Learning Assisted Agent-based Energy Optimization: A Reinforcement Learning Based Consensus + Innovations Approach

Ever-increasing penetration of distributed energy resources and adoption of advanced sensing and control technologies are fueling the transition of our centralized electric grid to a more distributed and multi-entity (agent) infrastructure. Distributed information processing methods provide scalability and robustness and lend themselves well to addressing the needs of our evolving grid. The performance of distributed decision-making methods, however, highly depends on the algorithmic design and parameter selection. This paper focuses on parameter selection and proposes using Machine Learning to speed up the convergence of distributed decision-making. To this end, we build on our prior works on the fully distributed Consensus + Innovations approach and adopt a Reinforcement Learning-based method to boost the convergence process.