A spacecraft safe orbit planning method based on debris position estimation

Abstract

The present invention belongs to the field of aerospace science and technology, and specifically relates to a method for planning a safe orbit of a spacecraft based on debris position estimation, comprising the following steps: S1: a debris position estimation model based on an LSTM neural network; S2, optimizing the debris position estimation model parameters based on a gradient descent method; S3, the spacecraft performs safe orbit planning using an artificial potential field method; S4, simulating related scenarios. In step S1, an LSTM neural network for real-time prediction of debris trajectories is also designed. The structure of the LSTM neural network is divided into an input layer, an LSTM layer, and an output layer. The input layer is used to receive historical data of debris trajectories, as follows: x(t)=[r d (t‑2p)r d (t‑p)r d (t)] T Among them, r d (t) represents the fragment position at time t, r d (t‑p) represents the fragment position at time t‑p, r d (t‑2p) represents the debris position at time t‑2p, and p represents the prediction interval. The present invention enables a spacecraft to make rapid and autonomous avoidance decisions in a complex and dynamic space environment by using its own sensors and computing resources.

Description

Technical Field

[0001] The present invention belongs to the field of aerospace science and technology, and in particular relates to a method for planning a safe orbit of a spacecraft based on debris position estimation. Background Art

[0002] As the spacecraft's operating cycle lengthens, the risk of a spacecraft running out of service life or accidentally colliding with other space objects, causing it to disintegrate and eventually form space debris, increases. These fragments can seriously affect the safety of spacecraft in orbit, and it is necessary to take certain measures to prevent debris from colliding with spacecraft.

[0003] In the existing technology, in order to solve the problem that space debris seriously threatens the safe operation of spacecraft, the traditional general network strategy is used to capture stationary or tumbling debris in orbit, but this method can only capture cataloged space debris. In addition, the adaptive feedback control strategy based on potential function and dynamic sliding manifold is used to avoid threats, but this method is computationally complex and the computing power of the spacecraft is limited.

[0004] Artificial Potential Field (APF) is a commonly used path planning method, which is particularly suitable for the problem of autonomous threat avoidance of spacecraft in dynamic environments.

[0005] In APF, space debris is considered a threat with a repulsive field, while the target location generates an attractive field. Under the combined effect of these potential fields, the spacecraft will move in the direction of decreasing potential energy, thereby bypassing the debris and moving towards the target. The advantage of this method is its simplicity and intuitiveness, which enables the spacecraft to use its own sensors and computing resources to make rapid and autonomous avoidance decisions in a complex and dynamic space environment. Summary of the invention

[0006] The purpose of the present invention is to provide a spacecraft safe orbit planning method based on debris position estimation, which can enable the spacecraft to use its own sensors and computing resources to achieve rapid and autonomous avoidance decisions in a complex and dynamic space environment.

[0007] The technical solution adopted by the present invention is as follows:

[0008] A spacecraft safe orbit planning method based on debris position estimation comprises the following steps:

[0009] S1: Debris location estimation model based on LSTM neural network;

[0010] S2, optimizing the debris position estimation model parameters based on the gradient descent method;

[0011] S3, spacecraft use artificial potential field method to plan safe orbit;

[0012] S4. Related scenario simulation.

[0013] In the step S1, an LSTM neural network is also designed for real-time prediction of debris trajectories, including an input layer, an LSTM layer and an output layer of the LSTM neural network.

[0014] The input layer, which is used to receive debris trajectory history data, is as follows:

[0015] x(t)=[r d (t-2p)r d (tp)r d (t)] T (35)

[0016] Among them, r d (t) represents the fragment position at time t, r d (tp) represents the fragment position at time tp, r d (t-2p) represents the fragment position at time t-2p, and p represents the prediction interval.

[0017] The LSTM layer is composed of several LSTM units, each of which includes three gate control structures: input gate I, forget gate f, output gate o and 1 LSTM unit state C, which are calculated as follows:

[0018] I(t)=sig(W I x(t)+R I h(t-1)+B I ) (36)

[0019] f(t)=sig(W f x(t)+R f h(t-1)+B f ) (37)

[0020] o(t)=sig(W o x(t)+R o h(t-1)+B o ) (38)

[0021] C(t)=tanh(W C x(t)+R C h(t-1)+B C ) (39)

[0022] Among them, h(t-1) is the output of the LSTM unit at time t-1, W I is the input matrix of the input gate, R I is the circulant matrix of the input gate, B I is the bias vector of the input gate, Wf is the input matrix of the forget gate, R f is the circulant matrix of the forget gate, B f is the bias vector of the forget gate, W o is the input matrix of the output gate, R o is the circulant matrix of the output gate, B o is the bias vector of the output gate, W C is the input matrix of the state, R C is the circulant matrix of the state, B C is the bias vector of the state,

[0023] Moreover, sig and tanh are both gating functions, and their expressions are as follows:

[0024]

[0025]

[0026] Where, exp = 2.7183;

[0027] The output value of the LSTM unit at time t is:

[0028] h(t)=tanh(c(t))⊙o(t) (42)

[0029] Among them, ⊙ is the Hadamard product operation, c(t) is the memory vector of the LSTM unit at time t, which can be expressed as:

[0030] c(t)=f(t)⊙c(t-1)+I(t)⊙C(t) (43)

[0031] Among them, c(t-1) is the memory vector of the LSTM unit at time t-1.

[0032] The output layer is a linear combination of the output values ​​of the LSTM units, i.e., the predicted position of the fragment at time t+p

[0033] y(t)=Wh(t) (44)

[0034] Among them, W is the output weight.

[0035] The step S2 specifically includes the following steps:

[0036] Calculate the loss function L of the LSTM neural network:

[0037]

[0038] Update LSTM neural network parameters W I , R I , BI , W f , R f , B f , W o , R o , B o , W C , R C , B C , W is updated based on the gradient, and the partial derivatives of L with respect to each parameter are:

[0039]

[0040]

[0041]

[0042]

[0043]

[0044]

[0045]

[0046]

[0047]

[0048]

[0049]

[0050]

[0051]

[0052] in, T represents the transpose operation of a matrix or vector, and 1 represents the all-1 matrix or vector of the corresponding matrix or vector operation.

[0053] The step S3 specifically comprises the following steps:

[0054] M1: Determine whether the spacecraft needs to plan a safe orbit or not;

[0055] M2: Calculate the attractive potential function U generated by the given orbit att (t), then, executing step M4;

[0056] M3: The spacecraft remains on the predetermined orbit, and then executes the step M4;

[0057] M4: Let t=t+1, and then go to step M1.

[0058] The step M1 also includes the following steps:

[0059]

[0060] Where P is a constant and is as follows:

[0061]

[0062] d(t)=||r s (t)-r d (t)|| (61)

[0063] Among them, min(·) means taking the smaller operation, r s (t) represents the position of the spacecraft at time t, d(t) represents the relative distance between the spacecraft and the debris at time t, ||·|| is the binary norm operation, represents the predicted relative distance between the spacecraft and the debris at time t+p, calculated according to formulas (35)-(44);

[0064] Prior to this, the relevant parameters of the LSTM neural network are updated according to formulas (45)-(58);

[0065] If the planning conditions are met, min (t)≤P, then execute the step M2; or if the planning condition d is not met min (t)≤P, execute step M3.

[0066] The attractive potential function U att (t), as follows:

[0067]

[0068] Among them, d s (t)=||r s A(t)-r(t)|| represents the distance between the actual position of the spacecraft at time t and the predetermined orbit, r(t) is the position of the spacecraft at time t. B is the threshold of the attraction transition zone, α is the attraction coefficient, which is specifically expressed as:

[0069]

[0070] Among them, β ∈ (0,1);

[0071] Calculate the gradient of the attractive potential function ▽U att (t), as follows:

[0072]

[0073] And, calculate the repulsive force potential function U generated by the fragments rep (t), as follows:

[0074]

[0075] Where, d(t) = ||r s (t)-r d (t)|| represents the distance between the actual position of the spacecraft and the debris at time t, R B is the threshold of the repulsive force transition zone, γ is the repulsive force coefficient;

[0076] Calculate the gradient of the repulsive force potential function as follows:

[0077]

[0078] The planned spacecraft position is r s (t+1), as follows:

[0079]

[0080] Then, execute the step M4.

[0081] After the spacecraft remains on the predetermined orbit in step M3, the planned spacecraft position is r s (t+1), as follows:

[0082] r s (t+1)=r(t+1) (68)

[0083] Then, execute the step M4.

[0084] The technical effects achieved by the present invention are:

[0085] The present invention discloses a spacecraft safe orbit planning method based on debris position estimation, designs a debris trajectory prediction model based on LSTM (long short-term memory) neural network, defines "active avoidance boundary" as a key decision parameter for performing evasive action, and proposes an improved piecewise attraction coefficient under the framework of artificial potential field method to optimize the transfer trajectory of the spacecraft, making the transfer trajectory smoother, thereby realizing online planning of the transfer trajectory when the spacecraft faces the threat of debris.

[0086] The present invention provides a spacecraft safe orbit planning method based on debris position estimation, which can provide the spacecraft with more sufficient avoidance time, reduce unnecessary large-scale debris avoidance actions, and alleviate on-orbit execution pressure.

[0087] The present invention provides a spacecraft safe orbit planning method based on debris position estimation, which needs to update the relevant parameters of the LSTM neural network according to formulas (45)-(58), has a self-regulating effect, can correct in real time, and reduce deviations.

[0088] The invention discloses a spacecraft safe orbit planning method based on debris position estimation. The safe orbit and the relative distance between the spacecraft and the debris obtained by the method of the invention are easy to calculate and can save the computing power of the spacecraft. BRIEF DESCRIPTION OF THE DRAWINGS

[0089] Figure 1 It is a flow chart of a method for planning safe orbit of a spacecraft based on debris position estimation of the present invention;

[0090] Figure 2 is a structural diagram of a debris position estimation model of the present invention;

[0091] Figure 3 is a schematic diagram of a spacecraft avoiding debris using the orbit planning method of the present invention;

[0092] Figure 4 The present invention is a curve diagram showing the relative distance variation between a spacecraft and debris using the orbit planning method of the present invention. DETAILED DESCRIPTION

[0093] In order to make the purpose and advantages of the present invention more clearly understood, the present invention is specifically described below in conjunction with embodiments. It should be understood that the following text is only used to describe one or several specific embodiments of the present invention, and does not strictly limit the scope of protection of the specific claims of the present invention.

[0094] like Figure 1-4 As shown, a spacecraft safe orbit planning method based on debris position estimation includes the following steps:

[0095] S1: Fragment location estimation model based on LSTM neural network, as follows:

[0096] like Figure 2 As shown in the figure, an LSTM neural network (Long Short-Term Memory) is designed for real-time prediction of debris trajectories. The structure of the LSTM neural network is divided into an input layer, an LSTM layer, and an output layer, specifically:

[0097] The input layer, which is used to receive the debris trajectory history data, is as follows:

[0098] x(t)=[r d (t-2p)r d (tp)r d (t)] T (35)

[0099] Among them, rd (t) represents the fragment position at time t, r d (tp) represents the fragment position at time tp, r d (t-2p) represents the fragment position at time t-2p, and p represents the prediction interval;

[0100] The LSTM layer consists of several LSTM units. Each LSTM unit includes three gate structures: input gate I, forget gate f, output gate o and 1 LSTM unit state C. Their calculation method is:

[0101] I(t)=sig(W I x(t)+R I h(t-1)+B I ) (36)

[0102] f(t)=sig(W f x(t)+R f h(t-1)+B f ) (37)

[0103] o(t)=sig(W o x(t)+R o h(t-1)+B o ) (38)

[0104] C(t)=tanh(W C x(t)+R C h(t-1)+B C ) (39)

[0105] Among them, h(t-1) is the output of the LSTM unit at time t-1, W I is the input matrix of the input gate, R I is the circulant matrix of the input gate, B I is the bias vector of the input gate, W f is the input matrix of the forget gate, R f is the circulant matrix of the forget gate, B f is the bias vector of the forget gate, W o is the input matrix of the output gate, R o is the circulant matrix of the output gate, B o is the bias vector of the output gate, W C is the input matrix of the state, R C is the circulant matrix of states, B C is the bias vector of the state,

[0106] Moreover, sig and tanh are both gating functions, and their expressions are as follows:

[0107]

[0108]

[0109] Where, exp = 2.7183;

[0110] The output value of the LSTM unit at time t is:

[0111] h(t)=tanh(c(t))⊙o(t) (42)

[0112] Among them, ⊙ is the Hadamard product operation, c(t) is the memory vector of the LSTM unit at time t, which can be expressed as:

[0113] c(t)=f(t)⊙c(t-1)+I(t)⊙C(t) (43)

[0114] Among them, c(t-1) is the memory vector of the LSTM unit at time t-1;

[0115] The output layer is a linear combination of the output values ​​of the LSTM units, i.e., the predicted position of the fragment at time t+p

[0116] y(t)=Wh(t) (44)

[0117] Where W is the output weight;

[0118] S2. Optimize the parameters of the debris position estimation model based on the gradient descent method, as follows:

[0119] Calculate the loss function L of the LSTM neural network:

[0120]

[0121] Update LSTM neural network parameters W I , R I , B I , W f , R f , B f , W o , R o , B o , W C , R C , B C , W is updated based on the gradient, and the partial derivatives of L with respect to each parameter are:

[0122]

[0123]

[0124]

[0125]

[0126]

[0127]

[0128]

[0129]

[0130]

[0131]

[0132]

[0133]

[0134]

[0135] in, T represents the transpose operation of a matrix or vector, and 1 represents the all-1 matrix or vector of the corresponding matrix or vector operation;

[0136] S3. The spacecraft uses the artificial potential field method to plan a safe orbit as follows:

[0137] ① Determine whether the spacecraft needs to plan a safe orbit:

[0138]

[0139] Where P is a constant and is as follows:

[0140]

[0141] d(t)=||r s (t)-r d (t)|| (61)

[0142] Among them, min(·) means taking the smaller operation, r s (t) represents the position of the spacecraft at time t, d(t) represents the relative distance between the spacecraft and the debris at time t, ||·|| is the binary norm operation, represents the predicted relative distance between the spacecraft and the debris at time t+p, which can be calculated according to formulas (35)-(44);

[0143] Prior to this, the relevant parameters of the LSTM neural network need to be updated according to formulas (45)-(58), which has a self-regulatory effect, real-time correction, and reduces deviations;

[0144] If the planning conditions are met, that is, d min (t)≤P, then execute step ②, otherwise execute step ③;

[0145] The following steps ② and ③ are in parallel relationship;

[0146] ② Calculate the attractive potential function U generated by the given orbit att (t), as follows:

[0147]

[0148] Among them, d s (t)=||r s A(t)-r(t)|| represents the distance between the actual position of the spacecraft at time t and the predetermined orbit, r(t) is the position of the spacecraft at time t. B is the threshold of the attraction transition zone, α is the attraction coefficient, which is specifically expressed as:

[0149]

[0150] Among them, β∈(0,1);

[0151] Calculate the gradient of the attractive potential function as follows:

[0152]

[0153] And, calculate the repulsive force potential function U generated by the fragments rep (t), as follows:

[0154]

[0155] Where, d(t) = ||r s (t)-r d (t)|| represents the distance between the actual position of the spacecraft and the debris at time t, R B is the threshold of the repulsive force transition zone, γ is the repulsive force coefficient;

[0156] Calculate the gradient of the repulsive force potential function as follows:

[0157]

[0158] The planned spacecraft position is r s (t+1), as follows:

[0159]

[0160] Then, execute step ④;

[0161] ③ The spacecraft remains on the predetermined orbit, and the planned spacecraft position is r s (t+1), as follows:

[0162] r s (t+1)=r(t+1) (68)

[0163] Then, execute step ④;

[0164] ④Let t=t+1, and then go to step ①;

[0165] S4, relevant scenario simulation, as follows:

[0166] Assume that the initial orbit of the spacecraft has the following six parameters: semi-major axis a = 8000000m, eccentricity e = 0.007, inclination i = π / 18rad, pericentric angle ω = π / 20rad, ascending node longitude Ω = π / 15rad, true anomaly angle The initial orbital parameters of the fragment are: semi-major axis a = 7944387.81801 m, eccentricity e = 0, inclination i = 0.05279 rad, pericentric angle ω = 0 rad, ascending node longitude Ω = -0.62832 rad, true anomaly angle

[0167] like Figure 3 and Figure 4 As shown, the safe orbit and the relative distance between the spacecraft and the debris obtained by applying the method of the present invention are simple to calculate and can save the computing power of the spacecraft.

[0168] The above is only a preferred embodiment of the present invention. It should be noted that, for those skilled in the art, several improvements and modifications can be made without departing from the principles of the present invention, and these improvements and modifications should also be considered as the protection scope of the present invention. The structures, devices and operating methods not specifically described and explained in the present invention shall be implemented according to the conventional means in the art unless otherwise specified and limited.

Claims

1. A spacecraft safe orbit planning method based on debris position estimation, characterized by: The following steps are involved: S1: Debris location estimation model based on LSTM neural network; In the step S1, an LSTM neural network is also designed for real-time prediction of debris trajectories, and the structure of the LSTM neural network is divided into an input layer, an LSTM layer and an output layer; The input layer, which is used to receive debris trajectory history data, is as follows: x(t)=[r d (t-2p)r d (t-p)r d (t)] T (35) Among them, r d (t) represents the fragment position at time t, r d (tp) represents the fragment position at time tp, r d (t-2p) represents the fragment position at time t-2p, and p represents the prediction interval; The LSTM layer is composed of several LSTM units, each of which includes three gate control structures: input gate I, forget gate f, output gate o and 1 LSTM unit state C, which are calculated as follows: I(t)=sig(W I x(t)+R I h(t-1)+B I ) (36) f(t)=sig(W f x(t)+R f h(t-1)+B f ) (37) o(t)=sig(W o x(t)+R o h(t-1)+B o ) (38) C(t)=tanh(W C x(t)+R C h(t-1)+B C ) (39) Among them, h(t-1) is the output of the LSTM unit at time t-1, W I is the input matrix of the input gate, R I is the circulant matrix of the input gate, B I is the bias vector of the input gate, W f is the input matrix of the forget gate, R f is the circulant matrix of the forget gate, B f is the bias vector of the forget gate, W o is the input matrix of the output gate, R o is the circulant matrix of the output gate, B o is the bias vector of the output gate, W C is the input matrix of the state, R C is the circulant matrix of states, B C is the bias vector of the state, Moreover, sig and tanh are both gating functions, and their expressions are as follows: Where, exp = 2.7183; The output value of the LSTM unit at time t is: h(t)=tanh(c(t))⊙o(t) (42) Among them, ⊙ is the Hadamard product operation, c(t) is the memory vector of the LSTM unit at time t, which can be expressed as: c(t)=f(t)⊙c(t-1)+I(t)⊙C(t) (43) Among them, c(t-1) is the memory vector of the LSTM unit at time t-1. The output layer is a linear combination of the output values ​​of the LSTM units, i.e., the predicted position of the fragment at time t+p y(t)=Wh(t) (44) Where W is the output weight; S2, optimizing the debris position estimation model parameters based on the gradient descent method; The specific steps include: Calculate the loss function L of the LSTM neural network: Update LSTM neural network parameters W I , R I , B I , W f , R f , B f , W o , R o , B o , W C , R C , B C , W is updated based on the gradient, and the partial derivatives of L with respect to each parameter are: in, T represents the transpose operation of a matrix or vector, and 1 represents the all-1 matrix or vector of the corresponding matrix or vector operation; S3, spacecraft use artificial potential field method to plan safe orbit; The specific steps include: M1: Determine whether the spacecraft needs to plan a safe orbit or not; M2: Calculate the attractive potential function U generated by the given orbit att (t), then, executing step M4; M3: The spacecraft remains on the predetermined orbit, and then executes the step M4; M4: let t=t+1, and then go to step M1; The step M1 also includes the following steps: Where P is a constant and is as follows: d(t)=||r s (t)-r d (t)|| (61) Among them, min(·) means taking the smaller operation, r s (t) represents the position of the spacecraft at time t, d(t) represents the relative distance between the spacecraft and the debris at time t, ||·|| is the binary norm operation, represents the predicted relative distance between the spacecraft and the debris at time t+p, calculated according to formulas (35)-(44); Prior to this, the relevant parameters of the LSTM neural network are updated according to formulas (45)-(58); If the planning conditions are met, min (t)≤P, then execute step M2; or if the planning condition d is not met min (t)≤P, execute step M3. The attractive potential function U att (t), as follows: Among them, d s (t)=||r s A(t)-r(t)|| represents the distance between the actual position of the spacecraft at time t and the predetermined orbit, r(t) is the position of the spacecraft at time t. B is the threshold of the attraction transition zone, α is the attraction coefficient, which is specifically expressed as: Among them, β∈(0,1); Calculate the gradient of the attractive potential function as follows: And, calculate the repulsive force potential function U generated by the fragments rep (t), as follows: Where, d(t) = ||r s (t)-r d (t)|| represents the distance between the actual position of the spacecraft and the debris at time t, R B is the threshold of the repulsive force transition zone, γ is the repulsive force coefficient; Calculate the gradient of the repulsive force potential function as follows: The planned spacecraft position is r s (t+1), as follows: Then, execute the step M4. After the spacecraft remains on the predetermined orbit in step M3, the planned spacecraft position is r s (t+1), as follows: r s (t+1)=r(t+1) (68) Then, execute the step M4S4, the relevant scene simulation.

Images