PUDL US Hourly Electricity Demand by State

Hourly Electricity Demand by State

This archive contains the output of the Public Utility Data Liberation (PUDL) Project state electricity demand allocation analysis, as of the v0.4.0 release of the PUDL Python package. Here is the script that produced this output. It was run using the Docker container and processed data that are included in PUDL Data Release v2.0.0.

The analysis uses hourly electricity demand reported at the balancing authority and utility level in the FERC 714 (data archive), and service territories for utilities and balancing authorities inferred from the counties served by each utility, and the utilities that make up each balancing authority in the EIA 861 (data archive), to estimate the total hourly electricity demand for each US state.

We used the total electricity sales by state reported in the EIA 861 as a scaling factor to ensure that the magnitude of electricity sales is roughly correct, and obtains the shape of the demand curve from the hourly planning area demand reported in the FERC 714. The scaling is necessary partly due to imperfections in the historical utility and balancing authority service territory maps which we have been able to reconstruct from the data reported in the EIA 861 Service Territories and Balancing Authority tables.

The compilation of historical service territories based on the EIA 861 data is somewhat manual and could be improved, but overall the results seem reasonable. Additional predictive spatial variables will be required to obtain more granular electricity demand estimates (e.g. at the county level).

FERC 714 Respondents

The file ferc714_respondents.csv links FERC Form 714 respondents to what we believe to be their corresponding EIA utilities or balancing authorities.

• eia_code: An integer ID reported in the FERC Form 714 corresponding to the respondent's EIA ID. In some cases this is a Utility ID, and in others it is a Balancing Authority ID, but which is not specified and so we have had to infer the type of entity which is responding. Note that in many cases the same company acts as both a utility and a balancing authority, and the integer ID associated with the company is often the same in both roles, but it does not need to be.
• respondent_type: Either balancing_authority or utility depending on which type of entity we believe was responding to the FERC 714.
• respondent_id_ferc714: The integer ID of the responding entity within the FERC 714.
• respondent_name_ferc714: The name provided by the respondent in the FERC 714.
• balancing_authority_id_eia: If the respondent was identified as a balancing authority, the EIA ID for that balancing authority, taken from the EIA Form 861.
• balancing_authority_code_eia: If the respondent was identified as a balancing authority, the EIA short code used to identify the balancing authority, taken from the EIA Form 861.
• balancing_authority_name_eia: If the respondent was identified as a balancing authority, the name of the balancing authority, taken from the EIA Form 861.
• utility_id_eia: If the respondent was identified as a utility, the EIA utility ID, taken from the EIA Form 861.
• utility_name_eia: If the respondent was identified as a utility, the name of the utility, taken from the EIA 861.

FERC 714 Respondent Service Territories

The file ferc714_service_territories.csv describes the historical service territories for FERC 714 respondents for the years 2006-2019. For each respondent and year, their service territory is composed of a collection of counties, identified by their 5-digit FIPS codes. The file contains the following columns, with each row associating a single county with a FERC 714 respondent in a particular year:

• respondent_id_ferc714: The FERC Form 714 respondent ID, which is also found in ferc714_respondents.csv
• report_date: The first day of the year for which the service territory is being described.
• state: Two letter abbreviation for the state containing the county, for human readability.
• county: The name of the county, for human readability.
• state_id_fips: The 2-digit FIPS state code.
• county_id_fips: The 5-digit FIPS county code for use with other geospatial data resources, like the US Census DP1 geodatabase.

State Hourly Electricity Demand Estimates

The file demand.csv contains hourly electricity demand estimates for each US state from 2006-2019. It contains the following columns:

• state_id_fips: The 2-digit FIPS state code.
• utc_datetime: UTC time at hourly resolution.
• demand_mwh: Electricity demand for that state and hour in MWh. This is an allocation of the electricity demand reported directly in the FERC Form 714.
• scaled_demand_mwh: Estimated total electricity demand for that state and hour, in MWh. This is the reported FERC Form 714 hourly demand scaled up or down linearly such that the total annual electricity demand matches the total annual electricity sales reported at the state level in the EIA Form 861.

A collection of plots are also included, comparing the original and scaled demand time series for each state.

Acknowledgements

This analysis was funded largely by GridLab, and done in collaboration with researchers at the Lawrence Berkeley National Laboratory, including Umed Paliwal and Nikit Abhyankar.

• Ethan Welty wrote the final code and most of the algorithms.
• Yash Kumar did initial data explorations and geospatial analyses.

The data screening methods were originally designed to identify unrealistic data in the electricity demand timeseries reported to EIA on Form 930, and have been applied here to data form the FERC Form 714.

They are adapted from code published and modified by:

• Tyler Ruggles
• Greg Schivley

And described at:

• Developing reliable hourly electricity demand data through screening and imputation
• EIA Cleaned Hourly Electricity Demand Code (Zenodo)
• EIA Cleaned Hourly Electricity Demand Code (GitHub)

The imputation methods were designed for multivariate time series forecasting.

They are adapted from code published by:

• Xinyu Chen chenxy346@gmail.com

And described at:

• Low-Rank Autoregressive Tensor Completion for Multivariate Time Series Forecasting
• Scalable Low-Rank Tensor Learning for Spatiotemporal Traffic Data Imputation
• Tensor Learning (张量学习)

About PUDL & Catalyst Cooperative

For additional information about this data and PUDL, see the following resources:

• The PUDL Repository on GitHub
• The PUDL Documentation
• Other Catalyst Cooperative data archives on Zenodo