Learning capabilities of neural networks and Keplerian dynamics

Research output: Chapter in Book/Report/Conference proceedingConference contribution

1 Citation (Scopus)

Abstract

Machine learning (ML) tools, especially deep neural networks (DNNs) have garnered significant attention in the last decade; however, it is not clear whether ML tools can learn the inherent characteristics of dynamical model (such as conservation laws) from the training data set. This paper considers the effectiveness of DNNs in learning dynamical system models by considering the Keplerian two-body problem. Training a DNN with data from a single revolution produces poor performance when predicting motion on subsequent revolutions. By incorporating deviations from constancy of angular momentum and total energy into the loss function for the DNN, predictive performance improves significantly. Further improvements appear when a richer training data set (generated from a number of orbits with different in orbital element values) is employed.

Original languageEnglish (US)
Title of host publicationAAS/AIAA Astrodynamics Specialist Conference, 2018
EditorsPuneet Singla, Ryan M. Weisman, Belinda G. Marchand, Brandon A. Jones
PublisherUnivelt Inc.
Pages2293-2310
Number of pages18
ISBN (Print)9780877036579
StatePublished - Jan 1 2018
EventAAS/AIAA Astrodynamics Specialist Conference, 2018 - Snowbird, United States
Duration: Aug 19 2018Aug 23 2018

Publication series

NameAdvances in the Astronautical Sciences
Volume167
ISSN (Print)0065-3438

Conference

ConferenceAAS/AIAA Astrodynamics Specialist Conference, 2018
CountryUnited States
CitySnowbird
Period8/19/188/23/18

Fingerprint

learning
Neural networks
machine learning
education
constancy
angular momentum
Learning systems
two body problem
orbital elements
Angular momentum
conservation laws
dynamical systems
energy
Conservation
Dynamical systems
Orbits
kinetic energy
orbits
deviation
Deep neural networks

All Science Journal Classification (ASJC) codes

  • Aerospace Engineering
  • Space and Planetary Science

Cite this

Guého, D., Singla, P., & Melton, R. G. (2018). Learning capabilities of neural networks and Keplerian dynamics. In P. Singla, R. M. Weisman, B. G. Marchand, & B. A. Jones (Eds.), AAS/AIAA Astrodynamics Specialist Conference, 2018 (pp. 2293-2310). (Advances in the Astronautical Sciences; Vol. 167). Univelt Inc..
Guého, Damien ; Singla, Puneet ; Melton, Robert G. / Learning capabilities of neural networks and Keplerian dynamics. AAS/AIAA Astrodynamics Specialist Conference, 2018. editor / Puneet Singla ; Ryan M. Weisman ; Belinda G. Marchand ; Brandon A. Jones. Univelt Inc., 2018. pp. 2293-2310 (Advances in the Astronautical Sciences).
@inproceedings{b0f69bf56741437ba97f18435312845c,
title = "Learning capabilities of neural networks and Keplerian dynamics",
abstract = "Machine learning (ML) tools, especially deep neural networks (DNNs) have garnered significant attention in the last decade; however, it is not clear whether ML tools can learn the inherent characteristics of dynamical model (such as conservation laws) from the training data set. This paper considers the effectiveness of DNNs in learning dynamical system models by considering the Keplerian two-body problem. Training a DNN with data from a single revolution produces poor performance when predicting motion on subsequent revolutions. By incorporating deviations from constancy of angular momentum and total energy into the loss function for the DNN, predictive performance improves significantly. Further improvements appear when a richer training data set (generated from a number of orbits with different in orbital element values) is employed.",
author = "Damien Gu{\'e}ho and Puneet Singla and Melton, {Robert G.}",
year = "2018",
month = "1",
day = "1",
language = "English (US)",
isbn = "9780877036579",
series = "Advances in the Astronautical Sciences",
publisher = "Univelt Inc.",
pages = "2293--2310",
editor = "Puneet Singla and Weisman, {Ryan M.} and Marchand, {Belinda G.} and Jones, {Brandon A.}",
booktitle = "AAS/AIAA Astrodynamics Specialist Conference, 2018",
address = "United States",

}

Guého, D, Singla, P & Melton, RG 2018, Learning capabilities of neural networks and Keplerian dynamics. in P Singla, RM Weisman, BG Marchand & BA Jones (eds), AAS/AIAA Astrodynamics Specialist Conference, 2018. Advances in the Astronautical Sciences, vol. 167, Univelt Inc., pp. 2293-2310, AAS/AIAA Astrodynamics Specialist Conference, 2018, Snowbird, United States, 8/19/18.

Learning capabilities of neural networks and Keplerian dynamics. / Guého, Damien; Singla, Puneet; Melton, Robert G.

AAS/AIAA Astrodynamics Specialist Conference, 2018. ed. / Puneet Singla; Ryan M. Weisman; Belinda G. Marchand; Brandon A. Jones. Univelt Inc., 2018. p. 2293-2310 (Advances in the Astronautical Sciences; Vol. 167).

Research output: Chapter in Book/Report/Conference proceedingConference contribution

TY - GEN

T1 - Learning capabilities of neural networks and Keplerian dynamics

AU - Guého, Damien

AU - Singla, Puneet

AU - Melton, Robert G.

PY - 2018/1/1

Y1 - 2018/1/1

N2 - Machine learning (ML) tools, especially deep neural networks (DNNs) have garnered significant attention in the last decade; however, it is not clear whether ML tools can learn the inherent characteristics of dynamical model (such as conservation laws) from the training data set. This paper considers the effectiveness of DNNs in learning dynamical system models by considering the Keplerian two-body problem. Training a DNN with data from a single revolution produces poor performance when predicting motion on subsequent revolutions. By incorporating deviations from constancy of angular momentum and total energy into the loss function for the DNN, predictive performance improves significantly. Further improvements appear when a richer training data set (generated from a number of orbits with different in orbital element values) is employed.

AB - Machine learning (ML) tools, especially deep neural networks (DNNs) have garnered significant attention in the last decade; however, it is not clear whether ML tools can learn the inherent characteristics of dynamical model (such as conservation laws) from the training data set. This paper considers the effectiveness of DNNs in learning dynamical system models by considering the Keplerian two-body problem. Training a DNN with data from a single revolution produces poor performance when predicting motion on subsequent revolutions. By incorporating deviations from constancy of angular momentum and total energy into the loss function for the DNN, predictive performance improves significantly. Further improvements appear when a richer training data set (generated from a number of orbits with different in orbital element values) is employed.

UR - http://www.scopus.com/inward/record.url?scp=85069490316&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=85069490316&partnerID=8YFLogxK

M3 - Conference contribution

AN - SCOPUS:85069490316

SN - 9780877036579

T3 - Advances in the Astronautical Sciences

SP - 2293

EP - 2310

BT - AAS/AIAA Astrodynamics Specialist Conference, 2018

A2 - Singla, Puneet

A2 - Weisman, Ryan M.

A2 - Marchand, Belinda G.

A2 - Jones, Brandon A.

PB - Univelt Inc.

ER -

Guého D, Singla P, Melton RG. Learning capabilities of neural networks and Keplerian dynamics. In Singla P, Weisman RM, Marchand BG, Jones BA, editors, AAS/AIAA Astrodynamics Specialist Conference, 2018. Univelt Inc. 2018. p. 2293-2310. (Advances in the Astronautical Sciences).