High-precision orbit prediction method for geosynchronous orbit targets using space-based visible light observation

Abstract

The present invention relates to a high-precision orbit prediction method for geosynchronous orbit targets by space-based visible light observation. First, a suitable long-short-term memory neural network is designed for high-precision orbit prediction of geosynchronous orbit satellite targets; then, target observation data acquired by a space-based visible light observation system, position data of a space-based satellite platform, and precise orbit data of geosynchronous orbit satellite targets are preprocessed, and a training set and a test set required for the long-short-term memory neural network are constructed and trained and tested to obtain a neural network model that approximates an orbital dynamics prediction error model; finally, the obtained neural network model is deployed to correct the orbit prediction based on the orbital dynamics model to obtain a high-precision orbit prediction result for geosynchronous orbit satellite targets. The method of the present invention has high orbit prediction accuracy for geosynchronous orbit satellite targets, can reduce the requirements for complex orbital dynamics modeling, and has strong universality.

Description

Technical Field

[0001] The present invention belongs to a space situation awareness technology based on visible light observation, and relates to a high-precision orbit prediction method for geosynchronous orbit targets by space-based visible light observation. Background Art

[0002] To manage space, perception comes first. Among the many observation methods of space-based space situational awareness, visible light observation has become the main observation method due to its advantages of low energy consumption, high reliability, long range, and easy miniaturization. It is to capture the sunlight reflected by space targets through visible light cameras carried by space-based satellite platforms for measurement and monitoring. Among them, geosynchronous orbit satellites are the most important monitoring targets. In the process of monitoring geosynchronous orbit satellite targets, the orbit prediction of the target is an indispensable key link and the most important basis for intelligence analysis and grasping the weather situation. The traditional orbit prediction method based entirely on the orbital dynamics model is completed by modeling the various perturbations received by the satellite during its operation. This method inevitably has the problem of low modeling accuracy due to the complexity of the satellite space force and the uncertainty of the satellite target parameters, which will directly cause the target satellite orbit prediction accuracy to be low. Therefore, it is necessary to improve the orbit prediction accuracy of space targets, especially geosynchronous orbit satellite targets, which is of great significance to improving my country's space situational awareness capabilities and national space security factors. Summary of the invention

[0003] Technical issues to be solved

[0004] In order to avoid the shortcomings of the prior art, the present invention proposes a high-precision orbit prediction method for geosynchronous orbit targets based on space-based visible light observation, which can achieve high-precision orbit prediction of geosynchronous orbit satellite targets based on space-based visible light observation.

[0005] Technical Solution

[0006] A high-precision orbit prediction method for geosynchronous orbit targets by space-based visible light observation, characterized by the following steps:

[0007] Step 1: Design a long short-term memory neural network for high-precision orbit prediction of geosynchronous orbit satellite targets:

[0008] The LSTM neural network has 7 layers, including 1 input layer, 5 connected hidden layers, and 1 output layer. In the forward calculation, the output h of the hidden layer t for:

[0009] h t =sigmoid(w ox x t +w oh h t-1+b o )·tanh(c t )

[0010] Where sigmoid is the sigmoid function, w ox ,w oh are the weights of the input and hidden layer output to the output gate, x t is the input at time t, h t-1 is the hidden layer output at time t-1, b o is the output bias, tanh is the tangent function, c t is the output of the memory unit at time t;

[0011] The output y of the output layer t for:

[0012] y t =sigmoid(w yh h t )

[0013] Where w yh The weight from the hidden layer output to the output layer output;

[0014] In the back-propagation calculation, the loss function E is defined as:

[0015]

[0016] Where y t is the expected output, To calculate the output;

[0017] Step 2: Preprocess the target observation data obtained by the space-based visible light observation system, the position data of the space-based satellite platform, and the precise orbit data of the geosynchronous orbit satellite target to construct the input training set and test set of the neural network model required by the long short-term memory neural network;

[0018] The long short-term memory neural network designed in step 1 is trained with the training set to obtain a neural network model that approximates the orbital dynamics prediction error model;

[0019] Step 3: Use the relevant data of the past p moments as input to the neural network model that approximates the orbital dynamics prediction error model, and obtain the orbital dynamics prediction error for the future q moments at the output of the neural network model; at the same time, predict the orbit at q moments through the orbital dynamics of the geosynchronous orbit satellite target, and use the prediction error to correct the orbit predicted by the orbital dynamics to obtain a high-precision orbit prediction.

[0020] The input for training the neural network model is: using the relevant data of the geosynchronous orbit satellite target at the past p moments to predict the orbital dynamics prediction error at the next q moments, the total time series length is T, and the input of the neural network model is:

[0021]

[0022] In the formula,

[0023] is the right ascension and declination j×5 seconds before the i-th moment obtained by the space-based visible light observation system, is the position of the space-based satellite platform in the J2000 coordinate system j×5 seconds before the i-th moment; are the right ascension and declination obtained by the space-based visible light observation system from the i-th moment, that is, 6 hours + j × 5 seconds before, is the position of the space-based satellite platform in the J2000 coordinate system 6 hours + j × 5 seconds before the i-th moment.

[0024] The output when training the neural network model is:

[0025]

[0026] In the formula, y k ,k=1,2…,T is the orbital dynamics prediction error at the kth moment, which is obtained by subtracting the precise orbit of the geosynchronous orbit satellite target from the orbit predicted based on the orbital dynamics model.

[0027] The test set obtained in step 2 is used to test the network.

[0028] 98% of the data in step 2 is used as a training set.

[0029] The data of step 2 is obtained by using a space-based visible light observation system to obtain 4,500 sets of data from different geosynchronous orbit target satellites in the past 90 days, of which 4,400 sets of data are used as training sets and 100 sets of data are used as test sets.

[0030] Beneficial Effects

[0031] The present invention proposes a high-precision orbit prediction method for geosynchronous orbit targets by space-based visible light observation. By adopting a long short-term memory neural network and utilizing space-based visible light observation means, high-precision orbit prediction of geosynchronous orbit satellite targets is achieved. First, a suitable long short-term memory neural network is designed for high-precision orbit prediction of geosynchronous orbit satellite targets; then, the target observation data acquired by the space-based visible light observation system, the position data of the space-based satellite platform, and the precise orbit data of the geosynchronous orbit satellite targets are preprocessed to construct the training set and test set required by the long short-term memory neural network, and the training and testing are carried out to obtain a neural network model that approximates the orbital dynamics prediction error model; finally, the obtained neural network model is deployed to correct the orbit prediction based on the orbital dynamics model, and a high-precision orbit prediction result of the geosynchronous orbit satellite target is obtained. The method of the present invention has high orbit prediction accuracy for geosynchronous orbit satellite targets, can reduce the requirements for complex orbital dynamics modeling, and has strong universality. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] Figure 1 This is a flow chart of the high-precision orbit prediction method for space-based visible light observation of geosynchronous orbit satellite targets of the present invention;

[0033] Figure 2 A training phase block diagram of the high-precision orbit prediction method for space-based visible light observation of geosynchronous orbit satellite targets of the present invention;

[0034] Figure 3 The present invention is a block diagram of the deployment phase of the high-precision orbit prediction method for space-based visible light observation of geosynchronous orbit satellite targets. DETAILED DESCRIPTION

[0035] The present invention will now be further described with reference to the embodiments and the accompanying drawings:

[0036] A high-precision orbit prediction method for geosynchronous orbit satellite targets by space-based visible light observation, the method comprising: S1, designing a suitable long short-term memory neural network for high-precision orbit prediction of geosynchronous orbit satellite targets; S2, preprocessing target observation data acquired by a space-based visible light observation system, position data of a space-based satellite platform, and precise orbit data of geosynchronous orbit satellite targets, constructing a training set and a test set required for the long short-term memory neural network, and performing training and testing to obtain a neural network model that approximates an orbit dynamics prediction error model; S3, deploying the obtained neural network model, correcting the orbit dynamics prediction, and obtaining a high-precision orbit prediction result for the geosynchronous orbit satellite target.

[0037] like Figure 1-Figure 3 In combination with the above, the present invention provides a method for high-precision orbit prediction of geosynchronous orbit satellite targets by space-based visible light observation, the method comprising the following steps:

[0038] S1. Design a suitable long short-term memory neural network for high-precision orbit prediction of geosynchronous orbit satellite targets;

[0039] S2. Preprocessing the target observation data obtained by the space-based visible light observation system, the position data of the space-based satellite platform, and the precise orbit data of the geosynchronous orbit satellite target, constructing the training set and test set required for the long short-term memory neural network, and conducting training and testing to obtain a neural network model that approximates the orbital dynamics prediction error model;

[0040] S3. Deploy the obtained neural network model to correct the orbital dynamics prediction and obtain high-precision orbit prediction results for geosynchronous orbit satellite targets.

[0041] The step S1 further comprises:

[0042] The designed LSTM neural network has 7 layers, including 1 input layer, 5 hidden layers, and 1 output layer. In the forward calculation, the output h of the hidden layer is t for:

[0043] h t =sigmoid(w ox x t +w oh h t-1 +b o )·tanh(c t ) (1)

[0044] Where sigmoid is the sigmoid function, w ox ,w oh are the weights of the input and hidden layer output to the output gate, x t is the input at time t, h t-1 is the hidden layer output at time t-1, b o is the output bias, tanh is the tangent function, c t is the output of the memory unit at time t.

[0045] The output y of the output layer t for:

[0046] y t =sigmoid(w yh h t ) (2)

[0047] Where w yh The weight from the hidden layer output to the output layer output.

[0048] In the back-propagation calculation, the loss function E is defined as:

[0049]

[0050] Where y t is the expected output, The calculated output.

[0051] The step S2 further comprises:

[0052] Assuming that the relevant data of the geosynchronous orbit satellite target at the past p moments are used to predict the orbital dynamics prediction error at the next q moments, and the total time series length is T, the input of the neural network model is:

[0053]

[0054] In the formula,

[0055] is the right ascension and declination j×5 seconds before the i-th moment obtained by the space-based visible light observation system, is the position of the space-based satellite platform in the J2000 coordinate system j×5 seconds before the i-th moment; are the right ascension and declination obtained by the space-based visible light observation system from the i-th moment before (6 hours + j × 5 seconds), is the position of the space-based satellite platform in the J2000 coordinate system from the i-th moment before (6 hours + j × 5 seconds).

[0056] The output of the neural network model is:

[0057]

[0058] In the formula, y k ,k=1,2…,T is the orbital dynamics prediction error at the kth moment, which is obtained by subtracting the precise orbit of the geosynchronous orbit satellite target from the orbit predicted based on the orbital dynamics model.

[0059] Through the independently developed space-based visible light observation system, we obtained 4,500 sets of data from different geosynchronous orbit target satellites in the past 90 days, 50 sets per day, and a total time series length of 4,500. Each set of data includes the following three parts:

[0060] (1) Orbital dynamics prediction error at this data point in time

[0061] like Figure 2 As shown, the orbital dynamics prediction error at this data moment is obtained by subtracting the precise orbit obtained by the space-based visible light observation system based on the target observation data from the orbital dynamics prediction orbit.

[0062] (2) Target observation data at a specific time

[0063] Target observation data refers to the target right ascension and declination data obtained by observation and calculation. The target observation data at a specific moment refers to the 36 sets of target right ascension and declination data before the moment of this set of data and the 36 sets of target right ascension and declination data before 6 hours before the moment of this set of data.

[0064] (3) Space-based satellite platform position data at a specific time

[0065] The space-based satellite platform position data at a specific moment refers to the 36 sets of space-based satellite platform position data before the moment of the set of data and the 36 sets of space-based satellite platform position data before 6 hours before the moment of the set of data.

[0066] 4400 sets of data in the dataset are used as training sets, and 100 sets of data are used as test sets.

[0067] The step S3 further comprises:

[0068] like Figure 3 As shown, the neural network model that approximates the orbital dynamics prediction error model obtained after training and testing is deployed, and the relevant data of the past p moments are input to obtain the orbital dynamics prediction error of the future q moments. At the same time, the orbit of the geosynchronous orbit satellite target orbital dynamics is predicted at q moments, and the orbit predicted by the orbital dynamics is corrected using the prediction error to obtain a high-precision orbit prediction.

[0069] In summary, the present invention constructs the required form of the long short-term memory neural network by preprocessing the observation data, the position data of the space-based satellite platform, and the precise orbit data of the geosynchronous orbit satellite target, and obtains the training set and the test set; then the long short-term memory neural network is used for training and testing to obtain a neural network model that approximates the orbital dynamics prediction error model; finally, the obtained neural network model is deployed to correct the orbital dynamics prediction, and a high-precision orbit prediction result for the geosynchronous orbit satellite target is obtained.

[0070] Although the content of the present invention has been described in detail through the above preferred embodiments, it should be appreciated that the above description should not be considered as a limitation of the present invention. After reading the above content, it will be apparent to those skilled in the art that various modifications and substitutions of the present invention will occur. Therefore, the protection scope of the present invention should be limited by the appended claims.

Claims

1. A high-precision orbit prediction method for geosynchronous orbit targets using space-based visible light observation, characterized in that Here are the steps: Step 1: Design a long short-term memory neural network for high-precision orbit prediction of geosynchronous orbit satellite targets: The long short-term memory neural network has 7 layers, including 1 input layer, 5 connected hidden layers, and 1 output layer. In the forward calculation, the output h of the hidden layer is t for: h t =sigmoid(w ox x t +w oh h t-1 +b o )·tanh(c t ) Where sigmoid is the sigmoid function, w ox ,w oh are the weights of the input and hidden layer output to the output gate, x t is the input at time t, h t-1 is the hidden layer output at time t-1, b o is the output bias, tanh is the tangent function, c t is the output of the memory unit at time t; The output y of the output layer t for: y t =sigmoid(w yh h t ) Where w yh The weight from the hidden layer output to the output layer output; In the back-propagation calculation, the loss function E is defined as: Where y t is the expected output, To calculate the output; Step 2: Preprocess the target observation data obtained by the space-based visible light observation system, the position data of the space-based satellite platform, and the precise orbit data of the geosynchronous orbit satellite target to construct the input training set and test set of the neural network model required by the long short-term memory neural network; The long short-term memory neural network designed in step 1 is trained with the training set to obtain a neural network model that approximates the orbital dynamics prediction error model; Step 3: Use the relevant data of the past p moments as input to the neural network model that approximates the orbital dynamics prediction error model, and obtain the orbital dynamics prediction error for the future q moments at the output of the neural network model; at the same time, predict the orbit at q moments through the orbital dynamics of the geosynchronous orbit satellite target, and use the prediction error to correct the orbit predicted by the orbital dynamics to obtain a high-precision orbit prediction.

2. The high-precision orbit prediction method for geosynchronous orbit targets by space-based visible light observation according to claim 1, characterized in that: The input for training the neural network model is: using the relevant data of the geosynchronous orbit satellite target at the past p moments to predict the orbital dynamics prediction error at the next q moments, the total time series length is T, and the input of the neural network model is: In the formula, is the right ascension and declination j×5 seconds before the i-th moment obtained by the space-based visible light observation system, is the position of the space-based satellite platform in the J2000 coordinate system j×5 seconds before the i-th moment; are the right ascension and declination obtained by the space-based visible light observation system from the i-th moment, that is, 6 hours + j × 5 seconds before, is the position of the space-based satellite platform in the J2000 coordinate system 6 hours + j × 5 seconds before the i-th moment.

3. The high-precision orbit prediction method for geosynchronous orbit targets by space-based visible light observation according to claim 1, characterized in that: The output when training the neural network model is: In the formula, y k ,k=1,2…,T is the orbital dynamics prediction error at the kth moment, which is obtained by subtracting the precise orbit of the geosynchronous orbit satellite target from the orbit predicted based on the orbital dynamics model.

4. The high-precision orbit prediction method for geosynchronous orbit targets by space-based visible light observation according to claim 1, characterized in that: The test set obtained in step 2 is used to test the network.

5. The high-precision orbit prediction method for geosynchronous orbit targets by space-based visible light observation according to claim 1, characterized in that: 98% of the data in step 2 is used as a training set.

6. The high-precision orbit prediction method for geosynchronous orbit targets by space-based visible light observation according to claim 1, characterized in that: The data of step 2 is obtained by using a space-based visible light observation system to obtain 4,500 sets of data from different geosynchronous orbit target satellites in the past 90 days, of which 4,400 sets of data are used as training sets and 100 sets of data are used as test sets.

Images