Optimizing building energy systems for grid-interactivity, comfort and resilience

With the proliferation of renewable energy sources such as solar photovoltaics, managing the complexity of building energy systems while ensuring grid stability, occupant comfort, and resilience to power outages has become increasingly challenging. To address this challenge, this study proposes a hierarchical control framework that optimally coordinates battery energy storage, heat pumps, and domestic hot water (DHW) systems across multiple residential buildings. Forecasting models for disturbances are developed using linear regression, k-nearest neighbors regression, and LightGBM. At the building level, a data-driven model predictive control (MPC) strategy optimally regulates heat pump operations to ensure occupant comfort, complemented by a rule-based controller for DHW storage scheduling. At the microgrid level, a physics-based MPC dispatches battery energy to achieve grid-level objectives such as peak shaving and emission reduction. Coordination between the two levels is achieved through a bottom-up structure: building-level controllers estimate their future electricity demand, which is passed as a disturbance input to the upper-level battery dispatch optimization. The framework performed effectively in the 2023 NeurIPS CityLearn Challenge, securing second place overall and achieving the best performance in public buildings across comfort, emissions, grid efficiency, and resilience metrics. This work provides an effective solution for community-scale energy management, emphasizing the importance of multi-level coordination between building systems and microgrids to support sustainable and resilient energy operations. Source code are available at: https://github.com/wfzheng/2nd-place-solution-neurips-citylearn2023-control