A Four-Year 30-Minute Simulated Residential HEMS Dataset for Six Heterogeneous Homes with PV, Battery, HVAC, and Shiftable Appliances
Description
This dataset provides four years of 30-minute simulated residential energy data for six heterogeneous homes in Sacramento, California, USA. It was developed to support research on home energy management systems (HEMS), residential demand-side flexibility, comfort-aware control, PV-battery scheduling, HVAC operation, and shiftable appliance scheduling. Each home represents a different residential scenario in terms of household type, number of occupants, building size, occupancy behavior, PV capacity, battery capacity, appliance usage, and HVAC characteristics. The released files include complete four-year home-level time series, predefined chronological train/evaluation/test splits, home metadata, summary statistics, a data dictionary, citation information, and license information. The time-series files include base electricity load, real-time electricity price, PV generation, washing machine and dishwasher loads and requests, occupancy, comfort activity, outdoor and indoor temperature, desired temperature, HVAC state, and HVAC electricity consumption. The dataset is simulation-based and does not contain measured private household data or personally identifiable information. External weather and solar resource signals were derived from NSRDB GOES Aggregated PSM v4 data for Sacramento, California, USA. Real-time electricity prices were derived from the CAISO TH_NP15 real-time price series accessed through EnergyOnline. All source signals were aligned to a 30-minute time grid and integrated into the residential simulation workflow. The dataset is intended for benchmarking rule-based, optimization-based, and learning-based residential energy management methods. It can be used to study trade-offs among electricity cost, grid import, peak demand, PV self-consumption, battery cycling, HVAC energy use, thermal comfort, and shiftable appliance service quality. Chronological splits are provided to support reproducible reinforcement learning and machine learning experiments: 2020–2021 for training, 2022 for evaluation, and 2023 for final testing.
Files
Institutions
- Ferdowsi University of MashhadRazavi Khorasan, Mashhad