Paper Abstract and Keywords |
Presentation |
2022-12-16 15:00
Learning of train control measures by means of Deep Q-Network
-- Preliminary study with a single train control -- Shogo Igarashi, Takumi Fukuda, Sei Takahashi, Hideo Nakamura (Nihon Univ), Tetsuya Takata (Kyosan Electric Manufacturing) DC2022-77 |
Abstract |
(in Japanese) |
(See Japanese page) |
(in English) |
Although the predictive fuzzy control technique has been put to practical use as a train control strategy for automatic train operation systems, it is difficult to control trains considering the speed limit and gradient of all running sections due to the complexity of the model. In this paper, we propose a method of learning single train control strategies for automatic train operation systems using Deep Q-Network, which learns control strategies based on the experience of simulation in advance, and confirm that the control strategies obtained by this method provide good control in terms of on-time performance, energy saving, and good ride quality. The control strategy obtained by this method is confirmed to provide good control in terms of punctuality, energy efficiency, and good ride quality. |
Keyword |
(in Japanese) |
(See Japanese page) |
(in English) |
Auto Train Control / Train Control / Machine Learning / Reinforcement Learning / Deep Q-Network / / / |
Reference Info. |
IEICE Tech. Rep., vol. 122, no. 318, DC2022-77, pp. 26-29, Dec. 2022. |
Paper # |
DC2022-77 |
Date of Issue |
2022-12-09 (DC) |
ISSN |
Online edition: ISSN 2432-6380 |
Copyright and reproduction |
All rights are reserved and no part of this publication may be reproduced or transmitted in any form or by any means, electronic or mechanical, including photocopy, recording, or any information storage and retrieval system, without permission in writing from the publisher. Notwithstanding, instructors are permitted to photocopy isolated articles for noncommercial classroom use without fee. (License No.: 10GA0019/12GB0052/13GB0056/17GB0034/18GB0034) |
Download PDF |
DC2022-77 |
|