Space target orbit state rapid forecasting method based on neural network and collocation point feedback iteration method

Abstract

The invention discloses a space target orbit state rapid forecasting method based on a neural network and a collocation point feedback iteration method, and solves the disadvantages of large orbit determination error and low calculation efficiency of an existing space target orbit determination and forecasting method based on a space optical detection result. And the actual demand of realizing high-speed and high-precision orbit forecasting on a space target on a low-performance space computing platform is difficult to meet. According to the method, the adopted neural network model has the advantages of high approximation precision and wide application range, the calculation efficiency and the calculation precision can be improved by combining the neural network model with the Chebyshev collocation iterative algorithm, and then rapid and accurate judgment of the position and the speed of the space target at the specific moment is achieved; and basic data is provided for rapid determination and analysis of space target approaching and collision threats.

Description

Technical Field

[0001] The present invention relates to a method for quickly predicting the orbital state of a space target, and in particular to a method for quickly predicting the orbital state of a space target based on a neural network and a point-matching feedback iteration method. Background Art

[0002] Space debris and failed spacecraft generated by human space activities pose a huge threat to the safe operation of spacecraft in orbit. In order to maintain the safety of space assets, it is necessary to accurately and quickly predict the orbital status of space debris and failed spacecraft at present and in the future, so as to provide information support for predicting the collision risk of related targets and generating response strategies for space collision events.

[0003] Space-based orbit determination and prediction systems for space targets have attracted widespread attention in recent years due to their wide monitoring range, long observation window, and immunity to atmospheric interference. As a key tool for space target orbit determination and prediction, space optical detection equipment, with its low power consumption, high resolution, and long monitoring range, plays a crucial role in this field. However, existing methods for space target orbit determination and prediction based on space optical detection often suffer from large orbit determination errors and low computational efficiency, making them difficult to meet the practical needs of achieving high-speed and high-precision orbit prediction for space targets on low-performance space computing platforms. Summary of the Invention

[0004] The purpose of the present invention is to solve the technical problem that the existing space target orbit determination and prediction methods based on space optical detection results usually face the disadvantages of large orbit determination errors and low computational efficiency, and are difficult to meet the actual needs of achieving high-speed and high-precision orbit prediction for space targets on low-performance space computing platforms. Instead, a method for rapid prediction of the orbit state of space targets based on a neural network and a point-matching feedback iterative method is provided.

[0005] In order to achieve the above object, the present invention adopts the following technical solutions: A method for rapid prediction of space target orbital state based on a neural network and a point-matching feedback iteration method is characterized in that it includes the following steps: Step 1: Build a neural network model; Step 2: Using the sight angle of the space target in the field of view of the observation star at multiple observation times, the position and velocity of the observation star in the geocentric inertial system at the observation time, and the position and velocity of the space target in the geocentric inertial system at the first observation time, a neural network model is trained to obtain a space target angle measurement and orbit determination model; Step 3: Deploy the space target angle measurement and orbit determination model to the orbit determination computer, input the angle information of the space target to be solved relative to the observation platform at multiple observation times into the space target angle measurement and orbit determination model, and solve the position and velocity of the space target to be solved in the geocentric inertial system at the starting observation time; Step 4: Using the position and velocity of the space target to be solved in the geocentric inertial system at the starting observation time as the initial value, and the length of the time interval between the expected time and the starting observation time as the step length, use the Chebyshev collocation iterative algorithm to calculate the position and velocity of the space target to be solved at a specific time.

[0006] Furthermore, in step 1, in the neural network model, the parameter optimizer is the Adam optimizer, the activation function is set to the ReLU function, and the loss function is set to the sum of the squares of the difference between the output result prediction value and the actual value and the regularization term.

[0007] Furthermore, step 2 is specifically as follows: Step 2.1, set the number of training cycles, initialize the weight matrix, bias matrix, and momentum parameter of the Adam optimizer of each layer of the neural network model, set the fixed parameters of the Adam optimizer, and the regularization coefficient of the regularization term in the loss function; Step 2.2: Set up a training data set. The training data set contains multiple sets of input and output data. In each set of input and output data, the input data includes the line-of-sight angle of the space target in the field of view of the observation star at multiple observation times, and the position and velocity of the observation star in the geocentric inertial system at the observation time. The output data is the position and velocity of the space target in the geocentric inertial system corresponding to the first observation time in the input data. Step 2.3: Use the training data set to train the neural network model established in step 1. Each training session first sends the input data from the input layer to the neural network. After calculations by each layer of the neural network, the predicted position and velocity of the space target in the geocentric inertial system at the first observation time are output; Step 2.4: Input the predicted position and speed from step 2.3 and the actual position and speed of the corresponding output data in the training data set into the loss function to calculate the loss function value; Step 2.5: Calculate the gradient of the loss function with respect to the weights and biases in each layer of the neural network model using the chain rule. Step 2.6: Use the Adam optimizer to update the weights and biases in each layer of the neural network model based on the gradient of the loss function with respect to the weights and biases. Step 2.7: Based on the updated neural network model, return to step 2.3 for cyclic training until the set number of cyclic training times is reached and then exit the training loop to obtain the space target angle measurement and orbit determination model.

[0008] Furthermore, step 4 is specifically as follows: Step 4.1: The time interval between the expected time and the starting observation time is expressed as , t For the time interval At any moment in the time interval After normalization, the interval As the calculation interval, The calculation interval after normalization is t corresponding moments; Step 4.2: Set the Chebyshev collocation point in the normalized calculation interval and calculate the Chebyshev collocation time , m represents the Chebyshev collocation number; Step 4.3, according to Chebyshev collocation time Construct a collocation time vector consisting of Chebyshev collocation values , construct the Chebyshev function matrix based on the collocation time vector C , Chebyshev function differential matrix D , Chebyshev function integral matrix P , M Representation interval The total number of Chebyshev points selected within; Step 4.4: Set the iteration error limit Initialize the position and velocity of the space target to be solved corresponding to all Chebyshev collocation moments in the calculation interval, so that the initial values of the position and velocity of the space target to be solved corresponding to all Chebyshev collocation moments are equal to the position and velocity of the space target to be solved in the geocentric inertial system at the starting observation moment, and obtain the iterative initial values of the position and velocity of the space target to be solved corresponding to all Chebyshev collocation moments ; Step 4.5, iteratively solve the position and velocity of the space target corresponding to all Chebyshev point moments; after each iteration, calculate the two-norm difference between the result of this iteration and the result of the previous iteration ; s Indicates the number of iterations; When the two norm of the difference between the result of this iteration and the result of the previous iteration is Not greater than the termination error limit When , the iteration ends, and the last iteration result is recorded as the spatial target position and velocity corresponding to each Chebyshev point moment. X w ;like Greater than the termination error limit , use the result of this iteration as the starting value for the next iteration and return to step 4.5; The final iteration result obtained by exiting the iteration is , ( ) represents the spatial target position corresponding to each Chebyshev collocation point, ( ) represents the space target velocity corresponding to each Chebyshev collocation point.

[0009] Further, in step 4.1, the The expression is: .

[0010] In step 4.2, the calculation formula of the Chebyshev collocation time is: ; Where, represents the Chebyshev matching moment, m represents the Chebyshev collocation number, M Representation interval The total number of Chebyshev points selected within.

[0011] Furthermore, in step 4.3, the Chebyshev function matrix C , Chebyshev function differential matrix D , Chebyshev function integral matrix P The calculation formulas are: ; ; ; in ; ; ; ; ; ; ; ; ; ; ; ; Where, Represents the collocation time vector The transposed vector of M Representation interval The total number of Chebyshev points selected in Represents the Chebyshev function differential matrix D The diagonal submatrix of Represents the Chebyshev function integral matrix P The diagonal submatrix of , containing superscript k Matrix express No. k column submatrices, l Represents a submatrix The line number of the inner element, with subscript Representation matrix Neidi l Row 1 column element, Represents the collocation time vector The l The value of the element, r is a submatrix 、 and The ordinal number of .

[0012] Furthermore, in step 4.4, the initial values of the spatial target position and velocity to be solved corresponding to all Chebyshev point matching moments are iterated as follows: The expression is: ; in, ; ; ; ; ; ; Where, and express The space target position vector in the geocentric inertial system at the moment x The vector consisting of the components on the axis, and express The space target position vector in the geocentric inertial system at the moment y The vector consisting of the components on the axis, and express The space target position vector in the geocentric inertial system at the moment z The vector consisting of the components on the axis, and express The velocity vector of the space target in the Earth-centered inertial system at the moment x The vector consisting of the components on the axis, and express The velocity vector of the space target in the Earth-centered inertial system at the moment y The vector consisting of the components on the axis, and express The velocity vector of the space target in the Earth-centered inertial system at the moment z A vector with components on the axis.

[0013] Furthermore, in step 4.5, the formula for iteratively solving the position and velocity of the space target to be solved corresponding to all Chebyshev collocation moments is:

[0014] Where, is the length of the calculation interval before normalization; , Indicates the s After iterative correction M The spatial target corresponding to the Chebyshev point matching moment x The position vector of the direction, Indicates the s After iterative correction M The spatial target corresponding to the Chebyshev point matching moment x The velocity vector in the direction of Indicates the s After iterative correction M The spatial target of a Chebyshev collocation moment y The position vector of the direction, Indicates the s After iterative correction M The spatial target corresponding to the Chebyshev point matching moment y The velocity vector in the direction of Indicates the s After iterative correction M The position vector of the space target in the z direction corresponding to the Chebyshev collocation point at that moment, Indicates the s After iterative correction M Space target at a Chebyshev collocation moment z The velocity vector in the direction of Represents the differentials of the target position and velocity calculated from the orbital dynamics equations.

[0015] Beneficial effects of the present invention: The present invention provides a method for rapidly predicting the orbital state of a space target based on a neural network and a collocation feedback iteration method. The neural network model adopted has the advantages of high approximation accuracy and a wide range of applications. When used in combination with the Chebyshev collocation iteration algorithm, it can improve computational efficiency and accuracy, thereby enabling rapid and accurate determination of the position and velocity of a space target at a specific moment, and providing basic data for rapidly measuring and analyzing the approach and collision threats of space targets. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 This is a flow chart of an embodiment of a method for rapid prediction of space target orbital status based on a neural network and a point-matching feedback iteration method provided by the present invention. DETAILED DESCRIPTION

[0017] The following will clearly and completely describe the technical solution of this embodiment in conjunction with the accompanying drawings and embodiments. Obviously, the embodiments described are only part of the embodiments of this embodiment, not all of them. Based on this embodiment, all other embodiments obtained by ordinary technicians in this field without making any creative efforts shall fall within the scope of protection of this embodiment.

[0018] To address the problem that the orbit determination accuracy and computational efficiency of existing space target orbit determination and prediction methods are unable to meet actual engineering needs, this embodiment provides a method for rapidly predicting the orbit state of a space target based on a neural network and a collocation feedback iterative method. This method, based on the angular measurement information of a space optical platform, can efficiently and accurately determine the orbit of a space target and predict its position, thereby meeting the actual needs of actual engineering tasks for rapidly and accurately determining the orbit of a space target and predicting its position. In this embodiment, the neural network model adopts a target state estimation network based on a deep neural network. The sum of the two-norm and the regularization term of the difference between the output prediction value and the true value is used as the neural network loss function. The position and velocity of the observation star in the Earth-centered inertial system and the line-of-sight angle of the space target in the field of view of the observation star are used as the neural network inputs. The position and velocity of the space target in the Earth-centered inertial system at the first observation moment are used as the neural network outputs. In combination with the Chebyshev collocation iterative algorithm, the calculation step length is selected as 2000 seconds, and the number of Chebyshev collocation points in the single-step calculation of the Chebyshev collocation iterative algorithm is selected as 13. The position and velocity of the space target 2000 seconds after the initial observation are rapidly and accurately predicted, including the following steps: Step 1: Figure 1 As shown, a neural network model is constructed; The neural network model used in this embodiment is a target state estimation network based on a deep neural network. First, the scale of the neural network input layer, output layer and hidden layer is set, including the input data dimension, output data dimension and hidden layer dimension. The parameter optimizer is set to the Adam optimizer, the activation function is set to the ReLU function, and the loss function is set to the sum of the squares of the difference between the output result predicted value and the actual value and the regularization term.

[0019] Step 2: Using the sight angle of the space target in the field of view of the observation star at multiple observation times, the position and velocity of the observation star in the Earth-centered inertial system at the observation time, and the position and velocity of the space target in the Earth-centered inertial system at the first observation time, a neural network model is trained to obtain a space target angle measurement and orbit determination model; specifically: Step 2.1. Set the number of training cycles; initialize the weight matrix and bias matrix of each layer of the neural network model, initialize the momentum parameter of the Adam parameter optimizer of the neural network, set the fixed parameters of the Adam parameter optimizer, and set the regularization coefficient of the regularization term in the loss function.

[0020] Step 2.2: Set up a training data set. The training data set contains multiple sets of input and output data. In each set of input and output data, the input data includes the line-of-sight angle of the space target in the field of view of the observation star at multiple observation times, and the position and velocity of the observation star in the geocentric inertial system at the observation time. The output data is the position and velocity of the space target in the geocentric inertial system corresponding to the first observation time in the input data. Step 2.3: Use the training data set to train the neural network model established in step 1. Each training session first sends the input data from the input layer to the neural network. After calculations by each layer of the neural network, the predicted position and velocity of the space target in the geocentric inertial system at the first observation time are output; Step 2.4: Input the predicted position and speed from step 2.3 and the actual position and speed of the corresponding output data in the training data set into the loss function to calculate the loss function value; Step 2.5: Calculate the gradient of the loss function with respect to the weights and biases in each layer of the neural network model using the chain rule. Step 2.6: Use the Adam optimizer to update the weights and biases in each layer of the neural network model based on the gradient of the loss function with respect to the weights and biases. Step 2.7: Based on the updated neural network model, return to step 2.3 for cyclic training until the set number of cyclic training times is reached and then exit the training loop to obtain the space target angle measurement and orbit determination model.

[0021] Step 3: Deploy the space target angle measurement and orbit determination model to the orbit determination computer, input the angle information of the space target to be solved relative to the observation platform at multiple observation times into the space target angle measurement and orbit determination model, and solve the position and velocity of the space target to be solved in the geocentric inertial system at the starting observation time; Step 4: Using the position and velocity of the target in the geocentric inertial system at the time of the initial observation as the initial value, and the length of the time interval between the expected time and the time of the initial observation as the step length, the Chebyshev collocation algorithm is used to calculate the position and velocity of the target at a specific time. Specifically: Step 4.1: The time interval between the expected time and the starting observation time is calculated as a step length of 2000s, expressed as , this embodiment selects , t The time interval is At any moment in the time interval After normalization, the interval As the calculation interval, The calculation interval after normalization is t corresponding moments; The expression is: .

[0022] Step 4.2: Set the Chebyshev collocation point in the normalized calculation interval and calculate the Chebyshev collocation time The calculation formula of Chebyshev collocation time is: ; Where, represents the Chebyshev matching moment, m represents the collocation number of Chebyshev polynomials, M Representation interval The total number of Chebyshev points selected in the example is set M The value is 13.

[0023] Step 4.3, according to Chebyshev collocation time Construct a collocation time vector consisting of Chebyshev collocation values , construct the Chebyshev function matrix based on the collocation time vector C , Chebyshev function differential matrix D , Chebyshev function integral matrix P ; Chebyshev function matrix C , Chebyshev function differential matrix D , Chebyshev function integral matrix P The calculation formulas are: ; ; ; in , , , ; ; ; ; ; ;

[0024] ; ; Where, Represents the collocation time vector The transposed vector of M Representation interval The total number of Chebyshev points selected in Represents the Chebyshev function differential matrix D The diagonal submatrix of Represents the Chebyshev function integral matrix P The diagonal submatrix of , containing superscript k Matrix express No. k column submatrices, l Represents a submatrix The line number of the inner element, with subscript Representation matrix Neidi l Row 1 column element, Represents the collocation time vector The l The value of the element, r is a submatrix 、 and The ordinal number of .

[0025] Step 4.4: Set the iteration error limit Initialize the position and velocity of the space target to be solved corresponding to all Chebyshev point matching moments within the calculation interval, so that the initial values of the position and velocity of the space target to be solved corresponding to all Chebyshev point matching moments are equal to the position and velocity of the space target to be solved in the geocentric inertial system at the starting observation moment, and obtain the iterative initial values of the position and velocity of the space target to be solved corresponding to all Chebyshev point matching moments ; The initial values of the position and velocity iteration of the space target corresponding to all Chebyshev collocation moments The expression is: ; in, ; ; ; ; ; ; Where, and express The space target position vector in the geocentric inertial system at the moment x The vector consisting of the components on the axis, and express The space target position vector in the geocentric inertial system at the moment y The vector consisting of the components on the axis, and express The space target position vector in the geocentric inertial system at the moment z The vector consisting of the components on the axis, and express The velocity vector of the space target in the Earth-centered inertial system at the moment x The vector consisting of the components on the axis, and express The velocity vector of the space target in the Earth-centered inertial system at the moment y The vector consisting of the components on the axis, and express The velocity vector of the space target in the Earth-centered inertial system at the moment z A vector with components on the axis.

[0026] Step 4.5: Iteratively solve the position and velocity of the space target at all Chebyshev collocation points; the formula is: ; Where, is the length of the calculation interval before normalization; , Indicates the s After iterative correction M The spatial target corresponding to the Chebyshev point matching moment x The position vector of the direction, Indicates the s After iterative correction M The spatial target corresponding to the Chebyshev point matching moment x The velocity vector in the direction of Indicates the s After iterative correction M The spatial target corresponding to the Chebyshev point matching moment y The position vector of the direction, Indicates the s After iterative correction M The spatial target corresponding to the Chebyshev point matching moment y The velocity vector in the direction of Indicates the s After iterative correction M The position vector of the space target in the z direction corresponding to the Chebyshev collocation point at that moment, Indicates the s After iterative correction M The spatial target corresponding to the Chebyshev point matching moment z The velocity vector in the direction of Represents the differentials of the target position and velocity calculated from the orbital dynamics equations.

[0027] After each iteration, calculate the two norm of the difference between the result of this iteration and the result of the previous iteration ; s Indicates the number of iterations.

[0028] When the two norm of the difference between the result of this iteration and the result of the previous iteration is Not greater than the termination error limit When , the iteration ends, and the last iteration result is recorded as the spatial target position and speed of each Chebyshev point. X w ;like Greater than the termination error limit , use the result of this iteration as the starting value for the next iteration and repeat step 4.5; If the final iteration result obtained by exiting the iteration is , ( ) represents the spatial target position corresponding to each Chebyshev collocation point, ( ) represents the space target velocity corresponding to each Chebyshev collocation point.

[0029] The data parameters for implementing a method for rapid prediction of space target orbital status based on a neural network and a point-matching feedback iteration method in this embodiment are shown in Table 1.

[0030] Table 1

[0031] According to Table 1, the method of this embodiment predicts the target position and speed 2000s after the initial observation time in one step, and the calculation time is 2.88e. -4 s, with faster on-orbit prediction speed and extremely high computational efficiency. Compared with traditional methods such as the Euler method and Runge-Kutta method based on Taylor expansion, which are limited by the relationship between the Taylor expansion residual error and the computational accuracy, it is impossible to directly calculate the orbital state after 2000s in one step. When the explicit fourth-order and fifth-order Runge-Kutta formula Dormand-Prince pair method widely used in commercial solvers is used, the orbital state after 2000s is obtained after multiple steps of recursion. Under the condition of achieving the same accuracy as the method of this embodiment, the calculation time of the explicit fourth-order and fifth-order Runge-Kutta formula Dormand-Prince pair method will reach 5.43e -4 s, the computational efficiency is only 53% of that of the present method. For those skilled in the art, a computational step size of 2000s can significantly demonstrate the advantages of the large computational step size of the present embodiment method, with extremely high computational efficiency, and has the potential to enable rapid determination and prediction of target orbits on space observation platforms with highly limited computing resources.

[0032] The above description is merely a specific embodiment of the present invention, and a comparison of the effects of the specific embodiment with the relevant comparative examples. However, the scope of protection of the present invention is not limited thereto. Any changes or substitutions within the technical scope disclosed in the present invention shall be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention shall be based on the scope of protection of the claims.

Claims

1. A method for rapid prediction of space target orbital state based on neural network and point-matching feedback iteration method, characterized in that: The following steps are involved: Step 1: Build a neural network model; Step 2: Using the sight angle of the space target in the field of view of the observation star at multiple observation times, the position and velocity of the observation star in the geocentric inertial system at the observation time, and the position and velocity of the space target in the geocentric inertial system at the first observation time, a neural network model is trained to obtain a space target angle measurement and orbit determination model; Step 3: Deploy the space target angle measurement and orbit determination model to the orbit determination computer, input the angle information of the space target to be solved relative to the observation platform at multiple observation times into the space target angle measurement and orbit determination model, and solve the position and velocity of the space target to be solved in the geocentric inertial system at the starting observation time; Step 4: Using the position and velocity of the space target to be solved in the geocentric inertial system at the starting observation time as the initial value, and the length of the time interval between the expected time and the starting observation time as the step length, use the Chebyshev collocation iterative algorithm to calculate the position and velocity of the space target to be solved at a specific time.

2. The method for rapid prediction of space target orbital status based on a neural network and a point-matching feedback iteration method according to claim 1, characterized in that: In step 1, in the neural network model, the parameter optimizer is the Adam optimizer, the activation function is set to the ReLU function, and the loss function is set to the sum of the squares of the difference between the output result prediction value and the actual value and the regularization term.

3. The method for rapid prediction of space target orbital state based on neural network and point-matching feedback iteration method according to claim 2 is characterized in that: Step 2 is as follows: Step 2.1, set the number of training cycles, initialize the weight matrix, bias matrix, and momentum parameter of the Adam optimizer of each layer of the neural network model, set the fixed parameters of the Adam optimizer, and the regularization coefficient of the regularization term in the loss function; Step 2.2: Set a training data set. The training data set includes multiple sets of input and output data. In each set of input and output data, the input data includes the line-of-sight angle of the space target in the field of view of the observation star at multiple observation times, and the position and velocity of the observation star in the geocentric inertial system at the observation time. The output data is the position and velocity of the space target in the geocentric inertial system corresponding to the first observation time in the input data. Step 2.3: Use the training data set to train the neural network model established in step 1. Each training session first sends the input data from the input layer to the neural network. After calculations by each layer of the neural network, the predicted position and velocity of the space target in the geocentric inertial system at the first observation time are output; Step 2.4: Input the predicted position and speed from step 2.3 and the actual position and speed of the corresponding output data in the training data set into the loss function to calculate the loss function value; Step 2.5: Calculate the gradient of the loss function with respect to the weights and biases in each layer of the neural network model using the chain rule. Step 2.6: Use the Adam optimizer to update the weights and biases in each layer of the neural network model based on the gradient of the loss function with respect to the weights and biases. Step 2.7: Based on the updated neural network model, return to step 2.3 for cyclic training until the set number of cyclic training times is reached and then exit the training loop to obtain the space target angle measurement and orbit determination model.

4. The method for rapid prediction of space target orbital state based on neural network and point-matching feedback iteration method according to claim 1, characterized in that: Step 4 is as follows: Step 4.1: The time interval between the expected time and the starting observation time is expressed as , t For the time interval At any moment in the time interval After normalization, the interval As the calculation interval, The calculation interval after normalization is t corresponding moments; Step 4.2: Set the Chebyshev collocation point in the normalized calculation interval and calculate the Chebyshev collocation time , m represents the Chebyshev collocation number; Step 4.3, according to Chebyshev collocation time Construct a collocation time vector consisting of Chebyshev collocation values , construct the Chebyshev function matrix based on the collocation time vector C , Chebyshev function differential matrix D , Chebyshev function integral matrix P , M Representation interval The total number of Chebyshev points selected within; Step 4.4: Set the iteration error limit Initialize the position and velocity of the space target to be solved corresponding to all Chebyshev collocation moments in the calculation interval, so that the initial values of the position and velocity of the space target to be solved corresponding to all Chebyshev collocation moments are equal to the position and velocity of the space target to be solved in the geocentric inertial system at the starting observation moment, and obtain the iterative initial values of the position and velocity of the space target to be solved corresponding to all Chebyshev collocation moments ; Step 4.5, iteratively solve the position and velocity of the space target corresponding to all Chebyshev point moments; after each iteration, calculate the two-norm difference between the result of this iteration and the result of the previous iteration ; s Indicates the number of iterations; When the two norm of the difference between the result of this iteration and the result of the previous iteration is Not greater than the termination error limit When , the iteration ends, and the last iteration result is recorded as the spatial target position and velocity corresponding to each Chebyshev point moment. X w ;like Greater than the termination error limit , use the result of this iteration as the starting value for the next iteration and return to step 4.5; The final iteration result obtained by exiting the iteration is , ( ) represents the spatial target position corresponding to each Chebyshev collocation point, ( ) represents the space target velocity corresponding to each Chebyshev collocation point.

5. The method for rapid prediction of space target orbital state based on neural network and point-matching feedback iteration method according to claim 4 is characterized in that: In step 4.1, The expression is: 。 6. The method for rapid prediction of space target orbital state based on neural network and point-matching feedback iteration method according to claim 4 is characterized in that: In step 4.2, the calculation formula of the Chebyshev collocation time is: ； Where, represents the Chebyshev matching moment, m represents the Chebyshev collocation number, M Representation interval The total number of Chebyshev points selected within.

7. The method for rapid prediction of space target orbital state based on neural network and point-matching feedback iteration method according to claim 4 is characterized in that: In step 4.3, the Chebyshev function matrix C , Chebyshev function differential matrix D , Chebyshev function integral matrix P The calculation formulas are: ； ； ； in, , , ; ； ； ； ； ； ； ； ； ； Where, Represents the collocation time vector The transposed vector of M Representation interval The total number of Chebyshev points selected in Represents the Chebyshev function differential matrix D The diagonal submatrix of Represents the Chebyshev function integral matrix P The diagonal submatrix of , containing superscript k Matrix express No. k column submatrices, l Represents a submatrix The line number of the inner element, with subscript Representation matrix Neidi l Row 1 column element, Represents the collocation time vector The l The value of the element, r is a submatrix 、 and The ordinal number of .

8. The method for rapid prediction of space target orbital status based on neural network and point-matching feedback iteration method according to claim 4 is characterized in that: In step 4.4, the initial values of the spatial target position and velocity to be solved corresponding to all Chebyshev point moments are The expression is: ； in, ; ； ； ； ； ； Where, and express The space target position vector in the geocentric inertial system at the moment x The vector consisting of the components on the axis, and express The space target position vector in the geocentric inertial system at the moment y The vector consisting of the components on the axis, and express The space target position vector in the geocentric inertial system at the moment z The vector consisting of the components on the axis, and express The velocity vector of the space target in the Earth-centered inertial system at the moment x The vector consisting of the components on the axis, and express The velocity vector of the space target in the Earth-centered inertial system at the moment y The vector consisting of the components on the axis, and express The velocity vector of the space target in the Earth-centered inertial system at the moment z A vector with components on the axis.

9. The method for rapid prediction of space target orbital status based on neural network and point-matching feedback iteration method according to claim 4, characterized in that: In step 4.5, the formula for iteratively solving the position and velocity of the space target to be solved corresponding to all Chebyshev collocation points is: ； Where, is the length of the calculation interval before normalization; , Indicates the s After iterative correction M The spatial target corresponding to the Chebyshev point matching moment x The position vector of the direction, Indicates the s After iterative correction M The spatial target corresponding to the Chebyshev point matching moment x The velocity vector in the direction of Indicates the s After iterative correction M The spatial target corresponding to the Chebyshev point matching moment y The position vector of the direction, Indicates the s After iterative correction M The spatial target corresponding to the Chebyshev point matching moment y The velocity vector in the direction of Indicates the s After iterative correction M The spatial target corresponding to the Chebyshev point matching moment z The position vector of the direction, Indicates the s After iterative correction M The spatial target corresponding to the Chebyshev point matching moment z The velocity vector in the direction of Represents the differentials of the target position and velocity calculated from the orbital dynamics equations.

Images