A Hybrid Benchmark Dataset for Multi-Home Residential Energy Management under Heterogeneous Building and Energy Conditions
Description
This dataset provides a hybrid benchmark testbed for multi-home residential energy management and intelligent control research. It was developed to support the evaluation of scalable optimization and deep reinforcement learning-based controllers under heterogeneous residential conditions. The dataset contains 100 residential homes with four years of synchronized half-hourly data from January 2020 to December 2023. Each home is provided as an individual CSV file (home_001.csv–home_100.csv), resulting in 7,012,800 home-level records and 70,128 regional time steps. The dataset combines two complementary components: (1) real regional signals, including weather and solar-resource data from the National Solar Radiation Database (NSRDB) and real-time electricity price signals derived from CAISO market data, and (2) empirically grounded household scenarios generated using published residential statistics, distributed energy resource references, and physics-based thermal models. Each household represents heterogeneous residential conditions, including occupancy patterns, building characteristics, HVAC behavior, photovoltaic generation, battery storage, base-load demand, and shiftable appliance operation. Indoor thermal dynamics are generated using reduced-order RC models with home-specific parameters to provide physically consistent comfort modeling. The dataset is organized into chronological training, validation, and testing subsets without random shuffling, enabling evaluation under future unseen conditions. Additional files include household metadata and real-time price forecasting data used for predictive energy management studies. This dataset is intended for research on multi-home energy management, constrained reinforcement learning, demand response, HVAC control, battery scheduling, photovoltaic integration, and scalable smart-grid optimization. The complete dataset documentation, including data sources, construction methodology, variable descriptions, and reference basis, is provided with this release.