Watch a how-to demonstration video: End-Use Load Profiles Dataset Access Demonstration
Documentation of the End-Use Load Profiles dataset can be found in the technical report, End-Use Load Profiles for the U.S. Building Stock: Methodology and Results of Model Calibration, Validation, and Uncertainty Quantification.
The residential (ResStock) dataset represents dwelling units in the contiguous United States, including single-family, multi-family (including high rise multi-family), and mobile homes. It does not include dormitories, prisons, assisted care facilities, and other congregate housing situations.
The commercial (ComStock) dataset represents 14 of the most common commercial building types – small office, medium office, large office, retail, strip mall, warehouse, primary school, secondary school, full-service restaurant, quick-service restaurant, small hotel, large hotel, hospital, and outpatient – which comprise about 65% of the commercial sector floor area in the United States according to CBECS.
The building stock represents, as closely as possible, the U.S. building stock as it was in 2018. The building stock representation is the same for the two weather years of end use load profile results (2018 and TMY3).
It depends on your application. Many applications just need an aggregate load shape. If analyzing scenarios that require realistic spikiness of individual dwelling unit or building loads, such as behind-the-meter solar plus storage, rate design involving real-time pricing or demand charges, or distribution system impacts, then we recommend using individual building profiles. These recommendations will be discussed in more detail in a forthcoming Applications and Opportunities report.
Descriptions of each of the building characteristics filters and the end-use categories can be found in the data_dictionary.tsv file (example). Descriptions of the values used in those filters can be found in the enumeration_dictionary.tsv (example). Both files can be opened with Excel or a text editor.
Water system energy consumption includes all building energy related to to residential water heating and commercial service water heating and pumping.
The list of included end-use categories can be found under the fields labeled "out.electricity", "out.natural_gas", etc. in the data_dictionary.tsv file (example).
In each of the published datasets there is a spatial lookup table: spatial_tract_lookup_table.csv (example). County and PUMA codes can be looked up using the "nhgis_county_gisjoin" and "nhgis_puma_gisjoin" columns, respectively. If you want to find the pre-aggregated timeseries file for a county or PUMA, you can use this lookup to find the code for the county of interest. To find the 5-digit PUMA code based on a city or place name, use this file from the U.S. Census Bureau: 2010_PUMA_Names.txt.
Aggregating by California climate zone is available for residential building profiles but not commercial building profiles.
To achieve this aggregation for the residential load profiles, use the resstock.nrel.gov website and select the dataset and region of interest (Example: the ResStock National Load Profiles by State 2018 dataset and the state of California.) At the bottom click the “Explore Timeseries” button. At the left side, halfway down, click the button “+ Add Filters”. In the “Filters” column, find and select the “Cec climate zone” filter. The options available are for CEC Climate Zones 1-16. Do not use the “None” option, as the option is for locations outside the state of California.
There are two ways to access the building characteristics data associated with an aggregate load profile:
The filename of the individual building (or dwelling unit) load profile's parquet files contains the building ID. Each of these building IDs corresponds to a row in the dataset's metadata, which is available in either .parquet or .tsv format (tab-separated value format that can be opened in Excel) (ResStock example) (ComStock example).
All downloaded energy data is in kWh, including all electricity, natural gas, propane, and fuel oil end uses, as documented in the data_dictionary.tsv files (example).
Timeseries energy consumption data viewed on the website are in metric units. The metric prefix is on the y-axis label (T for tera, G for giga, M for mega, etc.) and the rest of the unit information is in the y-axis label.
The timestamps of all load profiles have been converted to Eastern Standard Time, to prevent issues when aggregating across time zones.
The underlying modeling was done using local Standard Time for each location. In converting from local Standard Time to Eastern Standard Time, if necessary the last few hours of each dataset were moved to the beginning of the timeseries. For example, the first two hours of data from Colorado in Eastern Standard Time (Jan 1, midnight to 2 AM) were originally modeled as the last two hours of the year in Mountain Standard Time (Dec 31, 10 PM to midnight) using the corresponding weather.
Yes, the aggregates represent the total "floor_area_represented" for commercial or "units_represented" for residential.
Parquet files can be read using programming languages such as Python, using the pyarrow package. For other options, see https://arrow.apache.org/docs/index.html. There are a few third-party graphical tools for viewing parquet files, but we have not tested them and the third-party support is limited.
There are no plans for an API. However, we are currently developing documentation that will explain how to link one’s own Amazon Web Services account to this data, so the data can be queried by analytic tools like Athena. We will also be providing example SQL queries to help facilitate analyses.
Yes, there are solar PV profiles in the ResStock data but not the ComStock data.
No, we do not currently model EV charging in the dataset. For modeling aggregate EV load profiles for a city or state, we suggest using EVI-Pro Lite. Measured charging profile data for individual homes can be found in the NEEA HEMS data and Pecan Street Dataport. Email us at firstname.lastname@example.org if you have suggestions for other EV charging data sources.
Users should estimate standard error for metrics of interest using the standard deviation divided by the square root of the number of samples (i.e., profiles or models). As discussed in the methodology report (section 5.1.3), for residential units, a good rule of thumb is to use at least 1000 samples to maintain 15% or less sampling uncertainty for common quantities of interest. Queries in sparsely populated areas or with filters applied may have relatively few samples available. In these cases, samples from nearby locations can be grouped to increase the sample size.
As an example, if one is interested in the mean change in annual electricity costs in a certain county under a potential new rate structure and 500 samples are available in that county, the costs should be calculated for all 500 samples and the standard deviation of those costs can be used to calculate the standard error of the mean change in annual electricity costs.
We query several terabytes (TB) of data in real time to produce the time series graphs you see on the webpage. Running a query for California, Texas, New York, or Illinois takes around a minute, while running a query for a state like Colorado or Massachusetts takes about 10-20 seconds. However - if the graphs have previously been generated we have the data cached and can typically load the data in a few seconds. That's why sometimes loading new graphs is faster than other times.
The “Explore Timeseries” option is available once a specific geography (state or PUMA region) is selected.
Clicking on the end uses in the legend will toggle their inclusion in the visualization.
In the “Explore Timeseries” mode, use the Month Constraints sliders to select the month of the day you are interested in. Then choose “Export csv” and “15 minute resolution”. The resulting csv file will have 15 minute end use load profiles that are not aggregated over time.
The timestamp indicates the end of each 15-minute interval. So "12:15" represents the energy use between 12:00 and 12:15.
The 'sum' aggregation is the total energy consumption for all buildings that meet the filter criteria across all the occurrences of the given time step within the selected month(s). For example, in a day timeseries range for the month of July, the 7-7:15 AM hour time step shows the sum of all energy consumption between 7-7:15 AM in July, from buildings that meet the filter criteria.
The 'average' aggregation is the total energy consumption for all buildings that meet the filter criteria, averaged across all the occurrences of the given time step within the selected month(s).
Note that while each time step within a day or a year has the same number of occurrences within each dataset, but that each time step for a week does not - some days of the week occur more times than others in each year or month range (except for February).
The viewer allows aggregations of up to six locations (states or PUMAs, depending on the dataset). When viewing a single location, choose the “+ More Locations” option, add up to five additional locations, and choose “Update Search”.
Sums of more than six locations can be created manually by downloading sums of up to six locations and summing further on your local computer.
TMY3 weather is not aligned between locations. If your application requires aligned weather, either use the 2018 dataset, or filter by weather station and sum only within a single weather station’s PUMAs.
Downloading a csv of the relevant data is the best approach. The data visualizations in the web viewer that include PV have some UI complexities that are still being worked out. We are also aware that the plot axes cut off negative values.
The peak day is the day with the highest single-hour (peak) energy consumption within the selected months.
The min peak day is the day with the lowest single-hour energy consumption within the selected months.
This is not currently available in the web interface, but you can use the interface to download the full year of 15-min data and see which day is the peak day.
For commercial buildings, the pre-aggregated timeseries files include a floor area column, so it is straightforward to divide energy use by the floor area to get intensity. Floor area is not currently included in the residential aggregates, but the floor area can be calculated from the metadata.tsv file (example), by adding up the values in the "floor_area_conditioned_ft_2" column after filtering down to the building type and geographic region corresponding to the pre-aggregated file.
In the data viewer, the bar graphs can show energy use intensity by selecting "energy_consumption_intensity" from the Output drop down menu. Timeseries data for energy use intensity are not directly available, but you can use the Building Characteristics viewer to download floor area values for a filtered subset of buildings and use that to convert timeseries energy use to timeseries energy use intensity.
While the pre-aggregated files (example) contain a column with the "floor_area_represented" for commercial or "units_represented" for residential, aggregations generated by the web viewer don’t include the "floor_area_represented" or "units_represented" information currently. Instead, you can find this information in one of two ways:
This is the total energy consumption by that end use within the selected months.
Yes! Please visit the End Use Load Profiles for the U.S. Building Stock project website for links to the data hosted on data.openei.org.
While the validation effort was largely focused on electricity, we did make some comparisons to annual and monthly EIA survey data for natural gas. These comparisons, which we used to inform the model improvements made during calibration, are published in the technical Methodology and Results report linked at the top of this FAQ. We did not do any timeseries comparisons for propane, fuel oil, or other fuels, although these fuels are included in the models.
All comparisons we completed as part of the calibration and validation effort are published in the technical Methodology and Results report linked at the top of this FAQ. In general, the comparisons are against anonymous hourly utility meter data, EIA monthly/annual data, and various end-use metered datasets.
Not directly. We made OpenStudio model input files (.osm) available in the dataset (ResStock example, ComStock example), which generate the EnergyPlus model input files. The residential models require external schedule .csv files (example).
Weather data used for the modeling have been provided in .csv format for regression modeling, forecasting, or other analyses. The TMY3 weather files in EnergyPlus input format (EPW) can be downloaded here, with filenames that correspond to county IDs in the ResStock/ComStock metadata.
EPW format weather files for 2018 or other actual meteorological years have not been publicly released. These files can be purchased from private sector vendors. See https://energyplus.net/weather/simulation for a list of providers.