Resilience-Oriented Energy Infrastructure Planning in Oil & Gas Value Chains under Climate-Induced Disruptions: A Stochastic Multi-Agent Reinforcement Learning Framework for Adaptive Grid-Islanding and Microgrid Reconfiguration

The accelerating pace of climate change has rendered traditional energy infrastructure planning models obsolete, particularly in high-risk sectors such as oil, gas, and petrochemicals. This study introduces a novel resilience-oriented framework that leverages Stochastic Multi-Agent Reinforcement Learning (SMARL) to enable autonomous, real-time microgrid reconfiguration and adaptive grid-islanding in response to climate induced disruptions. The model integrates probabilistic climate forecasts (RCP ۸.۵), equipment failure rates, load criticality hierarchies, and distributed energy resource (DER) availability into a unified decision-making architecture. Validated using operational data from an offshore platform cluster in the Persian Gulf and an onshore refinery along the U.S. Gulf Coast, the framework demonstrates a ۴۷% reduction in Mean Time To Recovery (MTTR), ۳۱% decrease in production losses due to blackouts, and ۲۹% lower emergency diesel consumption during extreme weather events. Unlike rule-based or single-agent systems, the proposed SMARL approach enables coordinated, risk-aware responses across multiple microgrid zones, transforming passive redundancy into active, learning-based resilience. This research provides a scalable blueprint for climate adaptive energy management in critical industrial infrastructure.

Keywords:

Climate Resilience , Oil & Gas Infrastructure , Multi-Agent Reinforcement Learning , Microgrid Reconfiguration , Grid Islanding , Stochastic Optimization , Extreme Weather Adaptation