Minute-Level Dataset from a Naturally Ventilated Building for Benchmarking and Learning-Based Modeling

Area of Science:

• Building Science
• Data Science
• Energy Efficiency

Background:

• HouseZero® is a naturally ventilated, ultra-low-energy building in Cambridge, MA.
• High-resolution data from 190 sensors is crucial for understanding building performance.
• Raw sensor data often contains errors requiring rigorous filtering.

Purpose of the Study:

• To present a one-year, one-minute interval dataset from HouseZero®.
• To detail a multi-stage filtering methodology for processing large volumes of building sensor data.
• To establish a benchmark dataset for low-energy building research and data-driven modeling.

Main Methods:

• Development of a three-stage filtering process: system error, subsystem error, and sensor-level filters.
• Implementation of an automated algorithm for weekly data processing and storage.
• Utilization of various visualizations for data validation and analysis of feature relationships.

Main Results:

• A comprehensive, one-year dataset with one-minute intervals from HouseZero® was successfully generated.
• The filtering techniques effectively identified and removed erroneous data points from millions of raw sensor readings.
• The validated dataset provides high-fidelity data for advanced building performance analysis.

Conclusions:

• The HouseZero® dataset serves as a valuable benchmark for research on naturally ventilated and ultra-low-energy buildings.
• The developed data processing and filtering methodology can be applied to other building datasets.
• This data supports the advancement of data-driven and learning-based approaches in building thermal modeling.