Resilient Agentic AI-based Supply Chain Management and Disruption Control Framework

다국어 초록 (Multilingual Abstract)

Hybrid electric vehicle (HEV) supply chains are increasingly vulnerable to disruptions caused by international trade and geopolitical instability. These disruptions can affect critical raw materials, battery components, semiconductors, power electronics, and other key inputs required for HEV production. Risks such as tariffs, export restrictions, localization policies, political tensions, transportation delays, and logistics shocks can spread across interconnected supplier-buyer networks, causing shortages, cost increases, and production delays.
This study develops a hybrid multi-agent framework for resilient HEV supply chain management under multiple disruption scenarios. The proposed framework uses news-based information to generate disruption scenarios that differ by duration, affected production stage, and critical input category. These scenarios are then mapped onto a company-specific supply chain network to evaluate their operational impact and generate reconfiguration strategies.
The decision-making layer combines stochastic programming (SP) and deep reinforcement learning (DRL) to support supply chain reconfiguration under uncertainty. The architecture also integrates large and small language models to balance reasoning capability, cost efficiency, and protection of sensitive company data. Experimental results show that the Hybrid SP+DRL strategy achieved the best overall performance, reducing mean shortage by 61.5% and producing the highest composite resilience score. Compared with an LLM-only multi-agent architecture, the proposed framework also reduced execution cost and prevented company-sensitive supply chain information from being exposed to external large language models.

목차 (Table of Contents)

• Chapter I Introduction 1
• 1.1 Background and Research Motivation 1
• 1.2 Problem Statement and Research Gap 2
• 1.3 Purpose, Research Questions, and Objectives 5
• 1.4 Significance and Study Outline 8
• Chapter II Literature Review 11
• 2.1 Vulnerabilities in HEV Supply Chains 11
• 2.1.1 Structural and Critical Input Vulnerabilities 11
• 2.1.2 Trade-Related Disruptions and Multi-Scenario Exposure 13
• 2.2 Resilient Supply Chain 15
• 2.2.1 Resilience Capabilities and Quantitative Decision Support 15
• 2.2.2 Disruption Intelligence, Generative AI, and Multi-Agent Systems 18
• 2.3 Multi-Agent System 20
• 2.3.1 MASs and LLM-Based Agents in Supply Chain Decision-Making 20
• 2.3.2 Limitations of LLM-Only Systems 22
• Chapter III Research Methodology 26
• 3.1 Input Processing Step 29
• 3.2 Scenario Generation Step 30
• 3.3 Solution Modelling Step 32
• 3.3.1 Stochastic Programming Formulation 37
• 3.3.2 Deep Reinforcement Learning Formulation 39
• 3.3.3 Solver Agents 41
• 3.4 Decision Aggregation Step 42
• 3.3.1 Decision-Maker Agent 42
• 3.3.2 Explainer Agent 44
• 3.5 Feedback Incorporation Step 45
• Chapter IV Experimental Setup 47
• 4.1 HEV Supply Chain Data 47
• 4.2 Disruption News Data 49
• 4.3 Models and Hardware 49
• 4.4 Comparison Models 51
• Chapter V Results 53
• 5.1 Proposed Framework's Outputs 53
• 5.2 Comparative Evaluation 57
• 5.3 Managerial Implications and Implementations 61
• Chapter VI Conclusion 64
• [References] 66