Multi-drive track behavior intelligent planning method and system

Identify the flight intention of non-cooperative spacecraft through the LSTM layer, and combine the final guidance law of Lambert guidance and BP neural network optimization, solve the problem that our spacecraft is difficult to avoid when the space non-cooperative spacecraft approaches, and realizes intelligent avoidance and efficient maneuvering with smooth tracks.

CN118941101APending Publication Date: 2024-11-12NORTHWESTERN POLYTECHNICAL UNIV
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202410839475.3
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-06-26
Publication Date
2024-11-12

AI Technical Summary

Technical Problem

In the prior art, our spacecraft is difficult to effectively avoid when the space non-cooperative spacecraft approaches, and lacks the ability to intelligently identify enemy maneuver intentions, resulting in passive and inefficient avoidance measures.

Method used

The multi-driven orbital behavior intelligent planning method is adopted to identify the flight intention of non-cooperative spacecraft through the LSTM layer, and combine the Bang-Bang type variable structure final guidance law optimized by Lambert guidance and BP neural network to achieve intelligent maneuvering of the spacecraft.

Benefits of technology

It realizes dynamic identification and efficient avoidance of enemy spacecraft intentions, improves maneuverability, smoother orbits, and enhances the robustness of guidance parameters.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118941101A_ABST
    Figure CN118941101A_ABST
Patent Text Reader

Abstract

The invention discloses a multi-drive track behavior intelligent planning method and system, and belongs to the technical field of aerospace. According to the method, firstly, the intent of an enemy is judged by adopting a long-short-term memory neural network model, then avoidance measures are taken according to the judgment result, and in order to avoid approaching detection of the opposite side, the spacecraft in our part adopts Lambert orbit transfer to be transferred to the vicinity of a target orbit in the mid-guidance stage; the tail section reaches the target orbit by adopting a fuzzy variable structure terminal guidance law optimized by a BP neural network; the method combines a long-short term memory neural network, Lambert guidance and a sliding mode variable structure terminal guidance law optimized by a BP neural network. And an efficient and intelligent orbit avoidance maneuvering strategy is provided for our spacecraft.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention belongs to the field of aerospace technology, and in particular relates to a multi-drive track behavior intelligent planning method and system. Background Art

[0002] Space security is an important part of aerospace security. In recent years, space security has been constantly challenged, and there is an urgent need to build a complete and intelligent orbital behavior planning system to gain an advantage in space game confrontation. Most of the existing space avoidance is passive avoidance, which is implemented after the non-cooperative spacecraft has approached our spacecraft. The maneuvering intention of the non-cooperative spacecraft is not identified before avoidance. The space game situation is complex and changeable, and the intelligent identification of the maneuvering intention of the non-cooperative spacecraft is particularly important. Summary of the invention

[0003] The purpose of the present invention is to overcome the shortcomings of the above-mentioned prior art and to provide a multi-drive orbital behavior intelligent planning method and system to solve the problem in the prior art that once an enemy spacecraft approaches to detect our spacecraft, it is difficult for our spacecraft to avoid it.

[0004] In order to achieve the above object, the present invention adopts the following technical solutions:

[0005] A multi-actuated track behavior intelligent planning method comprises the following steps:

[0006] S1, based on the three types of features of the non-cooperative spacecraft, the flight intention of the non-cooperative spacecraft is obtained through the LSTM layer, and the tail of the LSTM layer is connected to a classifier; the three types of features of the non-cooperative spacecraft are relative position, relative speed and relative distance, and the flight intention is to mix, fly around or fly by;

[0007] S2, based on the flight intention of the non-cooperative spacecraft, confirm the final target point and intermediate target point of our spacecraft;

[0008] S3, through Lambert guidance, determine the velocity increment of the initial ignition of our spacecraft, the velocity increment of the ignition when reaching the intermediate target point, and the intermediate guidance trajectory;

[0009] S4, using the Bang-Bang type variable structure guidance law as the terminal guidance law of our spacecraft, and our spacecraft transfers from the intermediate target point to the final target point according to the terminal guidance law; the Bang-Bang type variable structure guidance law is optimized by BP neural network.

[0010] A further improvement of the present invention is:

[0011] Preferably, in step 1, in S1, based on the three types of features of the non-cooperative spacecraft, the specific process of obtaining the flight intention of the non-cooperative spacecraft through the LSTM layer is:

[0012] (1) After normalizing the three types of features, a dataset of temporal network input features is constructed;

[0013] (2) The seven-dimensional features in the dataset are input into the LSTM layer. The LSTM layer extracts spatial and temporal features from the seven-dimensional features and inputs the extracted features into the Softmax classifier to obtain the flight intention of the non-cooperative spacecraft.

[0014] Preferably, in S2, the intermediate target point is any point within a spherical range with the final target point as the center and a radius of 300-400 km.

[0015] Preferably, the specific process of S3 is: based on the initial position and the intermediate target point position of our spacecraft, a short-range orbit or a long-range orbit is obtained, and the initial point and the intermediate point speeds are obtained by calculation, thereby obtaining the initial acceleration and the intermediate target point acceleration of our spacecraft;

[0016] Through the initial position and initial velocity of our spacecraft, or the position and velocity of the intermediate target point, we calculate the orbital elements of the orbit transfer and then determine the trajectory of the mid-course guidance.

[0017] Preferably, in S3, the speed of the initial point and the speed of the intermediate point are:

[0018]

[0019] in, is the velocity at the initial point, is the speed of the midpoint, is the initial point position, is the position of the intermediate target point, l is the Lagrange number, g is the gravitational acceleration, k is the Lagrange coefficient, is the derivative of k.

[0020] Preferably, in S3, for short-distance tracks:

[0021]

[0022] For the "long-distance track",

[0023]

[0024] in,

[0025]

[0026] Preferably, in S4, the formula of the Bang-Bang type variable structure terminal guidance law is:

[0027]

[0028] Among them, a y is the acceleration of our spacecraft along the oy1 axis of the stellar coordinate system, a z is the acceleration of our spacecraft along the oz1 axis of the stellar coordinate system, M=const>0, is the acceleration generated by the thrust of the ideal orbit control engine on the satellite, δ ε is the on / off threshold of the orbit control engine in the oy1-axis direction of our spacecraft, δ β is the switch threshold of the orbit control engine in the oz1 axis direction of our spacecraft, and are the two normal components of the line-of-sight angular velocity between our spacecraft and the final target point.

[0029] Preferably, when the Bang-Bang variable structure guidance law is used as the terminal guidance law of our spacecraft, the control line of sight angular rate is input Approaching 0.

[0030] Preferably, the input of the BP neural network is the line of sight angular velocity of our spacecraft The two normal components of and Our spacecraft velocity v and line of sight angle q; the output variable is our spacecraft acceleration a w

[0031] A multi-actuated track behavior intelligent planning system, comprising:

[0032] A flight intention determination module is used to obtain the flight intention of the non-cooperative spacecraft through an LSTM layer based on three types of features of the non-cooperative spacecraft, and a classifier is connected to the tail of the LSTM layer; the three types of features of the non-cooperative spacecraft are relative position, relative speed and relative distance, and the flight intention is to mix, fly around or skim;

[0033] The target point determination module is used to confirm the final target point and intermediate target point of our spacecraft based on the flight intention of the non-cooperative spacecraft;

[0034] The mid-course guidance module is used to determine the velocity increment of the initial ignition of our spacecraft, the velocity increment of the ignition when reaching the intermediate target point, and the mid-course guidance trajectory through Lambert guidance;

[0035] The terminal guidance module is used to use the Bang-Bang type variable structure guidance law as the terminal guidance law of our spacecraft. Our spacecraft transfers from the intermediate target point to the final target point according to the terminal guidance law; the Bang-Bang type variable structure guidance law is optimized by BP neural network.

[0036] Compared with the prior art, the present invention has the following beneficial effects:

[0037] The invention discloses a multi-driven orbit behavior intelligent planning method. In order to dynamically identify the enemy's combat intention, the invention first adopts a long short-term memory neural network model to judge the enemy's intention, and then takes avoidance measures according to the judgment result, and establishes a long short-term memory neural network (LSTM) model. After completing the situation classification of the other party's spacecraft, in order to avoid the other party's approaching reconnaissance, our spacecraft first adopts Lambert orbit transfer to the vicinity of the target orbit, and the fuzzy variable structure terminal guidance law optimized by BP neural network is adopted in the terminal section to reach the target orbit; the method combines the long short-term memory neural network, Lambert guidance and the sliding mode variable structure terminal guidance law optimized by BP neural network to provide our spacecraft with an efficient and intelligent orbit avoidance maneuvering strategy, improves the efficiency of avoidance maneuvering and makes the orbit smoother. The method of the invention first intelligently identifies the intention of non-cooperative spacecraft, and then actively avoids according to the identification result, thereby destroying the approaching intention of non-cooperative spacecraft and taking avoidance measures in time.

[0038] Furthermore, in response to the accompanying, circling and flyby maneuvers taken by the non-cooperative spacecraft to achieve close reconnaissance and imaging of our spacecraft, the relative speed, relative position, solar incidence angle, observation distance change and other motion characteristics of the non-cooperative spacecraft and the corresponding intention labels are input as samples into the long short-term memory neural network model for training, so as to realize the memory function of the time series features, and use the trained model to predict the maneuvering intentions of the non-cooperative spacecraft, thereby completing the space situational awareness classification.

[0039] Furthermore, Lambert transfer guidance is used as the mid-stage strategy, and BP neural network optimized sliding mode variable structure terminal guidance is used as the terminal strategy. Simulation verification shows that this scheme can quickly maneuver to avoid the opponent's reconnaissance and enhance the robustness of guidance parameter changes. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] Figure 1 LSTM neural network model diagram used in the present invention;

[0041] Figure 2 This is a diagram of the LSTM-based situation classification model architecture of the present invention;

[0042] Figure 3This is a diagram of the LSTM-based situation classification model architecture of the present invention;

[0043] Figure 4 It is a variable structure terminal guidance law diagram based on BP neural network optimization of the present invention;

[0044] Figure 5 This is a flow chart of the multi-drive track behavior intelligent planning method of the present invention.

[0045] Figure 6 A schematic diagram of the guidance of the intelligent planning of the track behavior of multiple drives of the present invention. DETAILED DESCRIPTION

[0046] See also Figure 5 and Figure 6 The first aspect of the present invention discloses a multi-drive track behavior intelligent planning method, which specifically includes the following steps:

[0047] Step 1: Build a database of non-cooperative spacecraft maneuvers in space, pre-process the data and perform neural network training.

[0048] Firstly, five typical maneuvering intentions in space game scenarios are selected, namely, straight-line flyby, droplet flyby, fixed-point follow-up, ellipse enclosed follow-up, and ellipse non-enclosed follow-up. Among the numerous simulation data, three types of space non-cooperative spacecraft features related to maneuvering intentions are selected as classification data: relative position, relative speed, and relative distance.

[0049] Relative position: the three-dimensional relative position of the non-cooperative spacecraft relative to our spacecraft in the VVLH coordinate system;

[0050] Relative speed: the three-dimensional relative speed of the non-cooperative spacecraft relative to our spacecraft in the VVLH coordinate system;

[0051] Relative distance: the relative distance between non-cooperative spacecraft and our spacecraft.

[0052] It mainly includes two parts:

[0053] (1) Data preprocessing: In order to make the simulation data meet the input requirements of the model, the data needs to be preprocessed. The data includes the relative position, relative speed and relative distance mentioned above. The data preprocessing mainly includes normalization and constructing a data set of the input features of the time series network.

[0054] (2) The seven-dimensional features [3D relative position, 3D relative velocity, relative distance] corresponding to the processed data samples are input into the long short-term memory neural network. The spatial and temporal features of the data are extracted by using the structural characteristics of the LSTM layer. A Softmax classifier is connected after the LSTM layer to convert the maneuver intention recognition problem into a classification problem, and the recognition result is obtained, thereby obtaining whether the intention of the non-cooperative spacecraft is to mix, fly around, or fly by. Among them, the LSTM layer is a network trained with simulation data in the early stage.

[0055] See also Figure 1 , Long Short-Term Memory Network (LSTM) is a variant of the recurrent neural network, which introduces a set of memory cells and allows the network to learn to choose when to discard historical information and when to use new information to update memory cells. Therefore, LSTM can effectively learn the time-varying laws in the data features of time-series changes, and is a good solution to the problem of real-time situation recognition. Through LSTM, the situation recognition problem is effectively converted into a multi-classification problem based on time-series motion features. Its model is as follows Figure 1 shown.

[0056] First, we introduce the forget gate, as shown below:

[0057] f t =σ(W f ·[h t-1 ,x t ]+b f ) (1)

[0058] Where: f t Represents the output of the forget gate; W f is the weight matrix of the forget gate; h t-1 is the output of the previous moment; x t Indicates the input at the current moment; [h t-1 ,x t ] connects two vectors into a longer vector; b f is the bias term of the forget gate; σ is the sigmoid function.

[0059] If the input dimension is d x , the hidden layer (i.e. output) dimension is d h , the unit state dimension is d c , the weight matrix W of the forget gate f The dimensions are:

[0060] f=d c ×(d h +d x ) (2)

[0061] For the input gate:

[0062] i t =σ(W i ·[h t-1 ,x t ]+b i ) (3)

[0063] Where: W i is the weight matrix of the input gate; b i is the bias term of the input gate; i t It will determine how much of the information flow needs to be integrated at this moment.

[0064] Before deciding which information needs to be fused, the input needs to be processed first. For the processing of input units:

[0065]

[0066] Where: W c is the weight matrix of the memory unit; b c Represents the bias term of the memory unit.

[0067] According to the above calculation, the current output stream is:

[0068]

[0069] In this way, the current information and long-term memory c t-1 The forget gate ensures that the previous information can be used at this moment, and the input gate controls which current information needs to be retained to prevent useless information from entering the memory.

[0070] The output gate is represented as:

[0071] o t =σ(W o ·[h t-1 ,x t ]+b o ) (6)

[0072] Where: W o is the weight matrix of the output gate; b o is the bias term of the output gate.

[0073] The final output is composed of the output gate and the current state c t Jointly decided:

[0074] h t =o t ·tanh(c t ) (7)

[0075] The current state c tThen flow into the next moment.

[0076] See also Figure 2 ,h t The Softmax function is added at the end. The purpose of this function is to normalize the direct output of the model and convert the situation classification problem into a multi-classification probability maximization solution:

[0077]

[0078] Where: j is the number of categories; i is the category. The purpose is to make the sum of the output category probabilities equal to 1.

[0079] Step 2, complete the mid-range guidance through Lambert guidance.

[0080] The target transfer point is divided into an intermediate target point and a final target point. The intermediate target point is the switching point between the intermediate guidance strategy and the final guidance strategy, and is also the target point of the intermediate guidance strategy. The intermediate target point is within a spherical range with the final target point as the center and a radius of 300-400km.

[0081] According to the identification results, our spacecraft takes maneuvering measures to evade the non-cooperative spacecraft in order to destroy its intention. The ground control station selects the final target point designated by the ground control station according to the intention of the non-cooperative spacecraft. For the non-cooperative spacecraft's related flyby intention, our spacecraft can perform in-plane maneuvers to raise or lower the orbit within the orbital plane; for the non-cooperative spacecraft's related following intention, our spacecraft can perform out-of-plane maneuvers to change the orbital plane and get rid of the non-cooperative spacecraft's following. Figure 3 , then select an intermediate target point that meets the conditions near the final target point and determine the transfer time Δt required by Lambert. After selecting the intermediate target point, our spacecraft enters the mid-range guidance process, which uses Lambert guidance to guide it into an orbit containing the target transfer point (i.e., the intermediate target point). By using Lambert guidance, the accuracy of our spacecraft when reaching the intermediate target point can be improved.

[0082] The initial position of our spacecraft is known and speed It is required that after a flight time of Δt, the position of our spacecraft at the moment when it reaches the intermediate target point and speed When Δt is known, two transition orbits passing through these two points can be determined, namely "short-range orbit" or "long-range orbit". It is required to determine the velocity increments Δv1 and Δv2 of our spacecraft at the initial and terminal ignitions as well as the trajectory of the mid-range guidance.

[0083] Solve the Lambert problem. There are many mature theories and methods for solving the Lambert problem. The more commonly used methods are the Gauss method, the universal variable method, etc. This solution uses the universal variable method to solve the Lambert problem. The specific process is:

[0084] (1) Calculate r1 and r2, we have

[0085]

[0086] (2) Select “Short-range track” and “Long-range track” and calculate the central angle Δf of r1 and r2;

[0087] For "short-distance tracks",

[0088]

[0089] For the "long-distance track",

[0090]

[0091] in,

[0092]

[0093] (3) Calculate the intermediate variable A,

[0094]

[0095] Among them, the intermediate variable A has no actual physical meaning.

[0096] (4) Select the initial value of the universal variable z (z = -100, usually z is taken from negative

[0097] →(2π) 2 ).

[0098] (5) Using the initial value of z, we can iterate and calculate the following formula to find the value of z:

[0099]

[0100] In order to find the z value, use Newton's iteration method to establish the function:

[0101]

[0102] in

[0103]

[0104] Until the difference between the solved Δt value and the given flight time is within the error range (set to 10 -8 ) and take the z value at that time.

[0105] (6) Calculate the Lagrangian series l and k, and and

[0106]

[0107]

[0108] (7) From this, the speed of the initial point and the intermediate target point terminal point can be calculated

[0109]

[0110] (8) Our spacecraft is initially and terminal ignition speed They are:

[0111]

[0112] (9) By and or and The six orbital elements of the transfer orbit can be calculated, and the trajectory of the mid-course guidance can be determined.

[0113] Step 3: After our spacecraft completes the mid-range guidance and reaches the intermediate target point, and the relative distance and relative line-of-sight angular velocity between our spacecraft and the target transfer point can meet the handover conditions for mid-range guidance and terminal guidance, our spacecraft enters the terminal guidance stage and reaches the target transfer point through the terminal guidance process.

[0114] When our spacecraft transfers to the intermediate target point on the target orbit 300km away from the target transfer point, the mid-range guidance ends and enters the terminal guidance. First, the Bang-Bang variable structure guidance law along the line of sight system is given. The Bang-Bang variable structure terminal guidance law adopts the guidance strategy of zeroing the line of sight rotation rate.

[0115]

[0116]

[0117] Among them, a y is the acceleration of the spacecraft along the oy1 axis of the stellar coordinate system, a z is the acceleration of the spacecraft along the oz1 axis of the stellar coordinate system, M=const>0 is the acceleration generated by the thrust of the ideal orbit control engine on the satellite, δ ε is the on / off threshold of the orbit control engine in the oy1-axis direction of the spacecraft, δ βis the switch threshold of the orbit control engine in the oz1-axis direction of the spacecraft, and are the two normal components of the line-of-sight angular velocity between our spacecraft and the final target point:

[0118]

[0119] Among them, Δx, Δy, Δz, They are the projections of the relative positions and velocities of the satellite and the target transfer point in the launch inertial system.

[0120] The key to designing the guidance law is to control the line-of-sight angular rate through the input of the control variable. Approaches 0. According to the dynamic equation of line of sight rotation:

[0121]

[0122] in, R is the relative distance between the satellite and the target transfer point. The switch function is selected as Considering the limited control force, we have k2≤u≤k1, and k2<0<k1. For the needs of subsequent calculations, we set f2≤w≤f1. According to the arrival condition of the sliding mode And using the Bang-Bang variable structure method, it is calculated that:

[0123]

[0124] As long as equation (35) is satisfied, all points on the sliding mode s = 0 are end points, that is,

[0125] Step 6: Optimization of variable structure terminal guidance law based on BP neural network.

[0126] See also Figure 4 , BP neural network can be divided into two parts, namely the forward propagation of information and the back propagation of errors. BP neural network can learn and store a large number of input-output mode mapping relationships under the premise that the mapping relationship between input and output cannot be revealed. It uses the steepest descent method to continuously adjust the weights and thresholds of the network through back propagation to minimize the sum of squares of the network. BP neural network has strong nonlinear mapping capabilities and flexible network structure. The number of layers in the middle of the network and the number of neurons in each layer can be set according to the situation. Different structures have different performances.

[0127] In the network structure design, a single hidden layer BP neural network is used. There are three input variables: line of sight angular velocity The two normal components of and Our spacecraft velocity v and line of sight angle q; the output variable is our spacecraft acceleration a w , the acceleration a w Includes a y and a z Although the above terminal guidance law can enable our spacecraft to reach the final target point accurately, it requires our spacecraft to have strong maneuverability; neural networks happen to have strong nonlinear fitting capabilities. Using BP neural networks to optimize the terminal guidance law can reduce the overload of our spacecraft and make the terminal guidance trajectory as smooth as possible. The BP neural network is obtained after training and optimization through simulation data before use.

[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 principle of the present invention should be included in the protection scope of the present invention.

Claims

1. A multi-driven track behavior intelligent planning method, characterized in that: The following steps are involved: S1, based on the three types of features of the non-cooperative spacecraft, the flight intention of the non-cooperative spacecraft is obtained through the LSTM layer, and the tail of the LSTM layer is connected to a classifier; the three types of features of the non-cooperative spacecraft are relative position, relative speed and relative distance, and the flight intention is to mix, fly around or fly by; S2, based on the flight intention of the non-cooperative spacecraft, confirm the final target point and intermediate target point of our spacecraft; S3, through Lambert guidance, determine the velocity increment of the initial ignition of our spacecraft, the velocity increment of the ignition when reaching the intermediate target point, and the intermediate guidance trajectory; S4, using the Bang-Bang type variable structure guidance law as the terminal guidance law of our spacecraft, and our spacecraft transfers from the intermediate target point to the final target point according to the terminal guidance law; the Bang-Bang type variable structure guidance law is optimized by BP neural network.

2. A multi-drive track behavior intelligent planning method according to claim 1, characterized in that: In S1, based on the three types of features of non-cooperative spacecraft, the specific process of obtaining the flight intention of non-cooperative spacecraft through the LSTM layer is as follows: (1) After normalizing the three types of features, a dataset of temporal network input features is constructed; (2) The seven-dimensional features in the dataset are input into the LSTM layer. The LSTM layer extracts spatial and temporal features from the seven-dimensional features and inputs the extracted features into the Softmax classifier to obtain the flight intention of the non-cooperative spacecraft.

3. A multi-drive track behavior intelligent planning method according to claim 2, characterized in that: In S2, the intermediate target point is any point within a spherical range with the final target point as the center and a radius of 300-400 km.

4. The multi-drive track behavior intelligent planning method according to claim 1, characterized in that: The specific process of S3 is: based on the initial position of our spacecraft and the position of the intermediate target point, a short-range orbit or a long-range orbit is obtained, and the initial point and intermediate point velocities are obtained by calculation, thereby obtaining the initial acceleration of our spacecraft and the acceleration of the intermediate target point; Through the initial position and initial velocity of our spacecraft, or the position and velocity of the intermediate target point, we calculate the orbital elements of the orbit transfer and then determine the trajectory of the mid-course guidance.

5. The multi-drive track behavior intelligent planning method according to claim 4, characterized in that: In S3, the speed of the initial point and the speed of the intermediate point are: in, is the velocity at the initial point, is the speed of the midpoint, is the initial point position, is the position of the intermediate target point, l is the Lagrange number, g is the gravitational acceleration, k is the Lagrange coefficient, is the derivative of k.

6. The multi-drive track behavior intelligent planning method according to claim 4, characterized in that: In S3, for short-range tracks: For "long-distance tracks", in, 7. The multi-drive track behavior intelligent planning method according to claim 1, characterized in that: In S4, the formula of the Bang-Bang variable structure terminal guidance law is: Among them, a y is the acceleration of our spacecraft along the oy1 axis of the stellar coordinate system, a z is the acceleration of our spacecraft along the oz1 axis of the stellar coordinate system, M = const > 0, is the acceleration generated by the thrust of the ideal orbit control engine acting on the satellite, δ ε is the on / off threshold of the orbit control engine in the oy1-axis direction of our spacecraft, δ β is the switch threshold of the orbit control engine in the oz1 axis direction of our spacecraft, and are the two normal components of the line-of-sight angular velocity between our spacecraft and the final target point.

8. The multi-drive track behavior intelligent planning method according to claim 1, characterized in that: When the Bang-Bang variable structure guidance law is used as the terminal guidance law of our spacecraft, the input control line of sight angular rate Approaching 0.

9. The method for intelligent planning of track behavior of multiple drivers according to claim 1, characterized in that: The input of the BP neural network is the line of sight angular velocity of our spacecraft The two normal components of and Our spacecraft velocity v and line of sight angle q; the output variable is our spacecraft acceleration a w .

10. A multi-driven track behavior intelligent planning system, characterized in that: include: A flight intention determination module is used to obtain the flight intention of the non-cooperative spacecraft through an LSTM layer based on three types of features of the non-cooperative spacecraft, and a classifier is connected to the tail of the LSTM layer; the three types of features of the non-cooperative spacecraft are relative position, relative speed and relative distance, and the flight intention is to mix, fly around or skim; The target point determination module is used to confirm the final target point and intermediate target point of our spacecraft based on the flight intention of the non-cooperative spacecraft; The mid-course guidance module is used to determine the velocity increment of the initial ignition of our spacecraft, the velocity increment of the ignition when reaching the intermediate target point, and the mid-course guidance trajectory through Lambert guidance; The terminal guidance module is used to use the Bang-Bang type variable structure guidance law as the terminal guidance law of our spacecraft. Our spacecraft transfers from the intermediate target point to the final target point according to the terminal guidance law; the Bang-Bang type variable structure guidance law is optimized by BP neural network.