CityLearn v2: energy-flexible, resilient, occupant-centric, and carbon-aware management of grid-interactive communities

As more distributed energy resources become part of the demand-side\ninfrastructure, it is important to quantify the energy flexibility they provide\non a community scale, particularly to understand the impact of geographic,\nclimatic, and occupant behavioral differences on their effectiveness, as well\nas identify the best control strategies to accelerate their real-world\nadoption. CityLearn provides an environment for benchmarking simple and\nadvanced distributed energy resource control algorithms including rule-based,\nmodel-predictive, and reinforcement learning control. CityLearn v2 presented\nhere extends CityLearn v1 by providing a simulation environment that leverages\nthe End-Use Load Profiles for the U.S. Building Stock dataset to create virtual\ngrid-interactive communities for resilient, multi-agent distributed energy\nresources and objective control with dynamic occupant feedback. This work\ndetails the v2 environment design and provides application examples that\nutilize reinforcement learning to manage battery energy storage system\ncharging/discharging cycles, vehicle-to-grid control, and thermal comfort\nduring heat pump power modulation.\n