license: mit
Dataset Details
Dataset Description
The dataset is constructed based on the data dictionary released by Persagy in 2021 ( https://buildingdtp.com ), which defines objects and relationships in building operation systems. This dictionary includes 58 building types, 353 space types, 85 categories of mechanical and electrical systems, 558 equipment types, 95,000 attributes, and 81 relationship types. The dataset primarily focuses on HVAC-related objects and consists of both real-world empirical data and synthetic data generated through a simulation environment.
Simulation Data: The main dataset currently provided is based on 12 real-world project EnergyPlus simulation models. These simulations systematically vary 18 key influencing factors of building systems, including building location (city), building orientation, wall thermal transmittance, operating hours, source-side equipment capacity, distribution equipment capacity, and terminal equipment capacity, among others. This process generates 5,000 simulation scenario cases. We currently provide a subset of 50 simulation cases as an open dataset.
Real Project Data: The dataset includes operational data from 1,045 real-world building projects, covering subsystems such as energy management, security surveillance, facility management, and building automation. Among them, 508 projects contain relatively comprehensive information. Due to privacy concerns, the real-world dataset requires data anonymization , and a subset of the dataset will be open-sourced in the future.
This dataset supports machine learning applications in building operation systems, enabling research in energy efficiency optimization, predictive maintenance, and intelligent control strategies.
Dataset Structure
Data Organization
data/
├── sample_0001/
│ ├── graph/
│ │ ├── ACATACATFC-serve-building.csv
│ │ └── ...
│ └── timeseries/
│ ├── ACATACATAH.parquet
│ └── ...
├── sample_0002/
│ ├── graph/
│ └── timeseries/
└── ...
└── sample_0050/
- Each `sample_XXXX` folder represents an individual sample.
- Each sample contains two subdirectories:
- `graph/`: Contains CSV files representing adjacency matrices.
- File naming convention: `{source_node}-{edge_type}-{target_node}.csv`
- `timeseries/`: Contains time series data in Parquet format.