft orbits and providing safety warnings mainly focus on assessing the risk of collisions with space targets, commonly using regional collision warning methods and probability-based collision warning methods. However, these methods primarily use distance as an evaluation indicator and cannot provide a comprehensive threat assessment and warning of spacecraft. Therefore, to ensure the space safety of high-value satellites and the smooth implementation of on-orbit missions, it is necessary to understand the relative motion configurations of non-cooperative spacecraft, thereby providing information-level support for space safety warnings and avoidance. Summary of the Invention

[0003] In order to overcome the shortcomings of the prior art, the present invention aims to provide a spacecraft relative motion configuration recognition system and method based on neural networks, so as to solve the technical problem that distance, as an evaluation indicator for orbit prediction and safety early warning of spacecraft, cannot comprehensively assess and warn of threats to spacecraft.

[0004] To achieve the above objectives, the present invention employs the following technical solution:

[0005] A method for spacecraft relative motion configuration recognition based on neural networks, comprising:

[0006] Obtain the position information between the spacecraft and the central spacecraft, and determine the information and number of input nodes, hidden layer nodes, and output nodes of the neural network model, as well as the activation functions between each layer.

[0007] A dataset of spacecraft relative motion configurations is constructed based on the positional information between the spacecraft and the central spacecraft.

[0008] A neural network model was constructed based on the spacecraft relative motion configuration dataset.

[0009] The motion state information of the target spacecraft is input into a neural network model to identify the relative motion configuration of the spacecraft.

[0010] Preferably, the input nodes of the neural network model are selected as "relative position" and "relative velocity" information, and the number of output nodes of the prediction network model is the same as the number N of relative motion configurations; the hidden layer activation function f(·) and the output layer activation function g(·) use the Sigmoid function.

[0011] Preferably, the relative motion configuration of the spacecraft includes: rendezvous, forward drift, reverse drift, forward glide of water droplets, reverse glide of water droplets, forward circling of water droplets, reverse circling of water droplets, elliptical circling, elliptical forward glide, and elliptical backward glide.

[0012] Preferably, the construction of the spacecraft relative motion configuration dataset specifically includes:

[0013] Under near-circular orbit conditions, introducing the control variables into the CW equations yields the relative motion control equations as follows:

[0014]

[0015] Motion simulations were performed based on the characteristics of each relative configuration, and trajectory data were integrated to construct a spacecraft relative motion configuration dataset.

[0016] The meaning of the symbols is as follows:

[0017] x — the radial position component of the orbit relative to the coordinate system (LVLH system);

[0018] y — the position component of the flight direction in a relative coordinate system (LVLH system);

[0019] z — Position component of the orbital angular momentum direction relative to the coordinate system (LVLH system);

[0020] —The radial velocity component of the orbit in the relative coordinate system (LVLH system);

[0021] —The velocity component in the direction of flight in a relative coordinate system (LVLH system);

[0022] —The velocity component in the direction of the orbital angular momentum in a relative coordinate system (LVLH system);

[0023] —The radial acceleration component of the orbit in the relative coordinate system (LVLH system);

[0024] —Acceleration components in the direction of flight in a relative coordinate system (LVLH system);

[0025] —The acceleration component in the direction of orbital angular momentum in a relative coordinate system (LVLH system).

[0026] Preferably, the step of constructing a neural network model based on the spacecraft relative motion configuration dataset specifically includes:

[0027] S1: Randomly select data from the spacecraft relative motion configuration dataset Dataset to construct the training dataset Train_dataset and the test dataset Test_dataset;

[0028] S2: Normalize the training dataset Train_dataset to obtain training data between [0, 1];

[0029] S3: Use the normalized training data to train the neural network model and update the parameters;

[0030] S4: Test the trained neural network model using the test dataset Test_dataset. If the test passes, the neural network model is completed; otherwise, execute S3.

[0031] Preferably, in step S1, the ratio of the training dataset Train_dataset to the test dataset Test_dataset is 4:1.

[0032] Preferably, in the data normalization process of S2, a set of original data samples includes the following elements, which are divided into two independent vectors: position vector and velocity vector; then, the position vector is processed as follows:

[0033]

[0034]

[0035] Using vector magnitude and unit vector as positional feature information, the tanh function is chosen as the compression function.

[0036]

[0037] Therefore, the modulus is compressed.

[0038]

[0039] Can make

[0040]

[0041] but

[0042]

[0043] In the formula, LM K represents the maximum effective detection range. P The compression factor is 1. L1 represents the compression result of the position vector magnitude; L2 represents the position vector magnitude.

[0044] Preferably, in step S3, the normalized training data is used for neural network model training and parameter updating, specifically as follows:

[0045] The parameters of the neural network model for spacecraft relative motion configuration recognition are initialized. The input data in the training dataset Train_dataset is forward-propagated through the network structure to obtain the output value. The error between the output value and the corresponding labeled data in the labeled dataset labels is calculated using the cross-entropy loss function. Then, the chain rule is used to backpropagate this error to the neural network. The neural network weights and bias parameters are updated using an optimization algorithm.

[0046] This invention also discloses a spacecraft relative motion configuration recognition system based on neural networks, comprising:

[0047] The acquisition unit is used to acquire the position information between the spacecraft and the central spacecraft, and to determine the information and number of input nodes, hidden layer nodes, and output nodes of the neural network model, as well as the activation functions between each layer.

[0048] The data construction unit is used to construct a dataset of the relative motion configuration of the spacecraft based on the positional information between the spacecraft and the central spacecraft.

[0049] The model building unit is used to build a neural network model based on the spacecraft relative motion configuration dataset.

[0050] The identification unit is used to input the motion state information of the target spacecraft into the neural network model to identify the relative motion configuration of the spacecraft.

[0051] Compared with the prior art, the present invention has the following beneficial effects:

[0052] This invention presents a method for identifying the relative motion configuration of non-cooperative spacecraft. It proposes a neural network-based method for spacecraft relative motion configuration identification, which, by constructing a dataset of spacecraft relative motion configurations, enables the identification of relative motion configurations even when domain expert knowledge is insufficient. By using a neural network model to identify the relative motion configuration of target spacecraft, the relative motion configuration of non-cooperative spacecraft can be mastered, thereby providing information-level support for space safety early warning and avoidance. Attached Figure Description

[0053] Figure 1A schematic diagram of a scenario involving a central spacecraft (reference point) S and a non-cooperative spacecraft G;

[0054] Figure 2 This is a schematic diagram of the relative configuration of the spacecraft;

[0055] Figure 3 The training loss curve for the neural network;

[0056] Figure 4 Train the neural network to identify the accuracy data curve;

[0057] Figure 5 This is a schematic diagram of the process of the present invention. Detailed Implementation

[0058] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.

[0059] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this invention are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of the invention described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover a non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0060] The present invention will now be described in further detail with reference to the accompanying drawings:

[0061] Studying the relative motion configurations of spacecraft is a crucial step in understanding space situational information and holds significant theoretical value. Its significance lies primarily in two aspects: First, it allows for the systematic classification of the natural configurations of non-cooperative target spacecraft that pose a potential threat to on-orbit spacecraft, a vital prerequisite for studying the threat posed by relative trajectories. Second, it enables the identification of the relative motion configurations of non-cooperative spacecraft within a specific near-circular orbit, providing a basis for space safety early warning and maneuver avoidance.

[0062] See Figure 5A method for spacecraft relative motion configuration recognition based on neural networks includes the following steps:

[0063] S1. Establish a set of spacecraft relative motion configuration patterns and construct a spacecraft relative motion configuration dataset.

[0064] The specific steps are as follows:

[0065] See Figure 1 In a near-circular orbit, introducing the control variable into the CW equation yields the relative motion control equation:

[0066]

[0067] The meaning of the symbols is as follows:

[0068] x — the radial position component of the orbit relative to the coordinate system (LVLH system);

[0069] y — the position component of the flight direction in a relative coordinate system (LVLH system);

[0070] z — Position component of the orbital angular momentum direction relative to the coordinate system (LVLH system);

[0071] —The radial velocity component of the orbit in the relative coordinate system (LVLH system);

[0072] —The velocity component in the direction of flight in a relative coordinate system (LVLH system);

[0073] —The velocity component in the direction of the orbital angular momentum in a relative coordinate system (LVLH system);

[0074] —The radial acceleration component of the orbit in the relative coordinate system (LVLH system);

[0075] —Acceleration components in the direction of flight in a relative coordinate system (LVLH system);

[0076] —The acceleration component in the direction of orbital angular momentum in a relative coordinate system (LVLH system);

[0077] See Figure 2 The set of spacecraft relative motion configuration patterns is defined as follows:

[0078] (1) Rendezvous: When the minimum distance between a spacecraft and a reference point is less than the critical value ε, this relative motion orbital configuration is called "rendezvous".

[0079] (2) Follow-through: The spacecraft passes through the region near the reference point in a curved configuration, and the distance between the two points at any given time is greater than the critical value ε. (Spacecraft relative velocity components) Greater than zero.

[0080] (3) Reverse drift: The spacecraft passes through the region near the reference point in a curved configuration, and the distance between the two points at any given time is greater than the critical value ε. (Spacecraft relative velocity components) Less than zero.

[0081] (4) Water droplet flying in the forward direction: The spacecraft flies over the area near the spacecraft in a water droplet configuration. During this process, the projection of the target trajectory on the orbital plane is located in the third and fourth quadrants of the xy coordinate system, and the distance between the two at any time is greater than the critical value ε.

[0082] (5) Water droplet reverse fly-by: The spacecraft flies past the area near the spacecraft in a water droplet configuration. During this process, the projection of the target trajectory on the orbital plane is located in the first and second quadrants of the xy coordinate system, and the distance between the two at any time is greater than the critical value ε.

[0083] (6) Droplet-shaped forward orbit: The spacecraft orbits the spacecraft and its surrounding area in a droplet configuration. The mission spacecraft is located between the x-axis boundaries of the target, forming a non-closed orbital trajectory in three-dimensional space, and the distance between the two at any time is greater than the critical value ε. The spacecraft moves forward in a droplet configuration along the positive y-axis.

[0084] (7) Teardrop Reverse Orbit: The spacecraft orbits the spacecraft and its surrounding area in a teardrop configuration. The mission spacecraft is located between the x-axis boundaries of the target spacecraft, forming a non-closed orbital trajectory in three-dimensional space, and the distance between the two spacecraft at any time is greater than the critical value ε. The spacecraft moves in the reverse direction along the negative y-axis in a teardrop configuration.

[0085] (8) Elliptical orbit: The spacecraft surrounds the spacecraft and its surrounding area in an elliptical configuration, forming a closed orbit in three-dimensional space, and the distance between the two sides at any time is greater than the critical value ε.

[0086] (9) Elliptical forward flyby: The spacecraft approaches the spacecraft with an elliptical configuration period. The spacecraft trajectory is located in the positive y-axis direction of the spacecraft coordinate system, and the distance between the two sides at any time is greater than the critical value ε.

[0087] (10) Elliptical backward flyby: The spacecraft approaches the other spacecraft in an elliptical configuration. The spacecraft's trajectory is located in the negative y-axis direction of the spacecraft's coordinate system, and the distance between the two spacecraft at any time is greater than the critical value ε.

[0088] Based on the characteristics of the above relative configurations, motion simulation was performed, trajectory data was integrated, and a spacecraft relative motion configuration dataset was constructed.

[0089] S2 uses a 4:1 ratio of training dataset to test dataset, and randomly selects data from the dataset of different relative motion configurations of spacecraft to construct the training dataset Train_dataset and the test dataset Test_dataset;

[0090] S3 requires normalization of the training data before network training.

[0091] Data normalization transforms all data into values ​​between [0,1] to eliminate magnitude differences between dimensions and prevent large prediction errors caused by significant variations in the input data magnitude. The data normalization method employs a "scale compression" approach to preprocess the original data, compressing it to the range of 0 to 1 without losing feature information. It also considers spatial perception constraints and is correlated with the detection range of onboard equipment. A set of original data samples includes the following elements, which are divided into two independent vectors: position and velocity. The position vector is then processed as follows:

[0092]

[0093]

[0094] In the formula, L M K represents the maximum effective detection range. P The compression factor is 1. L1 represents the compression result of the position vector magnitude; L2 represents the position vector magnitude.

[0095] Vector magnitude and unit vector are used as location feature information. However, since the magnitudes of location vectors vary greatly among different samples, large differences in the magnitudes of different input data for a single data node can lead to network instability. Therefore, a compression function is used to compress the vector magnitude to a range of 0 to 1. Here, the tanh function is chosen as the compression function.

[0096]

[0097] Therefore, the modulus is compressed, where K P Compression factor

[0098]

[0099] Can make

[0100]

[0101] Among them, L M To achieve the maximum effective detection range, then

[0102]

[0103] The preprocessed data not only ensures the boundedness of the feature information, but also processes the velocity vector in the same way. After preprocessing, the input information is increased from 6-dimensional information to 8-dimensional data. Therefore, the number of input nodes for the prediction network model is 8.

[0104] S4, determine the information and number of input nodes, hidden layer nodes, and output nodes of the neural network model, as well as the activation functions between each layer;

[0105] The neural network model uses "relative position" and "relative velocity" information as input nodes. After preprocessing, the input information is upgraded from 6-dimensional to 8-dimensional data; therefore, the prediction network model has 8 input nodes. There are 3 hidden layers, each with 45 nodes. The number of output nodes is the same as the number of relative motion configurations, N. The hidden layer activation function f(·) and the output layer activation function g(·) use the Sigmoid function.

[0106] S5 uses the training data Train_dataset for model training and parameter updates;

[0107] The parameters of the neural network model for spacecraft relative motion configuration recognition are initialized. The input data in the training dataset is forward-propagated through the network structure to obtain the output value. The error between the output value and the corresponding labeled data in the labeled dataset is calculated using the cross-entropy loss function. Then, the error is backpropagated to the neural network using the chain rule. The weights and bias parameters of the neural network are updated using the optimization algorithm to complete one round of learning.

[0108] S6 uses the Test_dataset to test the training effect, identifying the relative motion configuration of the spacecraft based on single-moment motion state information. After training the neural network model parameters (weights and bias parameters), the test dataset is used to evaluate its performance. See also Figure 3 The training loss curve of the neural network shows that the training loss converges, demonstrating the feasibility of this method; see also Figure 4 As shown in the figure, the accuracy curve of the neural network model in recognizing relative motion types can ultimately reach 95%.

[0109] This invention also discloses a spacecraft relative motion configuration recognition system based on neural networks, comprising:

[0110] The acquisition unit is used to acquire the position information between the spacecraft and the central spacecraft, and to determine the information and number of input nodes, hidden layer nodes, and output nodes of the neural network model, as well as the activation functions between each layer.

[0111] The data construction unit is used to construct a dataset of the relative motion configuration of the spacecraft based on the positional information between the spacecraft and the central spacecraft.

[0112] The model building unit is used to build a neural network model based on the spacecraft relative motion configuration dataset.

[0113] The identification unit is used to input the motion state information of the target spacecraft into the neural network model to identify the relative motion configuration of the spacecraft.

[0114] The above content is only for illustrating the technical concept of the present invention and should not be construed as limiting the scope of protection of the present invention. Any modifications made to the technical solution based on the technical concept proposed in this invention shall fall within the scope of protection of the claims of this invention.

Claims

1. A method for spacecraft relative motion configuration recognition based on neural networks, characterized in that, include: Obtain the position information between the spacecraft and the central spacecraft, and determine the information and number of input nodes, hidden layer nodes, and output nodes of the neural network model, as well as the activation functions between each layer. A dataset of relative motion configurations of spacecraft is constructed based on the positional information between the spacecraft and the central spacecraft. Specifically, it includes: S1: Randomly select data from the spacecraft relative motion configuration dataset to construct the training dataset and the test dataset; S2: Normalize the training dataset to obtain training data between [0, 1]; a set of original data samples includes the following elements, which are divided into two independent vectors, namely position vector and velocity; then the position vector is processed as follows: （2） （3） Using vector magnitude and unit vector as positional feature information, the tanh function is chosen as the compression function. （4） Therefore, the modulus is compressed. （5） Can make （6） but （7） In the formula, This is the maximum effective detection range; The compression factor is 1. This is the result of compression of the position vector magnitude; The magnitude of the position vector; S3: Use the normalized training data for neural network model training and parameter updates; specifically: The parameters of the neural network model for the spacecraft relative motion configuration recognition method are initialized. The input data in the training dataset is forward-propagated through the network structure to obtain the output value. The cross-entropy loss function is used to calculate the error between the output value and the corresponding labeled data in the labeled dataset. Then, the chain rule is used to backpropagate this error to the neural network. The optimization algorithm is used to update the weights and bias parameters of the neural network. S4: Test the trained neural network model using the test dataset. If the test passes, the neural network model is built. If the test fails, proceed to S3. A neural network model was constructed based on a dataset of spacecraft relative motion configurations. The motion state information of the target spacecraft is input into a neural network model to identify the relative motion configuration of the spacecraft.

2. The method for spacecraft relative motion configuration recognition based on neural networks according to claim 1, characterized in that, The input nodes of the neural network model are selected as "relative position" and "relative velocity" information, and the number of output nodes is the same as the number of relative motion configurations N; the activation functions of the hidden layer and the output layer use the Sigmoid function.

3. The method for spacecraft relative motion configuration recognition based on neural networks according to claim 1, characterized in that, The relative motion configurations of the spacecraft include: rendezvous, forward drift, reverse drift, forward glide of a droplet, reverse glide of a droplet, forward orbit of a droplet, reverse orbit of a droplet, elliptical orbit, forward glide of an ellipse, and backward glide of an ellipse.

4. The method for spacecraft relative motion configuration recognition based on neural networks according to claim 1, characterized in that, The construction of the spacecraft relative motion configuration dataset specifically includes: Under near-circular orbit conditions, introducing the control variables into the CW equations yields the relative motion control equations as follows: （1） Motion simulations are performed based on the characteristics of each relative configuration, and trajectory data is integrated to construct a dataset of spacecraft relative motion configurations. The meaning of the symbols is as follows: —The radial position component of the orbit in the relative coordinate system; —Position components of the flight direction in a relative coordinate system; —Position component of the orbital angular momentum direction in the relative coordinate system; —The radial velocity component of the track in the relative coordinate system; —The velocity component of the flight direction in the relative coordinate system; —The velocity component in the direction of the orbital angular momentum in the relative coordinate system; —The radial acceleration component of the track in the relative coordinate system; —Acceleration components in the direction of flight in a relative coordinate system; —The acceleration component in the direction of the orbital angular momentum in the relative coordinate system.

5. The method for spacecraft relative motion configuration recognition based on neural networks according to claim 1, characterized in that, In S1, the ratio of the training dataset to the test dataset is 4:

1.

6. A spacecraft relative motion configuration recognition system based on neural networks, characterized in that, The method for recognizing the relative motion configuration of a spacecraft based on a neural network as described in any one of claims 1 to 5 includes: The acquisition unit is used to acquire the position information between the spacecraft and the central spacecraft, and to determine the information and number of input nodes, hidden layer nodes, and output nodes of the neural network model, as well as the activation functions between each layer. The data construction unit is used to construct a dataset of the relative motion configuration of spacecraft based on the positional information between the spacecraft and the central spacecraft. The model building unit is used to build a neural network model based on a dataset of spacecraft relative motion configurations. The identification unit is used to input the motion state information of the target spacecraft into the neural network model to identify the relative motion configuration of the spacecraft.

Images