EnergyBench Dataset

A open-source large-scale energy meter dataset designed to support a variety of energy analytics applications, including load profile analysis, and load forecasting. It compiles over 60 detailed electricity consumption datasets encompassing approximately 78,000 real buildings representing the global building stock, offering insights into the temporal and spatial variations in energy consumption. The buildings are classified into two categories: Commercial, which includes 2,900 buildings, and Residential, which includes 75,100 buildings, with a total of 65,000 building-days of time-series data. These buildings span diverse geographies, climatic zones, and operational profiles, providing a broad spectrum of usage patterns. Additional to the real buildings, it also includes synthetic and simulated buildings electricity consumption data.

All processed datasets contain hourly data, with some offering additional 15-minute and 30-minute resolution. To ensure efficient storage, we saved the datasets in Parquet format rather than CSV. For datasets involving a large number of buildings, we partitioned the data into multiple Parquet files.

Metadata of each datasets are provided in metadata.

Table 1: Summary of EnergBench dataset (Real Buildings)

Category	# Datasets	# Buildings	# Days	# Hourly Obs
Commercial	20	2,916	16,706	62M
Residential	47	75,121	43,852	1.2B

Table 2: Summary of EnergBench dataset (Synthetic and Simulated Buildings)

Category	# Datasets	# Buildings	# Days	# Hourly Obs
Commercial	1	207,559	365	~1.7B
Residential	4	31M	966	~718.7B