Transactive multi-agent reinforcement learning for distributed energy price localization

Abstract

Energy demand response is projected to play a crucial role in facilitating increased renewable penetration. Current energy pricing mechanisms are simple and not optimized to the way consumers respond financially; adaptive and dynamic pricing may be pivotal in maximizing for deferability of demand. I have demonstrated the efficacy of a price-setting reinforcement learning controller in an office building simulation above baseline controls in shifting demand and lowering the building's energy costs. Abstracting to a microgrid level, I have demonstrated that a similar architecture is also effective. As a next step, I am interested in a larger scale simulation of a distributed controller network. With the appropriate simulation setup, quantifying the effects of numerous localized price-setters would be possible.