A Deep Q-Network based optimized modulation scheme for Dual-Active-Bridge converter to reduce the RMS current

Yuanhong Tang; Weihao Hu; Jian Xiao; Zhengdong Lu; Zhou Liu; Zhe Chen; Frede Blaabjerg

doi:10.1016/j.egyr.2020.11.055

A Deep Q-Network based optimized modulation scheme for Dual-Active-Bridge converter to reduce the RMS current

Yuanhong Tang, Weihao Hu, Jian Xiao, Zhengdong Lu, Zhou Liu, Zhe Chen, Frede Blaabjerg

Research output: Contribution to journal › Conference article in Journal › peer-review

2 Downloads (Pure)

Abstract

In order to reduce the conduction losses of the Dual-Active-Bridge (DAB) converter, this paper proposes an optimized modulation scheme based on deep reinforcement learning (DRL). Owing to the Extended-Phase-Shift (EPS) modulation based Deep Q-Network (DQN) algorithm, the optimal phase-shift-angles can be defined, which reduces the root-mean-square (RMS) current tremendously. Moreover, the zero-voltage-switching (ZVS) performance can be guaranteed for the whole operation conditions. A 200 W prototype of the DAB converter is built and tested to prove the effectiveness of the proposed optimized modulation scheme. Experimental results demonstrates that the proposed optimized modulation scheme can obtain lower RMS current and higher operation efficiency in comparison to other three modulations.

Original language	English
Journal	Energy Reports
Volume	6
Issue number	9
Pages (from-to)	1192-1198
Number of pages	7
DOIs	https://doi.org/10.1016/j.egyr.2020.11.055
Publication status	Published - Dec 2020
Event	7th International Conference on Power and Energy Systems Engineering (CPESE 2020) - Fukuoka, Japan Duration: 26 Sep 2020 → 29 Sep 2020

Conference

Conference	7th International Conference on Power and Energy Systems Engineering (CPESE 2020)
Country	Japan
City	Fukuoka
Period	26/09/2020 → 29/09/2020

Keywords

DAB converter
RMS current
deep reinforcement learning
deep Q-network
ZVS

Access to Document

10.1016/j.egyr.2020.11.055Licence: CC BY-NC-ND 4.0

1-s2.0-S2352484720314803-mainFinal published version, 1.29 MBLicence: CC BY-NC-ND 4.0

Cite this

@inproceedings{1af2e75a0dd74d92a0f37a425e247258,

title = "A Deep Q-Network based optimized modulation scheme for Dual-Active-Bridge converter to reduce the RMS current",

abstract = "In order to reduce the conduction losses of the Dual-Active-Bridge (DAB) converter, this paper proposes an optimized modulation scheme based on deep reinforcement learning (DRL). Owing to the Extended-Phase-Shift (EPS) modulation based Deep Q-Network (DQN) algorithm, the optimal phase-shift-angles can be defined, which reduces the root-mean-square (RMS) current tremendously. Moreover, the zero-voltage-switching (ZVS) performance can be guaranteed for the whole operation conditions. A 200 W prototype of the DAB converter is built and tested to prove the effectiveness of the proposed optimized modulation scheme. Experimental results demonstrates that the proposed optimized modulation scheme can obtain lower RMS current and higher operation efficiency in comparison to other three modulations.",

keywords = "DAB converter, RMS current, deep reinforcement learning, deep Q-network, ZVS",

author = "Yuanhong Tang and Weihao Hu and Jian Xiao and Zhengdong Lu and Zhou Liu and Zhe Chen and Frede Blaabjerg",

year = "2020",

month = dec,

doi = "10.1016/j.egyr.2020.11.055",

language = "English",

volume = "6",

pages = "1192--1198",

journal = "Energy Reports",

issn = "2352-4847",

publisher = "Elsevier",

number = "9",

note = "7th International Conference on Power and Energy Systems Engineering (CPESE 2020), CPESE 2020 ; Conference date: 26-09-2020 Through 29-09-2020",

}

TY - GEN

T1 - A Deep Q-Network based optimized modulation scheme for Dual-Active-Bridge converter to reduce the RMS current

AU - Tang, Yuanhong

AU - Hu, Weihao

AU - Xiao, Jian

AU - Lu, Zhengdong

AU - Liu, Zhou

AU - Chen, Zhe

AU - Blaabjerg, Frede

PY - 2020/12

Y1 - 2020/12

N2 - In order to reduce the conduction losses of the Dual-Active-Bridge (DAB) converter, this paper proposes an optimized modulation scheme based on deep reinforcement learning (DRL). Owing to the Extended-Phase-Shift (EPS) modulation based Deep Q-Network (DQN) algorithm, the optimal phase-shift-angles can be defined, which reduces the root-mean-square (RMS) current tremendously. Moreover, the zero-voltage-switching (ZVS) performance can be guaranteed for the whole operation conditions. A 200 W prototype of the DAB converter is built and tested to prove the effectiveness of the proposed optimized modulation scheme. Experimental results demonstrates that the proposed optimized modulation scheme can obtain lower RMS current and higher operation efficiency in comparison to other three modulations.

AB - In order to reduce the conduction losses of the Dual-Active-Bridge (DAB) converter, this paper proposes an optimized modulation scheme based on deep reinforcement learning (DRL). Owing to the Extended-Phase-Shift (EPS) modulation based Deep Q-Network (DQN) algorithm, the optimal phase-shift-angles can be defined, which reduces the root-mean-square (RMS) current tremendously. Moreover, the zero-voltage-switching (ZVS) performance can be guaranteed for the whole operation conditions. A 200 W prototype of the DAB converter is built and tested to prove the effectiveness of the proposed optimized modulation scheme. Experimental results demonstrates that the proposed optimized modulation scheme can obtain lower RMS current and higher operation efficiency in comparison to other three modulations.

KW - DAB converter

KW - RMS current

KW - deep reinforcement learning

KW - deep Q-network

KW - ZVS

U2 - 10.1016/j.egyr.2020.11.055

DO - 10.1016/j.egyr.2020.11.055

M3 - Conference article in Journal

VL - 6

SP - 1192

EP - 1198

JO - Energy Reports

JF - Energy Reports

SN - 2352-4847

IS - 9

T2 - 7th International Conference on Power and Energy Systems Engineering (CPESE 2020)

Y2 - 26 September 2020 through 29 September 2020

ER -

A Deep Q-Network based optimized modulation scheme for Dual-Active-Bridge converter to reduce the RMS current

Abstract

Conference

Keywords

Access to Document

Fingerprint

Cite this