The HouseZero® two‑year dataset (June 2022 – May 2024) provides a comprehensive, high-resolution collection of environmental and operational data from HouseZero®, an ultra-low-energy living-laboratory building located in Cambridge, Massachusetts. The dataset is recorded at hourly intervals and includes:
Outdoor sensor data: such as temperatures, microclimate variables, and façade surface conditions (e.g., exterior temperatures)
Indoor environmental measurements: including indoor air temperature, relative humidity, and CO₂ concentration
Systems and operations logs: covering the status and performance of integrated building systems—such as operable windows/skylights, heat pump and geothermal systems, thermally activated building systems (TABS), zone-level BTU meters, valve positions, and slab temperatures.
This dataset is invaluable for research in data-driven management of low-energy buildings, enabling model development for natural ventilation control, geothermal/TABS dynamics, environmental comfort prediction, energy system optimization, and benchmarking of ultra-efficient building operation.
Use case:
Data-Driven Management and Optimization of Ultra-Low-Energy Buildings
| Example Observations (≈ 60 features / time step) | External conditions: temperature, humidity, solar irradiance, wind speed. Internal states per zone (≈ 15 zones): temperature, RH, CO₂, luminosity, binary occupancy or counting. System signals: current positions of hot/cold water valves, motorized window openings, fan speed, heat pump power, external blinds state. Timestamp: hour, day of the week, season. |
| Example Actions (controls available in the dataset) | Zone temperature set-point [18 – 26 °C] Airflow set-point (%) / fan speed (0–100 %) Motorized window opening (0–100 %) External blinds position (0–100 %) Heating/cooling valve set-points (0–100 %) |
| Example Reward | rₜ = − (Energyₜ / E_max) − λ⋅ComfortPenaltyₜ where Energyₜ = electrical + thermal consumption (kWh) extracted from meters, ComfortPenaltyₜ = Σ_z max(0, |T_zₜ − T_set| − δ) (δ ≈ 1 °C), λ governs the comfort/energy trade-off. |
| Example Transitions | Sampling frequency: 1 min (internal loggers). Duration: 24 months ⇒ ≈ 1,050,000 transitions (1 min) or 210,000 (5 min if aggregated). Data already cleaned, timestamped, and synchronized in 22 CSV files. |