Ishaan
Goel
Multi-Agent Reinforcement Learning for Markets of Limited Natural Resources
Abstract profile. Full document pending author claim.
Authors:
Ishaan Goel
Date Created:
Not specified
Course Title:
Professor:
Not specified
About Paper:
California's Sustainable Groundwater Management Act (SGMA) was enacted to address groundwater overextraction and aquifer depletion, motivating the development of regulated groundwater markets among basin stakeholders. Building on the stochastic groundwater market framework, this research addresses the computational challenges inherent in solving Generalized Nash Equilibrium Problems (GNEPs), where multiple agents interact within a shared environment characterized by coupled physical and regulatory constraints [1]. Unlike standard Nash equilibria, GNEPs require agents to optimize their individual objectives while operating within feasible strategy sets that are directly influenced by the actions of others, often leading to highly complex and nonconvex solution spaces [2]. To navigate these dynamics, the model employs a Reinforcement Learning (RL) framework built around the Soft Actor-Critic (SAC) algorithm and its multi-agent extensions. By leveraging SAC's off-policy actor-critic architecture and maximum entropy objective, the framework promotes robust exploration and training stability, enabling the discovery of equilibrium strategies even in environments with intricate and interdependent constraints. The implemented system provides a modular and scalable architecture that progresses from single-agent optimization to sophisticated multi-agent coordination, supported by dedicated training pipelines and evaluation scripts. Through the systematic deployment of SAC-based agents in single-agent, two-agent, and larger multi-agent configurations, the framework enables empirical analysis of convergence behavior and constraint satisfaction across diverse resource-sharing scenarios. Compared to traditional equilibrium computation and numerical optimization methods, the proposed approach offers substantially improved scalability and computational efficiency. By providing a practical and high-fidelity computational tool for analyzing groundwater trading dynamics, this work supports policymakers in designing and evaluating sustainable groundwater management policies under SGMA and similar regulatory frameworks.
Source:
Illinois Institute of Technology
Topics:
No topics listed
Co-authors:
Ishaan Goel