Toward low-carbon and risk-aware mobile energy storage systems control: A parallel hierarchical multi-objective deep reinforcement learning method
低炭素でリスクを考慮した移動式エネルギー貯蔵システム制御に向けて: 並列階層的多目的深層強化学習手法 (AI 翻訳)
Xuanang Gui, Qianlong Wang, Yuheng Cheng, Tong Zhao, Jin Zhao
🤖 gxceed AI 要約
日本語
本論文は、移動式バッテリーエネルギー貯蔵システム（MBESS）の運用最適化において、経済性・低炭素・安全性の複数目的を同時に達成するため、深層強化学習（DRL）に基づく並列階層的多目的マルコフ決定過程（PHMO-MDP）フレームワークを提案。GANによるデータ補完とモデル枝刈りにより、現実的な制約下での性能を向上させ、IEEE 30バス系統で有効性を検証した。
English
This paper proposes a Parallel Hierarchical Multi-objective MDP framework based on deep reinforcement learning for optimizing mobile battery energy storage systems (MBESS) operation, balancing economic efficiency, low-carbon operation, and safety. It integrates GAN-based data imputation and model pruning for robustness and efficiency, tested on the IEEE 30-bus system.
Unofficial AI-generated summary based on the public title and abstract. Not an official translation.
📝 gxceed 編集解説 — Why this matters
日本のGX文脈において
日本でも系統用蓄電池の導入が進み、移動式ストレージの活用が検討され始めている。本手法は、複数目的のトレードオフを明示的に扱う点で、日本の配電系統における需給調整や再エネ導入拡大に貢献する可能性がある。
In the global GX context
Globally, mobile battery storage is gaining attention for grid flexibility. This DRL-based multi-objective framework offers a practical approach to balancing economic and decarbonization goals, relevant for power system operators and energy storage asset managers.
👥 読者別の含意
🔬研究者:Provides a novel hierarchical multi-objective DRL method with GAN imputation and pruning for energy storage control.
🏢実務担当者:Offers a framework for optimizing MBESS dispatch to achieve cost savings and emissions reduction simultaneously.
🏛政策担当者:Demonstrates AI-driven control can enhance grid resilience and renewable integration, supporting policy for storage deployment.
📄 Abstract（原文）
Mobile battery energy storage systems (MBESS) utilize temporal and spatial flexibility to enhance power grid resilience, integrate renewables, and provide economic benefits. However, optimizing MBESS operation involves balancing multiple conflicting objectives—namely economic efficiency, low-carbon operation, and operational safety—under significant uncertainties from market prices, traffic conditions, and potential data sparsity. To address these multifaceted challenges, this paper proposes a Parallel Hierarchical Multi-objective Markov Decision Process (PHMO-MDP) framework based on deep reinforcement learning (DRL). This framework employs a hierarchical structure with parameterized scalarization to explicitly manage the trade-offs between low-level operational tasks and high-level strategic goals. Furthermore, a Generative Adversarial Network (GAN)-based data imputation module is integrated to handle missing or sparse input data effectively, enhancing state perception robustness. Lastly, a magnitude-based model pruning strategy is incorporated to improve the computational efficiency of the DRL agent, accelerating convergence and reducing inference latency for real-time applicability. The proposed framework is tested on the IEEE 30-bus system. Results demonstrate that the proposed methods achieve a superior balance across objectives, yielding higher overall performance and stability compared to baseline methods, while validating the effectiveness of the GAN imputation and model pruning components.
🔗 Provenance — このレコードを発見したソース
- openalex https://doi.org/10.1016/j.rineng.2026.111785first seen 2026-07-05 05:03:21
🔔 こうした論文の新着を逃したくない方は キーワードアラート に登録（無料・3キーワードまで）。
gxceed は公開メタデータに基づく研究支援データセットです。要約・翻訳・解説は AI 支援で生成されています。 最終的な解釈・検証は利用者が原典資料に基づいて行うことを前提とします。