Virtualized twin control framework for optimal ammonia combustion

Magoshi, Ryotaro; Sawada, Shinya; Konishi, Kokoro; Wakashiro, Fugo; Nakatsuka, Noriaki; Hori, Tsukasa; Akamatsu, Fumiteru

doi:https://doi.org/10.1016/j.ijhydene.2025.152109

Virtualized twin control framework for optimal ammonia combustion

Magoshi, Ryotaro; Sawada, Shinya; Konishi, Kokoro et al.

International Journal of Hydrogen Energy, 2025, 189, 152109

アクセス数:1,065件（2026-02-15 22:59 集計）

固定URL: https://hdl.handle.net/11094/103336

閲覧可能ファイル

ファイル	フォーマット	利用条件	サイズ	閲覧回数	利用開始日	説明	information
IntJHydrogenEnergy_189	pdf	なし	4.74 MB		2027.11.17

Creative Commons : 表示 - 非営利 - 改変禁止

論文情報

ファイル出力 EndNote Basic出力 Mendeley出力

タイトル	Virtualized twin control framework for optimal ammonia combustion
著者	著者名 Magoshi, Ryotaro
	NRID 1000070969683 1000070969683 ORCID 0000-0002-2264-4910 0000-0002-2264-4910 researchmap permalink sawadashinya sawadashinya Scopus Author ID 57210562192 57210562192 著者名 Sawada, Shinya
	著者名 Konishi, Kokoro
	著者名 Wakashiro, Fugo
	NRID 1000040634457 1000040634457 著者名 Nakatsuka, Noriaki
	NRID 1000040744066 1000040744066 ORCID 0000-0001-7545-0055 0000-0001-7545-0055 researchmap permalink 7000017271 7000017271 Scopus Author ID 55017652500 55017652500 著者名 Hori, Tsukasa
	NRID 1000010231812 1000010231812 researchmap permalink read0013928 read0013928 Scopus Author ID 56071092200 56071092200 著者名 Akamatsu, Fumiteru
キーワード等	Ammonia combustion optimization
	Data-efficient modeling and control
	Engineering sensitivity analysis
	Hierarchical deep reinforcement learning (HDRL)
	Machine learning
	Mathematical optimization
抄録	The utilization of ammonia as a fuel in industrial furnaces has attracted significant attention in the pursuit of a carbon-neutral society. However, compared with fossil fuels, ammonia produces higher NOx emissions and exhibits lower radiative heat transfer from the flame, posing challenges for heating efficiency. Moreover, industrial applications must simultaneously consider safety, product quality, cost, and environmental regulations, which complicates the realization of high-precision real-time optimal control. To address these challenges, this study develops and validates a Hierarchical Deep Reinforcement Learning (HDRL) control architecture based on Proximal Policy Optimization (PPO). The framework separates a prediction-oriented agent for forecasting process variations and an optimization-oriented agent for determining control actions, thereby enhancing both control performance and interpretability. The predictive agent is further enriched with engineering sensitivity information extracted from the facility and supported by a rationally designed data acquisition strategy, including quasi-steady state data extraction. These additions enable accurate and reliable model construction from limited operational data, facilitating rapid controller development even during the commissioning phase. A 100-kW-class ammonia combustion furnace was employed to demonstrate the framework. A data-driven multi-objective optimization control strategy was implemented to concurrently minimize NOx emissions and maximize thermal efficiency under constraints such as furnace temperature, temperature uniformity, and unburned ammonia emissions. Demonstration experiments confirmed that the proposed controller can simultaneously achieve multiple conflicting objectives, validating its practical applicability. This study establishes a practical and comprehensive data-driven framework that integrates sensitivity information with rational feature design, enabling self-adaptive, flexible, and real-time multi-objective optimization. The proposed methodology provides both a guideline and a pathway toward the sustainable implementation of data-driven industrial process control systems.
出版者	Elsevier Ltd
収録物名	International Journal of Hydrogen Energy
巻(号)	189
ページ	152109
刊行年月	2025-11-17
言語	英語
ハンドルURL	https://hdl.handle.net/11094/103336
PISSN	03603199
関連情報 (isVersionOf)	DOI (出版社版) https://doi.org/10.1016/j.ijhydene.2025.152109
アクセス権	embargoed access
権利情報	©2025. This manuscript version is made available under the CC-BY-NC-ND 4.0 license https://creativecommons.org/licenses/by-nc-nd/4.0/
権利情報	The full-text file will be made open to the public on 17 November 2027 in accordance with the publisher's policy.
出版タイプ	Accepted Manuscript
カテゴリ	学術雑誌論文 Journal Article

資源タイプ	学術雑誌論文
ローカル資源タイプ	学術雑誌論文
DCMI資源タイプ	text
DCTERMS.bibliographicCitation	International Journal of Hydrogen Energy. 2025, 189, p. 152109
DC.title	Virtualized twin control framework for optimal ammonia combustion
DC.creator	Magoshi, Ryotaro
	Sawada, Shinya
	Konishi, Kokoro
	Wakashiro, Fugo
	Nakatsuka, Noriaki
	Hori, Tsukasa
	Akamatsu, Fumiteru
DC.publisher	Elsevier Ltd
DC.language' scheme='DCTERMS.RFC1766	英語
DC.type' scheme='DCTERMS.DCMIType	text
DCTERMS.issued' scheme='DCTERMS.W3CDTF	2025-11-17
DC.identifier	https://doi.org/10.1016/j.ijhydene.2025.152109
DC.subject	Ammonia combustion optimization
	Data-efficient modeling and control
	Engineering sensitivity analysis
	Hierarchical deep reinforcement learning (HDRL)
	Machine learning
	Mathematical optimization
DCTERMS.abstract	The utilization of ammonia as a fuel in industrial furnaces has attracted significant attention in the pursuit of a carbon-neutral society. However, compared with fossil fuels, ammonia produces higher NOx emissions and exhibits lower radiative heat transfer from the flame, posing challenges for heating efficiency. Moreover, industrial applications must simultaneously consider safety, product quality, cost, and environmental regulations, which complicates the realization of high-precision real-time optimal control. To address these challenges, this study develops and validates a Hierarchical Deep Reinforcement Learning (HDRL) control architecture based on Proximal Policy Optimization (PPO). The framework separates a prediction-oriented agent for forecasting process variations and an optimization-oriented agent for determining control actions, thereby enhancing both control performance and interpretability. The predictive agent is further enriched with engineering sensitivity information extracted from the facility and supported by a rationally designed data acquisition strategy, including quasi-steady state data extraction. These additions enable accurate and reliable model construction from limited operational data, facilitating rapid controller development even during the commissioning phase. A 100-kW-class ammonia combustion furnace was employed to demonstrate the framework. A data-driven multi-objective optimization control strategy was implemented to concurrently minimize NOx emissions and maximize thermal efficiency under constraints such as furnace temperature, temperature uniformity, and unburned ammonia emissions. Demonstration experiments confirmed that the proposed controller can simultaneously achieve multiple conflicting objectives, validating its practical applicability. This study establishes a practical and comprehensive data-driven framework that integrates sensitivity information with rational feature design, enabling self-adaptive, flexible, and real-time multi-objective optimization. The proposed methodology provides both a guideline and a pathway toward the sustainable implementation of data-driven industrial process control systems.
DC.rights	©2025. This manuscript version is made available under the CC-BY-NC-ND 4.0 license https://creativecommons.org/licenses/by-nc-nd/4.0/
DC.rights	The full-text file will be made open to the public on 17 November 2027 in accordance with the publisher's policy.
DC.format	103336
citation_title	Virtualized twin control framework for optimal ammonia combustion
citation_author	Magoshi, Ryotaro
	Sawada, Shinya
	Konishi, Kokoro
	Wakashiro, Fugo
	Nakatsuka, Noriaki
	Hori, Tsukasa
	Akamatsu, Fumiteru
citation_publisher	Elsevier Ltd
citation_language	英語
citation_date	2025-11-17
citation_journal_title	International Journal of Hydrogen Energy
citation_volume	189
citation_firstpage	152109
citation_doi	https://doi.org/10.1016/j.ijhydene.2025.152109
citation_public_url	https://hdl.handle.net/11094/103336

Virtualized twin control framework for optimal ammonia combustion

閲覧可能ファイル

論文情報

論文詳細を表示