Automated pipeline framework for processing of large-scale building energy time series data

Area of Science:

• Building energy efficiency
• Data science
• Computational science

Background:

• Commercial buildings consume one-third of US electricity, with substantial energy waste.
• Non-intrusive, automated methods are needed for widespread energy audits of buildings.
• Virtual energy audits can identify inefficiencies and savings opportunities without physical inspection.

Purpose of the Study:

• To demonstrate virtual energy audits on large-scale building datasets.
• To develop and apply an automated Building Energy Analytics (BEA) pipeline.
• To analyze HVAC operational hours and identify energy-saving targets in commercial buildings.

Main Methods:

• A systematic approach using a fully automated BEA pipeline.
• Utilizing a non-relational data warehouse for efficient data unification, cleaning, storage, and analysis.
• Employing a custom compute job scheduler for parallel processing on a high-performance computing cluster.
• Implementing a data qualification tool with hierarchical clustering for error correction and abnormality detection.

Main Results:

• The BEA pipeline processed 816 buildings' data in 34 minutes, 85 times faster than sequential processing.
• Data quality was improved, and building operations were analyzed efficiently.
• Food sales buildings were identified as prime targets for HVAC energy savings, with 17.75 hours of daily cooling operation.

Conclusions:

• The automated BEA pipeline enables statistically significant, population-based studies of building energy data at scale.
• Virtual energy audits can effectively identify energy-saving opportunities in large building populations.
• This data-driven approach facilitates a new generation of building energy analysis and efficiency improvements.