RADR-MO：グラフアテンションネットワークと多目的メタ強化学習による強靭性を考慮した動的再ルーティング
RADR-MO: Resilience-Aware Dynamic Re-routing via Graph Attention Networks and Multi-Objective Meta-Reinforcement Learning for Supply Chain Resilience (原題)
Qi Chen
🤖 gxceed AI 要約
日本語
本論文は、サプライチェーンの混乱（港湾閉鎖、異常気象、供給者障害）に対応するため、コスト・強靭性・炭素排出を同時に最適化する動的再ルーティング手法RADR-MOを提案する。グラフアテンションネットワークとメタ強化学習を統合し、未知の混乱シナリオにも迅速に適応する。実データでの評価では、強靭性スコア0.943、コスト33.7%削減、再計画遅延5.1倍高速化を達成。
English
This paper proposes RADR-MO, a dynamic re-routing framework for supply chain resilience that jointly optimizes cost, resilience, and carbon emissions using graph attention networks and meta-reinforcement learning. It adapts to novel disruptions with few gradient steps, achieving a resilience score of 0.943, 33.7% cost reduction, and 5.1x faster re-planning on real-world logistics data.
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
日本企業にとって、サプライチェーン強靭化は経済安全保障上の重要課題であり、物流ネットワークの炭素排出削減はGX投資の対象でもある。本手法は、SSBJ開示やScope 3排出量削減目標の達成に寄与する可能性があり、物流・製造業の実務に示唆を与える。
In the global GX context
Globally, this work aligns with the growing emphasis on climate-resilient supply chains under frameworks like TCFD and CSRD, where physical climate risks and transition risks (including carbon emissions) must be disclosed. The multi-objective optimization approach offers a practical method for companies to balance cost, resilience, and decarbonization, supporting transition finance and ESG ratings.
👥 読者別の含意
🔬研究者:Provides a novel integration of GAT and meta-RL for multi-objective supply chain optimization, offering a benchmark for future research on AI-driven climate-resilient logistics.
🏢実務担当者:Offers a concrete tool for logistics and supply chain managers to reduce costs and emissions while enhancing resilience, potentially supporting Scope 3 reporting and customer disclosure requests.
🏛政策担当者:Highlights the potential of AI to support supply chain resilience and decarbonization, informing policies that encourage adoption of such technologies for climate adaptation.
📄 Abstract（原文）
Abstract–Supply chain resilience—the capacity to maintain core operations under real-world disruptions such as port blockages, extreme weather, and supplier failures—has become a critical concern in global logistics management. Existing dynamic routing methods suffer from three key limitations: (1) they rely on static topology representations that cannot capture real-time IoT-driven disruption signals, (2) they optimize a single objective (cost or delivery rate) while ignoring the inherent trade-offs among cost, resilience, and carbon emission, and (3) they require full retraining when confronted with previously unseen disruption scenarios, resulting in prohibitive latency. To address these gaps, we propose RADR-MO, a Risk-Aware Dynamic Re-routing framework with Multi-Objective optimization that integrates three innovations: (1) a Graph Attention Network encoder with IoT-driven edge weight updating for real-time vulnerability estimation, (2) a Pareto-guided vector reward formulation jointly optimizing cost, resilience score, and carbon emission, and (3) a Model-Agnostic Meta-Learning policy that adapts to novel disruption tasks within a few gradient steps. Extensive experiments on the Solomon benchmark and the real-world SCG logistics dataset across five disruption types demonstrate that RADR-MO achieves a resilience score of 0.943, reduces routing cost by up to \(33.7\%\), and is 5.1 × faster in re-planning latency compared to the strongest baseline, while maintaining robust performance at network scales up to 1,600 nodes.
🔗 Provenance — このレコードを発見したソース
- semanticscholar https://doi.org/10.1145/3821966.3821992first seen 2026-09-05 05:30:35 · last seen 2026-09-21 04:54:06
🔔 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。