System and method for predicting trajectory of object

Patent metadata

Field	Value
Publication number	US-12539983-B2
Application number	US-202217901011-A
Country	US
Kind code	B2
Filing date	Sep 1, 2022
Priority date	Sep 3, 2021
Publication date	Feb 3, 2026
Grant date	Feb 3, 2026

Abstract

A method for predicting trajectory of an object includes constructing a training data set using past actual orbital information of a target object, wherein the training data set includes a plurality of pairs of input sequence data corresponding to a trajectory in a first section before a reference point, and output sequence data corresponding to a trajectory in a second section after the reference point, training an object trajectory prediction model using the training data set, and predicting the trajectory of the prediction target object after the reference point, by inputting input sequence data corresponding to an actual trajectory of the prediction target object before the reference point into the object trajectory prediction model.

First claim

The invention claimed is:

1 . A method for predicting trajectory of an object, the method comprising:
constructing, at a data processing unit, a training data set using past actual orbital information of a target object, wherein the training data set includes a plurality of pairs of input sequence data corresponding to a trajectory in a first section before a reference point, and output sequence data corresponding to a trajectory in a second section after the reference point; training, at an artificial neural network unit, an object trajectory prediction model using the training data set; and predicting, at the artificial neural network unit, the trajectory of the prediction target object after the reference point, by inputting input sequence data corresponding to an actual trajectory of the prediction target object before the reference point into the object trajectory prediction model, wherein the constructing the training data set includes: obtaining altitude data of the target object at a specific point in time from the past actual orbital information of the target object; determining an approximation function that expresses an altitude of the target object over time by using curve fitting on the obtained altitude data; generating, using the approximation function, altitude profile sequence data sequentially listing an elapsed time at each of altitudes divided at equal intervals in order from a first altitude to a second altitude, wherein the elapsed time is a time taken for the target object from the first altitude to reach each of the altitudes; and dividing the altitude profile sequence data into input sequence data corresponding to a first section before the reference point, and output sequence data corresponding to a second section after the reference point.

2 . The method according to claim 1 , wherein the training data set is obtained for each type of shape of a predetermined object, and the object trajectory prediction model is trained for each type of shape of the object.

3 . The method according to claim 1 , wherein the object trajectory prediction model is any one of Recurrent Neural Network (RNN), Gated Recurrent Unit (GRU), Long Short Term Memory (LSTM), and Sequence-to-Sequence (Seq2Seq).

4 . The method according to claim 1 , wherein the object is a re-entering object to the Earth.

5 . The method according to claim 4 , wherein the past actual orbital information of the target object is basic orbital information (TLE) data of the re-entering object.

6 . The method according to claim 5 , comprising obtaining altitude data of the re-entering object using a mean motion value included in the basic orbital information data of the re-entering object, or obtaining the altitude data of the re-entering object using the mean motion value in conjunction with at least one of a first derivative value of the mean motion and a B* parameter value.

7 . The method according to claim 6 , wherein the training data set is obtained for each type of shape of a predetermined object, and the object trajectory prediction model is trained for each type of shape of the object.

8 . The method according to claim 7 , wherein the object trajectory prediction model is any one of Recurrent Neural Network (RNN), Gated Recurrent Unit (GRU), Long Short Term Memory (LSTM), and Sequence-to-Sequence (Seq2Seq).

9 . A system for predicting trajectory of an object, the system comprising a processor and instructions executable by the processor to implement:
a data processing unit configured to construct a training data set using past actual orbital information of a target object, wherein the training data set includes a plurality of pairs of input sequence data corresponding to a trajectory in a first section before a reference point, and output sequence data corresponding to a trajectory in a second section after the reference point, and further configured to execute the instructions to: (i) obtain altitude data of the target object at a specific point in time from the past actual orbital information of the target object; (ii) determine an approximation function that expresses an altitude of the target object over time by using curve fitting on the obtained altitude data; (iii) generate, using the approximation function, altitude profile sequence data sequentially listing an elapsed time at each of altitudes divided at equal intervals in order from a first altitude to a second altitude, wherein the elapsed time is a time taken for the target object from the first altitude to reach each of the altitudes; and (iv) divide the altitude profile sequence data into input sequence data corresponding to a first section before the reference point, and output sequence data corresponding to a second section after the reference point; and an artificial neural network unit configured to execute instructions to train an object trajectory prediction model using the training data set constructed by the data processing unit by applying a deep learning algorithm implemented as one of a Recurrent Neural Network (RNN) Gated Recurrent Unit (GRU), Long Short Term Memory (LSTM), or Sequence-to-Sequence (Seq2Seq), and predict the trajectory of the prediction target object after the reference point, by inputting input sequence data corresponding to an actual trajectory of the prediction target object before the reference point into the object trajectory prediction model.

10 . The system according to claim 9 , wherein the training data set is obtained for each type of shape of a predetermined object, and the object trajectory prediction model is trained for each type of shape of the object.

11 . The system according to claim 9 , wherein the object is a re-entering object to the Earth.

12 . The system according to claim 11 , wherein the past actual orbital information of the target object is basic orbital information (TLE) data of the re-entering object.

13 . The system according to claim 12 , obtaining altitude data of the re-entering object using a mean motion value included in the basic orbital information data of the re-entering object, or obtaining the altitude data of the re-entering object using the mean motion value in conjunction with at least one of a first derivative value of the mean motion and a B* parameter value.

14 . The system according to claim 13 , wherein the training data set is obtained for each type of shape of a predetermined object, and the object trajectory prediction model is trained for each type of shape of the object.

Assignees

• Korea Aerospace Res Inst

Inventors

• Jung Okchul
• Seong Jaedong
• Jung Youyeun

Classifications

• B64G1/62

  Systems for re-entry into the earth's atmosphere; Retarding or landing devices · CPC title

• G06N3/08

  Learning methods · CPC title

• B64G1/244 Primary

  Spacecraft control systems · CPC title

• B64G3/00 Primary

  Observing or tracking cosmonautic vehicles · CPC title

• G06N3/0442

  characterised by memory or gating, e.g. long short-term memory [LSTM] or gated recurrent units [GRU] · CPC title