A fast prediction method for the transfer trajectory fuel of cluster spacecraft based on DNN

Abstract

The present application relates to a fast method for transferring trajectory fuel of cluster spacecraft based on DNN. The method includes: setting the constraint conditions of the optimal trajectory optimization problem of relative motion fuel according to boundary condition constraints, relative motion dynamics constraints and control saturation constraints; and setting the objective function of the optimal trajectory optimization problem of relative motion fuel according to the fuel consumed by the spacecraft maneuvering to the observation point; converting the optimal trajectory optimization problem of relative motion fuel into a convex optimization problem, solving the convex optimization problem using the convex optimization toolbox of MATLAB to obtain the fuel consumption index; constructing an initial training set according to multiple orbital parameters and corresponding fuel consumption indicators and then normalizing it, using the normalized training set and test set to train and test the deep neural network, and performing rapid fuel prediction according to the optimized deep neural network. The present method can be used for fuel prediction.

Description

Technical Field

[0001] The present application relates to the field of spacecraft technology, and in particular to a method and device for rapid prediction of fuel for cluster spacecraft transfer trajectories based on DNN. Background Art

[0002] With the rapid development of network technology and the increasing maturity of space missions, cluster spacecraft have attracted extensive attention from scholars due to their flexibility, robustness, and adaptability. Fuel consumption is a key indicator in the task allocation of cluster spacecraft. The computational efficiency of traditional numerical optimization algorithms is usually low, so a method for rapid prediction of fuel indicators is needed. Summary of the invention

[0003] Based on this, it is necessary to provide a DNN-based cluster spacecraft transfer trajectory fuel rapid method and device that can perform fuel prediction in order to address the above technical problems.

[0004] A fast method for transferring trajectory fuel of cluster spacecraft based on DNN, the method comprising:

[0005] Obtain the randomly generated orbital parameters of the cluster spacecraft;

[0006] Construct the initial and terminal boundary conditions of relative motion according to the orbital parameters, and use the initial and terminal boundary conditions of relative motion to construct boundary condition constraints;

[0007] According to the boundary condition constraints, relative motion dynamics constraints and control saturation constraints, set the constraint conditions of the relative motion fuel optimal trajectory optimization problem; and set the objective function of the relative motion fuel optimal trajectory optimization problem based on the fuel consumed by the spacecraft maneuvering to the observation point; construct the relative motion fuel optimal trajectory optimization problem using the constraint conditions and objective function;

[0008] The relative motion fuel optimal trajectory optimization problem is transformed into a convex optimization problem, and the convex optimization toolbox of MATLAB is used to solve the convex optimization problem to obtain the fuel consumption index corresponding to the orbit parameters.

[0009] An initial training set is constructed according to a plurality of orbital parameters and corresponding fuel consumption indicators, and the initial training set is normalized to obtain a normalized training set and a test set;

[0010] Construct a deep neural network; the deep neural network includes an input layer, an output layer, and multiple hidden layers; use the normalized training set and test set to train and test the deep neural network to obtain an optimized deep neural network;

[0011] Fast fuel prediction based on optimized deep neural network.

[0012] In one embodiment, the boundary condition constraints are constructed using the relative motion initial and terminal boundary conditions, including:

[0013] Using the initial and terminal boundary conditions of relative motion, the boundary condition constraints are constructed as x i (0) = x i0 ,x i (M) = x jf , where x i0 represents the initial state of spacecraft i, x jf Represents the relative state of the jth observation point.

[0014] In one embodiment, the relative motion fuel optimal trajectory optimization problem is constructed using constraint conditions and objective functions, including:

[0015] The optimal trajectory optimization problem of relative motion fuel is constructed by using constraints and objective functions to minimize

[0016] Satisfy x i (k+1)=A d x i (k)+B d u i (k)(i=1,...,Ns)

[0017] ||u i (k)|| ∞ ≤T max

[0018] x i (0) = x i0 ,x i (M) = x jf (j=1,...,Nt)

[0019] Among them, T max is the maximum thrust of the propeller, F ij represents the fuel consumed by spacecraft i to the jth observation point, M represents the total number of discrete steps, Nt represents the total number of observation points, and x i (k+1)=A d x i (k)+B d u i (k) is the relative motion dynamics constraint, x i (k) and u i (k) represent the relative motion state and relative control vector, A d and B d Respectively represent the discrete relative state and relative control coefficient, ||u i (k)|| ∞ ≤T maxTo control the saturation constraint.

[0020] In one embodiment, the relative motion fuel optimal trajectory optimization problem is transformed into a convex optimization problem, and the convex optimization toolbox of MATLAB is used to solve the convex optimization problem to obtain the fuel consumption index corresponding to the orbital parameters, including:

[0021] According to the optimization objective and inequality constraints of the relative motion fuel optimal trajectory optimization problem as convex functions, and the equality constraints as affine functions, the relative motion fuel optimal trajectory optimization problem is transformed into a convex optimization problem. The fuel consumption value corresponding to each set of fly-by orbit parameters is calculated using the convex optimization toolbox of MATLAB, and the fuel consumption index corresponding to the fly-by orbit parameters is obtained.

[0022] In one embodiment, the normalization process is

[0023]

[0024] Among them, p is the original data, p max and p min Represent the maximum and minimum values ​​of the original data respectively.

[0025] In one embodiment, the dimensions of the input layer and the output layer are consistent with the dimensions of the input data and the output data; there are multiple hidden layers between the input layer and the output layer; and the hidden layer includes multiple neurons.

[0026] In one embodiment, the deep neural network is trained and tested using the normalized training set and test set to obtain an optimized deep neural network, including:

[0027] According to the optimization algorithm of the neural network, the mean square error between the training samples and the predicted values ​​is used as the optimization target of the deep neural network for training, and the performance of the trained deep neural network is tested using the test set to obtain the optimized deep neural network.

[0028] In one embodiment, performing rapid fuel prediction based on an optimized deep neural network includes:

[0029] The orbital parameters of the cluster spacecraft to be predicted are input into the optimized deep neural network to obtain the fuel prediction results.

[0030] A DNN-based cluster spacecraft transfer trajectory fuel rapid prediction device, the device comprising:

[0031] Construct a constraint condition module to obtain the randomly generated orbital parameters of the cluster spacecraft; construct the initial and terminal boundary conditions of relative motion according to the orbital parameters, and use the initial and terminal boundary conditions of relative motion to construct boundary condition constraints;

[0032] A relative motion fuel optimal trajectory optimization problem module is set up, which is used to set the constraint conditions of the relative motion fuel optimal trajectory optimization problem according to boundary condition constraints, relative motion dynamics constraints and control saturation constraints; and to set the objective function of the relative motion fuel optimal trajectory optimization problem based on the fuel consumed by the spacecraft maneuvering to the observation point; and to construct the relative motion fuel optimal trajectory optimization problem using the constraint conditions and the objective function;

[0033] The problem solving module is used to transform the relative motion fuel optimal trajectory optimization problem into a convex optimization problem, and solve the convex optimization problem using the convex optimization toolbox of MATLAB to obtain the fuel consumption index corresponding to the orbit parameters;

[0034] A training sample construction module is used to construct an initial training set according to a plurality of orbital parameters and corresponding fuel consumption indicators, and to normalize the initial training set to obtain a normalized training set and a test set;

[0035] The neural network training module is used to construct a deep neural network; the deep neural network includes an input layer, an output layer, and multiple hidden layers; the deep neural network is trained and tested using the normalized training set and test set to obtain an optimized deep neural network;

[0036] The fuel prediction module is used to make fast fuel predictions based on the optimized deep neural network.

[0037] The above-mentioned DNN-based cluster spacecraft trajectory fuel rapid transfer method, device, computer equipment and storage medium, the present application constructs boundary condition constraints through the orbital parameters of the cluster spacecraft, and then sets the constraint conditions of the relative motion fuel optimal trajectory optimization problem according to the relative motion dynamics constraints and the control saturation constraints; and sets the objective function of the relative motion fuel optimal trajectory optimization problem based on the fuel consumed by the spacecraft maneuvering to the observation point, by converting the relative motion fuel optimal trajectory optimization problem into a convex optimization problem, using the convex optimization toolbox of MATLAB to solve multiple convex optimization problems, the fuel consumption index corresponding to each set of orbital parameters can be obtained, and finally an initial training set is constructed according to the multiple orbital parameters and the corresponding multiple initial fuel consumption indicators, the initial training set is normalized to obtain a normalized training set and a test set, and the normalized training set is used to train the deep neural network to improve the accuracy of network training, and the fuel is quickly predicted according to the optimized deep neural network, and the deep neural network is used to directly predict the fuel consumption of the relative motion of the spacecraft, thereby replacing the numerical optimization process and improving the computational efficiency of fuel prediction. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] Figure 1A flowchart of a fast method for transferring trajectory fuel of cluster spacecraft based on DNN in one embodiment;

[0039] Figure 2 A schematic diagram of a deep neural network structure in one embodiment;

[0040] Figure 3 A structural block diagram of a DNN-based cluster spacecraft transfer trajectory fuel rapid device in one embodiment. DETAILED DESCRIPTION

[0041] In order to make the purpose, technical solution and advantages of the present application more clearly understood, the present application is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0042] In one embodiment, Figure 1 As shown, a fast method for transferring trajectory fuel of cluster spacecraft based on DNN is provided, including the following steps:

[0043] Step 102, obtaining the randomly generated flyby orbit parameters of the cluster spacecraft; constructing the initial and terminal boundary conditions of the relative motion according to the flyby orbit parameters, and constructing the boundary condition constraints using the initial and terminal boundary conditions of the relative motion.

[0044] The orbital parameters include [b0,c0,φ0,ψ0,b f ,c f ,φ f ,ψ f ], where b is the semi-minor axis of the ellipse projected on the orbital plane, c is the maximum value of the orbit in the z-axis direction, and y is c is the displacement of the flyby center in the y-axis direction, φ and ψ are the phase angles in the orbital plane and out of the orbital plane, respectively. The subscripts 0 and f represent the initial orbit and the terminal orbit, respectively.

[0045] Determine the range of orbital parameters, and construct the initial and terminal boundary conditions of relative motion based on the randomly generated orbital parameters, including:

[0046] Determine the value range of the initial and terminal orbital parameters, and randomly generate a series of orbital parameters [b0, c0, φ0, ψ0, b f ,c f ,φ f ,ψ f ], the initial relative state of the spacecraft Relative status to the terminal This set of characteristic parameters can be expressed

[0047]

[0048] Where: n is the average orbital angular velocity of the target spacecraft, b is the semi-minor axis of the ellipse projected on the orbital plane, c is the maximum value of the orbit in the z-axis direction, y is c is the displacement of the flyby center in the y-axis direction, φ and ψ are the phase angles in and out of the orbital plane, respectively.

[0049] Given the initial orbital parameters [b0, c0, φ0, ψ0] and the terminal orbital parameters [b f ,c f ,φ f ,ψ f ]Get the initial state x of spacecraft i i0 and the expected terminal state x at observation point j jf , thus constructing the initial and terminal boundary conditions x i (0) = x i0 ,x i (M) = x jf , where i = 1, ..., Ns, j = 1, ..., Nt, k = 0, 1, ..., M, Ns and Nt represent the number of satellites and observation points respectively, M is the total number of discrete steps, x i (k) represents the relative state of spacecraft i at the kth step.

[0050] Step 104, according to the boundary condition constraints, relative motion dynamics constraints and control saturation constraints, set the constraint conditions of the relative motion fuel optimal trajectory optimization problem; and set the objective function of the relative motion fuel optimal trajectory optimization problem based on the fuel consumed by the spacecraft maneuvering to the observation point; and construct the relative motion fuel optimal trajectory optimization problem using the constraint conditions and the objective function.

[0051] Step 106, converting the relative motion fuel optimal trajectory optimization problem into a convex optimization problem, solving the convex optimization problem using the convex optimization toolbox of MATLAB, and obtaining the fuel consumption index corresponding to the orbital parameters.

[0052] The relative motion fuel optimal trajectory optimization problem is converted into a convex optimization problem. The convex optimization problem is solved using MATLAB's convex optimization toolbox CVX to obtain the corresponding fuel consumption index. This process is a prior art and will not be described in detail in this application.

[0053] Step 108, constructing an initial training set according to a plurality of orbital parameters and corresponding fuel consumption indicators, and normalizing the initial training set to obtain a normalized training set and a test set.

[0054] Since the orders of magnitude of the parameters vary greatly, the maximum-minimum normalization method is used to convert the original data into the interval [0,1], which can improve the learning accuracy and efficiency of the neural network.

[0055] Step 110, constructing a deep neural network; the deep neural network includes an input layer, an output layer and multiple hidden layers; using the normalized training set and test set to train and test the deep neural network to obtain an optimized deep neural network.

[0056] In this application, the deep neural network uses a fully connected neural network, and the activation function between the hidden layers is the Relu function. The neural network optimization goal is the mean square error between the training sample and the predicted value:

[0057]

[0058] Where: N represents the number of samples, and i Represent the neural network prediction and true value respectively. The optimization algorithm is Adam, and the initial learning rate is 1×10 -4 , the number of mini-batch samples is 200, and the maximum number of iterations is 500.

[0059] The scale of the neural network mainly depends on the number of layers and neurons in the hidden layer. A network that is too large will lead to a complex structure, increased training time, and even overfitting. A network that is too small will make it difficult to fit the mapping relationship between input and output, and the training accuracy cannot be met. In this application, a neural network with 7 hidden layers and 100 neurons in each hidden layer is selected for fuel prediction.

[0060] Step 112, perform rapid fuel prediction based on the optimized deep neural network.

[0061] By inputting the orbital parameters generated by the cluster spacecraft to be predicted into the optimized deep neural network, the fuel prediction results can be obtained.

[0062] In the above-mentioned DNN-based cluster spacecraft transfer trajectory fuel rapid prediction method, the present application constructs boundary condition constraints through the orbital parameters of the cluster spacecraft, and then sets the constraint conditions of the relative motion fuel optimal trajectory optimization problem according to the relative motion dynamics constraints and the control saturation constraints; and sets the objective function of the relative motion fuel optimal trajectory optimization problem based on the fuel consumed by the spacecraft maneuvering to the observation point. By converting the relative motion fuel optimal trajectory optimization problem into a convex optimization problem, and using the convex optimization toolbox of MATLAB to solve multiple convex optimization problems, the fuel consumption index corresponding to each set of orbital parameters can be obtained. Finally, an initial training set is constructed according to the multiple orbital parameters and the corresponding multiple initial fuel consumption indicators, and the initial training set is normalized to obtain a normalized training set and test set. The normalized training set is used to train the deep neural network, which can improve the accuracy of network training, and the fuel is rapidly predicted according to the optimized deep neural network. The deep neural network is used to directly predict the fuel consumption of the relative motion of the spacecraft, thereby replacing the numerical optimization process and improving the computational efficiency of fuel prediction.

[0063] In one embodiment, the boundary condition constraints are constructed using the relative motion initial and terminal boundary conditions, including:

[0064] Using the initial and terminal boundary conditions of relative motion, the boundary condition constraints are constructed as x i (0) = x i0 ,x i (M) = x jf , where x i0 represents the initial state of spacecraft i, x jf Represents the relative state of the jth observation point.

[0065] In one embodiment, the relative motion fuel optimal trajectory optimization problem is constructed using constraint conditions and objective functions, including:

[0066] The optimal trajectory optimization problem of relative motion fuel is constructed by using constraints and objective functions to minimize

[0067] Satisfy x i (k+1)=A d x i (k)+B d u i (k)(i=1,...,Ns)

[0068] ||u i (k)|| ∞ ≤T max

[0069] x i (0) = xi0 ,x i (M) = x jf (j=1,...,Nt)

[0070] Among them, T max is the maximum thrust of the propeller, F ij represents the fuel consumed by spacecraft i to the jth observation point, M represents the total number of discrete steps, Nt represents the total number of observation points, and x i (k+1)=A d x i (k)+B d u i (k) is the relative motion dynamics constraint, x i (k) and u i (k) represent the relative motion state and relative control vector, A d and B d Respectively represent the discrete relative state and relative control coefficient, ||u i (k)|| ∞ ≤T max To control the saturation constraint.

[0071] In one embodiment, the relative motion fuel optimal trajectory optimization problem is transformed into a convex optimization problem, and the convex optimization toolbox of MATLAB is used to solve the convex optimization problem to obtain the fuel consumption index corresponding to the orbital parameters, including:

[0072] According to the optimization objective and inequality constraints of the relative motion fuel optimal trajectory optimization problem as convex functions, and the equality constraints as affine functions, the relative motion fuel optimal trajectory optimization problem is transformed into a convex optimization problem. The fuel consumption value corresponding to each set of fly-by orbit parameters is calculated using the convex optimization toolbox of MATLAB, and the fuel consumption index corresponding to the fly-by orbit parameters is obtained.

[0073] In one embodiment, the normalization process is

[0074]

[0075] Among them, p is the original data, p max and p min Represent the maximum and minimum values ​​of the original data respectively.

[0076] In a specific embodiment, the normalized data is constructed into a training set and a test set in a ratio of 8:2.

[0077] In one embodiment, the dimensions of the input layer and the output layer are consistent with the dimensions of the input data and the output data; there are multiple hidden layers between the input layer and the output layer; and the hidden layer includes multiple neurons.

[0078] In one embodiment, the deep neural network is trained and tested using the normalized training set and test set to obtain an optimized deep neural network, including:

[0079] In a specific embodiment, as shown in Table 1, a relative error comparison of prediction results of networks of different scales is shown.

[0080] Table 1

[0081]

[0082] It can be seen from Table 1 that the prediction error of the neural network obeys the normal distribution, and the mean of the error is close to zero. Comparing the simulations of groups 1 to 4 in Table 1, as the number of hidden layers increases, the variance of the relative error first decreases and then remains basically unchanged. Comparing the simulations of groups 3 and 5 to 8 in Table 2, it can be seen that when the number of hidden layers remains unchanged, as the number of hidden layer neurons increases, the variance of the relative error shows the same change pattern. Taking into account the training accuracy and training efficiency, this application selects the neural network "7×100" for predicting the fuel consumption of the transfer trajectory in subsequent simulations. The 3σ interval of the relative error predicted by the network is [-3.9%, 3.7%,] that is, the probability that the relative error is less than 4% is 99.73%.

[0083] According to the optimization algorithm of the neural network, the mean square error between the training samples and the predicted values ​​is used as the optimization target of the deep neural network for training, and the performance of the trained deep neural network is tested using the test set to obtain the optimized deep neural network.

[0084] In one embodiment, performing rapid fuel prediction based on an optimized deep neural network includes:

[0085] The orbital parameters generated by the cluster spacecraft to be predicted are input into the optimized deep neural network to obtain the fuel prediction results.

[0086] In a specific embodiment, it is assumed that b0=1000, c0=900, ψ0=-π / 6, b f =1000, c f =800,ψ f =π / 4,φ0 and φ f The calculation results of convex optimization and the fuel prediction results of DNN when taking different values ​​are shown in Table 2.

[0087]

[0088] As shown in Table 2, the prediction results of DNN are similar to those of convex optimization, with the maximum relative error being 4.13%. It takes 0.07s for DNN to predict fuel consumption, while 4.84s for convex optimization. The simulation results show that DNN can accurately predict relative motion fuel consumption indicators, and its calculation efficiency is much higher than that of convex optimization calculation method.

[0089] It should be understood that although Figure 1 The steps in the flowchart are shown in sequence as indicated by the arrows, but these steps are not necessarily executed in the order indicated by the arrows. Unless otherwise specified in this document, there is no strict order restriction for the execution of these steps, and these steps can be executed in other orders. Moreover, Figure 1 At least part of the steps may include multiple sub-steps or multiple stages. These sub-steps or stages are not necessarily executed at the same time, but can be executed at different times. The execution order of these sub-steps or stages is not necessarily sequential, but can be executed in turn or alternately with other steps or at least part of the sub-steps or stages of other steps.

[0090] In one embodiment, Figure 3 As shown, a DNN-based cluster spacecraft transfer trajectory fuel rapid device is provided, including: a constraint condition construction module 302, a relative motion fuel optimal trajectory optimization problem setting module 304, a problem solving module 306, a training sample construction module 308, a neural network training module 310 and a fuel prediction module 312, wherein:

[0091] The constraint condition construction module 302 is used to obtain the randomly generated orbital parameters of the cluster spacecraft; construct the initial and terminal boundary conditions of the relative motion according to the orbital parameters, and construct the boundary condition constraints using the initial and terminal boundary conditions of the relative motion;

[0092] The module 304 for setting the relative motion fuel optimal trajectory optimization problem is used to set the constraint conditions of the relative motion fuel optimal trajectory optimization problem according to the boundary condition constraints, the relative motion dynamics constraints and the control saturation constraints; and to set the objective function of the relative motion fuel optimal trajectory optimization problem based on the fuel consumed by the spacecraft maneuvering to the observation point; and to construct the relative motion fuel optimal trajectory optimization problem using the constraint conditions and the objective function;

[0093] The problem solving module 306 is used to transform the relative motion fuel optimal trajectory optimization problem into a convex optimization problem, solve the convex optimization problem using the convex optimization toolbox of MATLAB, and obtain the fuel consumption index corresponding to the orbital parameters;

[0094] The training sample construction module 308 is used to construct an initial training set according to a plurality of orbital parameters and corresponding fuel consumption indicators, and normalize the initial training set to obtain a normalized training set and a test set;

[0095] A neural network training module 310 is used to construct a deep neural network; the deep neural network includes an input layer, an output layer and multiple hidden layers; the deep neural network is trained and tested using a normalized training set and a test set to obtain an optimized deep neural network;

[0096] The fuel prediction module 312 is used to perform rapid fuel prediction based on the optimized deep neural network.

[0097] In one embodiment, the constraint condition building module 302 is further used to build boundary condition constraints using the initial and terminal boundary conditions of the relative motion, including:

[0098] Using the initial and terminal boundary conditions of relative motion, the boundary condition constraints are constructed as x i (0) = x i0 ,x i (M) = x jf (j=1,...,Nt), where x i0 represents the initial state of spacecraft i, x jf Represents the relative state of the jth observation point.

[0099] In one embodiment, the relative motion fuel optimal trajectory optimization problem module 304 is further configured to construct the relative motion fuel optimal trajectory optimization problem using constraint conditions and objective functions, including:

[0100] The optimal trajectory optimization problem of relative motion fuel is constructed by using constraints and objective functions to minimize

[0101] Satisfy x i (k+1)=A d x i (k)+B d u i (k)(i=1,...,Ns)

[0102] ||u i (k)|| ∞ ≤T max

[0103] x i (0) = x i0 ,x i (M) = x jf (j=1,...,Nt)

[0104] Among them, Tmax is the maximum thrust of the propeller, F ij represents the fuel consumed by spacecraft i to the jth observation point, M represents the total number of discrete steps, Nt represents the total number of observation points, and x i (k+1)=A d x i (k)+B d u i (k) is the relative motion dynamics constraint, x i (k) and u i (k) represent the relative motion state and relative control vector, A d and B d Respectively represent the discrete relative state and relative control coefficient, ||u i (k)|| ∞ ≤T max To control the saturation constraint.

[0105] In one embodiment, the problem solving module 306 is further used to transform the relative motion fuel optimal trajectory optimization problem into a convex optimization problem, and solve the convex optimization problem using the convex optimization toolbox of MATLAB to obtain the fuel consumption index corresponding to the orbital parameters, including:

[0106] According to the optimization objective and inequality constraints of the relative motion fuel optimal trajectory optimization problem as convex functions, and the equality constraints as affine functions, the relative motion fuel optimal trajectory optimization problem is transformed into a convex optimization problem. The fuel consumption value corresponding to each set of fly-by orbit parameters is calculated using the convex optimization toolbox of MATLAB, and the fuel consumption index corresponding to the fly-by orbit parameters is obtained.

[0107] In one embodiment, the dimensions of the input layer and the output layer are consistent with the dimensions of the input data and the output data; there are multiple hidden layers between the input layer and the output layer; the hidden layer includes multiple neurons

[0108] In one embodiment, the neural network training module 310 is further used to train and test the deep neural network using the normalized training set and test set to obtain an optimized deep neural network, including:

[0109] According to the optimization algorithm of the neural network, the mean square error between the training samples and the predicted values ​​is used as the optimization target of the deep neural network for training, and the performance of the trained deep neural network is tested using the test set to obtain the optimized deep neural network.

[0110] In one embodiment, the fuel prediction module 312 is further configured to perform rapid fuel prediction based on the optimized deep neural network, including:

[0111] The orbital parameters generated by the cluster spacecraft to be predicted are input into the optimized deep neural network to obtain the fuel prediction results.

[0112] For the specific definition of the DNN-based cluster spacecraft transfer trajectory fuel rapid device, please refer to the definition of the DNN-based cluster spacecraft transfer trajectory fuel rapid method above, which will not be repeated here. Each module in the above-mentioned DNN-based cluster spacecraft transfer trajectory fuel rapid device can be implemented in whole or in part through software, hardware and a combination thereof. The above-mentioned modules can be embedded in or independent of the processor in the computer device in the form of hardware, or can be stored in the memory of the computer device in the form of software, so that the processor can call and execute the operations corresponding to the above modules.

[0113] The technical features of the above embodiments may be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0114] The above embodiments only express several implementation methods of the present application, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the invention patent. It should be pointed out that, for a person of ordinary skill in the art, several variations and improvements can be made without departing from the concept of the present application, and these all belong to the protection scope of the present application. Therefore, the protection scope of the patent of the present application shall be subject to the attached claims.

Claims

1. A DNN-based cluster spacecraft transfer trajectory fuel rapid prediction method, characterized in that: The method comprises: Obtain the randomly generated orbital parameters of the cluster spacecraft; Constructing initial and terminal boundary conditions of relative motion according to the orbital parameters, and constructing boundary condition constraints using the initial and terminal boundary conditions of relative motion; According to the boundary condition constraints, relative motion dynamics constraints and control saturation constraints, setting constraints for the relative motion fuel optimal trajectory optimization problem; and setting an objective function for the relative motion fuel optimal trajectory optimization problem based on the fuel consumed by the spacecraft maneuvering to the observation point; constructing the relative motion fuel optimal trajectory optimization problem using the constraints and objective function; The relative motion fuel optimal trajectory optimization problem is transformed into a convex optimization problem, and the convex optimization toolbox of MATLAB is used to solve the convex optimization problem to obtain the fuel consumption index corresponding to the orbit parameters. constructing an initial training set according to a plurality of orbital parameters and corresponding fuel consumption indicators, and normalizing the initial training set to obtain a normalized training set and a test set; Constructing a deep neural network; the deep neural network includes an input layer, an output layer and multiple hidden layers; using the normalized training set and test set to train and test the deep neural network to obtain an optimized deep neural network; Rapid fuel prediction is performed based on the optimized deep neural network.

2. The method according to claim 1, characterized in that The relative motion initial and terminal boundary conditions are used to construct boundary condition constraints, including: Using the initial and terminal boundary conditions of the relative motion, the boundary condition constraint is constructed as x i (0) = x i0 ,x i (M) = x jf j=1,...,Nt, where x i0 represents the initial state of spacecraft i, x jf Represents the relative state of the jth observation point.

3. The method according to claim 2, characterized in that The constraint conditions and objective function are used to construct the relative motion fuel optimal trajectory optimization problem, including: The optimal trajectory optimization problem of relative motion fuel is constructed by using the constraints and objective function to minimize Satisfy x i (k+1)=A d x i (k)+B d u i (k) i=1, ... ,Ns ||in i (k)|| ∞ ≤T max x i (0)=x i0 ,x i (M)=x jf j=1,...,Nt Among them, T max is the maximum thrust of the propeller, F ij represents the fuel consumed by spacecraft i to the jth observation point, M represents the total number of discrete steps, Nt represents the total number of observation points, and x i (k+1)=A d x i (k)+B d u i (k) is the relative motion dynamics constraint, x i (k) and u i (k) represent the relative motion state and relative control vector, A d and B d Respectively represent the discrete relative state and relative control coefficient, ||u i (k)|| ∞ ≤T max To control the saturation constraint.

4. The method according to any one of claims 1 to 3, characterized in that: The relative motion fuel optimal trajectory optimization problem is transformed into a convex optimization problem. The convex optimization toolbox of MATLAB is used to solve the convex optimization problem and obtain the fuel consumption indicators corresponding to the orbit parameters, including: According to the optimization objective and inequality constraints of the relative motion fuel optimal trajectory optimization problem being convex functions, and the equality constraints being affine functions, the relative motion fuel optimal trajectory optimization problem is transformed into a convex optimization problem, and the fuel consumption value corresponding to each set of fly-by orbit parameters is calculated using the convex optimization toolbox of MATLAB to obtain the fuel consumption index corresponding to the fly-by orbit parameters.

5. The method according to claim 1, characterized in that The normalization process is Among them, p is the original data, p max and p min Represent the maximum and minimum values ​​of the original data respectively.

6. The method according to claim 1, characterized in that The dimensions of the input layer and the output layer are consistent with the dimensions of the input data and the output data; a plurality of hidden layers are included between the input layer and the output layer; and the hidden layer includes a plurality of neurons.

7. The method according to claim 6, characterized in that The normalized training set and test set are used to train and test the deep neural network to obtain an optimized deep neural network, including: According to the optimization algorithm of the neural network, the mean square error between the training samples and the predicted values ​​is used as the optimization target of the deep neural network for training, and the performance of the trained deep neural network is tested using the test set to obtain the optimized deep neural network.

8. The method according to claim 7, characterized in that The fuel rapid prediction is performed according to the optimized deep neural network, including: The orbital parameters generated by the cluster spacecraft to be predicted are input into the optimized deep neural network to obtain the fuel prediction result.

9. A DNN-based cluster spacecraft transfer trajectory fuel rapid prediction device, characterized in that: The device comprises: Constructing a constraint condition module, used to obtain the randomly generated orbital parameters of the cluster spacecraft; constructing the initial and terminal boundary conditions of the relative motion according to the orbital parameters, and constructing the boundary condition constraints using the initial and terminal boundary conditions of the relative motion; A relative motion fuel optimal trajectory optimization problem module is provided, which is used to set the constraint conditions of the relative motion fuel optimal trajectory optimization problem according to the boundary condition constraints, relative motion dynamics constraints and control saturation constraints; and set the objective function of the relative motion fuel optimal trajectory optimization problem based on the fuel consumed by the spacecraft maneuvering to the observation point; and construct the relative motion fuel optimal trajectory optimization problem using the constraint conditions and the objective function; The problem solving module is used to transform the relative motion fuel optimal trajectory optimization problem into a convex optimization problem, and solve the convex optimization problem using the convex optimization toolbox of MATLAB to obtain the fuel consumption index corresponding to the orbit parameters; A training sample construction module is used to construct an initial training set according to a plurality of orbital parameters and corresponding fuel consumption indicators, and to normalize the initial training set to obtain a normalized training set and a test set; A neural network training module is used to construct a deep neural network; the deep neural network includes an input layer, an output layer and multiple hidden layers; the deep neural network is trained and tested using the normalized training set and test set to obtain an optimized deep neural network; The fuel prediction module is used to perform rapid fuel prediction based on the optimized deep neural network.

Images