Collaborative Research: UTILITARIANS: mUlti-agent contexT-aware human-In-the-Loop strategIes for markeT pARticipants In transActive eNergy systemS

About This Grant

This project aims to transform how individuals and communities participate in electricity exchange by introducing an advanced artificial intelligence (AI) framework. As distributed energy resources become more prevalent, consumers increasingly act as “prosumers” who both generate and consume electricity. Furthermore, new entities such as aggregators are emerging to represent groups of prosumers in the energy trades. These evolutions turn traditional distribution systems into dynamic, decentralized networks that support local, peer-to-peer energy trading, which is a core concept of transactive energy. However, modeling these entities in energy trades is challenging due to the nonlinear characteristics of the distribution networks, as well as the uncertainty in the states of the system, driven by environmental conditions and the diverse behavior of market participants. The intellectual merit of this project includes: (1) development of an AI-driven framework that determines the model for aggregators, learns optimal bidding and trading strategies for the participants, and adapts to the evolving system conditions; (2) proposed domain adaptation to efficiently transfer knowledge between problem setups, such as simulated and real-world environments or different conditions and market participant behaviors; (3) introduction of explainable deep reinforcement learning using an innovative “attention map” mechanism to visualize learned knowledge and effectively incorporate human feedback. The framework and the proposed approach provide transparent reasoning behind the decisions that enables human experts to validate and refine the outcomes. Additionally, with the proposed domain adaptation mechanism to transfer knowledge from simulated environments to real-world scenarios, it will support the design efficient and reliable transactive energy architectures. The broader impacts of this project include strengthening graduate and undergraduate curricula in power systems and AI, advancing cross-disciplinary research, and expanding STEM education outreach for K–12 students. This project presents a novel AI-driven framework to enhance the operation of transactive energy systems by integrating advanced machine learning with domain knowledge and human-in-the-loop interpretability. It addresses the complexity of decentralized energy markets, where prosumers engage in peer-to-peer energy trading across dynamic, physically constrained distribution networks. The proposed solution includes four major thrusts: (1) a physics-informed generative graph neural network to create surrogate models that capture nonlinear network characteristics such as power loss; (2) a distributionally robust multi-agent reinforcement learning algorithm that enables agents to learn risk-aware bidding strategies under uncertain and evolving market conditions; (3) a domain adaptation mechanism that allows the trained model to generalize from simulated environments to real-world systems using an adversarial autoencoder; and (4) an interpretable AI framework incorporating attention mechanisms and graph recurrent network to visualize and explain agent decisions, allowing human experts to validate and refine policies. This comprehensive approach promotes robust, efficient, and explainable energy trading strategies, enabling scalable, real-time decision-making in future transactive energy systems. This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.