Computer-based systems configured for space object orbital trajectory predictions and methods thereof

Patent metadata

Field	Value
Publication number	US-12046129-B2
Application number	US-202117406917-A
Country	US
Kind code	B2
Filing date	Aug 19, 2021
Priority date	Aug 20, 2020
Publication date	Jul 23, 2024
Grant date	Jul 23, 2024

Abstract

Systems and methods of the present disclosure enable machine learning-based refinement of trajectory predictions using a processor to determine a future trajectory associated with an object state using an iterative physics algorithm. The processor utilizes a trajectory error prediction machine learning model to predict a trajectory error for the future trajectory determined for the object state. The processor determines a pseudo-measurement representative of the trajectory error based at least in part on the trajectory error for the future trajectory and determines a pseudo-measurement noise based at least in part on the pseudo-measurement. The processor determines an updated future trajectory for the future trajectory based on the pseudo-measurement and the pseudo-measurement noise of the trajectory error for the future trajectory, and causes to display a trajectory notification associated with future trajectory on a screen of a user computing device associated with a user.

First claim

What is claimed is:

1. A method, comprising:
receiving, by at least one processor, an object state dataset comprising at least one object state data item representing at least one object state associated with the object;
wherein the at least one object state comprises at least one trajectory of the object in an orbital frame of reference;

determining, by the at least one processor, a future trajectory associated with each object state data item in the object state dataset using an iterative physics-based algorithm;
utilizing, by the at least one processor, a trajectory error prediction machine learning model to output at least one respective trajectory error measurement for a respective trajectory error for each respective future trajectory determined for each respective object state data item in the object state dataset;
wherein the trajectory error prediction machine learning model is trained using a known trajectory error paired with each historical object state data in a historical epoch object state dataset to learn correlations between state and trajectory error based on trajectory parameters;
wherein each historical object state data comprises a historical two-line element (TLE) associated with a particular object;

determining, by the at least one processor, a respective pseudo-measurement associated with the respective trajectory error for each respective future trajectory based at least in part on the respective trajectory error for each respective future trajectory;
wherein the respective pseudo-measurement represents the respective trajectory error as a physical measurement in the orbital frame of reference;

determining, by the at least one processor, a respective pseudo-measurement noise associated with a trajectory error covariance based at least in part on the respective pseudo-measurement representative of the respective trajectory error for each respective future trajectory;
wherein the respective pseudo-measurement noise represents a physical noise measurement of the trajectory error in the orbital frame of reference;

determining, by the at least one processor, a fusion of the respective pseudo-measurement and the respective pseudo-measurement noise into the future trajectory to generate a respective modification to each respective future trajectory for each respective future trajectory based on the respective pseudo-measurement and the respective pseudo-measurement noise of the respective trajectory error for each respective future trajectory; and
causing to display, by the at least one processor, at least one trajectory notification associated with at least one future trajectory on a screen of a user computing device associated with a user.

2. The method as recited in claim 1 , wherein the trajectory error comprises an error term and an uncertainty term.

3. The method as recited in claim 2 , further comprising:
fusing, by the at least one processor, the respective pseudo-measurement and the respective pseudo-measurement noise associated with each respective trajectory error with a respective future trajectory state and its corresponding covariance;
wherein the respective future trajectory state and its corresponding covariance are generated using Kalman filter-related methods; and

determining, by the at least one processor, a respective state and its corresponding uncertainty associated the respective future trajectory.

4. The method as recited in claim 3 , further comprising re-training the trajectory error prediction machine learning model with a new object state dataset based on monitoring the respective uncertainty of each respective updated future trajectory, a respective precision of each respective updated future trajectory, or both.

5. The method as recited in claim 3 , further comprising:
determining, by the at least one processor, a respective pseudo-innovation of each respective future trajectory based at least on each respective trajectory error;
determining, by the at least one processor, a respective pseudo-innovation covariance associated with each respective pseudo-innovation based at least in part on a respective machine learning covariance;
determining, by the at least one processor, a respective gain associated with each respective fusion of each respective trajectory error with each respective future trajectory based at least in part on a respective pseudo-measurement matrix associated with each respective pseudo-innovation covariance; and
determining, by the at least one processor, each respective future trajectory based at least in part on each respective gain.

6. The method as recited in claim 1 , further comprising re-training the trajectory error prediction machine learning model using training data comprising a moving window of testing object state data items in the object state dataset.

7. The method as recited in claim 1 , wherein the trajectory error prediction machine learning model comprises Gaussian Processes.

8. The method as recited in claim 1 , wherein the trajectory error prediction machine learning model comprises at least one model associated with each respective at least one object state data item.

9. The method as recited in claim 1 , wherein the iterative physics-based algorithm comprises extended Kalman filtering.

10. The method as recited in claim 1 , further comprising:
determining, by the at least one processor, a future collision with an object based at least in part on the at least one future trajectory; and
causing to display, by the at least one processor, a collision alert associated with the future collision on the screen of the user computing device associated with the user.

11. A system, comprising:
an object state database to collect an object state dataset comprising at least one object state data item representing at least one object state associated with the object;
wherein the at least one object state comprises at least one trajectory of the object in an orbital frame of reference; and

a Monitored Machine Learning (MML) database comprising at least one processor;
wherein the at least one processor is configured to execute instructions stored in a non-tangible storage medium, the instructions comprising:
receiving the object state dataset from the object state database;
determining a future trajectory associated with each object state data item in the object state dataset using an iterative physics-based algorithm;
utilizing a trajectory error prediction machine learning model to output at least one respective trajectory error measurement for a respective trajectory error for each respective future trajectory determined for each respective object state data item in the object state dataset;
wherein the trajectory error prediction machine learning model is trained a known trajectory error paired with each historical object state data in a historical epoch object state dataset to learn correlations between state and trajectory error based on trajectory parameters;
wherein each historical object state data comprises a historical two-line element (TLE) associated with a particular object;

determining a respective pseudo-measurement associated with the respective trajectory error for each respective future trajectory based at least in part on the respective trajectory error for each respective future trajectory;
wherein the respective pseudo-measurement represents the respective trajectory error as a physical measurement in the orbital frame of reference;

determining a respective pseudo-measurement noise associated with a respective trajectory error covariance based at least in part on the respective pseudo-measurement representative of the respective trajectory error for each respect

Assignees

• Univ Rutgers

Inventors

• Bai Xiaoli
• Peng Hao

Classifications

• G06N3/09

  Supervised learning · CPC title

• G06N20/00

  Machine learning · CPC title

• G01C21/24

  specially adapted for cosmonautical navigation · CPC title

• G08B21/18

  Status alarms (G08B21/02 takes precedence) · CPC title

• G06N7/01

  Probabilistic graphical models, e.g. probabilistic networks · CPC title