Skip to main content
U.S. flag

An official website of the United States government

TELL: A Python Package for Predicting the Short- and Long-Term Evolution of Total Electricity Loads in the United States

Presentation Date
Tuesday, December 14, 2021 at 4:00pm
Convention Center - Poster Hall, D-F

Forecasting changes in electricity loads in response to anthropogenic and natural stressors is necessary for promoting energy system resilience. Given the pressures of aging infrastructure and the increasing integration of renewables, accurate load forecasts are critical for maintaining a stable grid and as a basis for long-term planning. Within the past two decades there have been rapid advances in both short-term (minutes to hours ahead) and long-term (months to years ahead) probabilistic load forecasting approaches. The general structure of these types of models are, understandably, quite different. Short- and medium-term load models most commonly relate meteorology and day-of-week parameters to loads. Longer-term models also use meteorology/climate as explanatory variables, but typically require bringing in “macro” variables like the decadal evolution of population, number of customers, or economic indicators. The Total ELectricity Load (TELL) model provides a framework that integrates aspects of both short- and long-term predictions of electricity demand in a coherent and scalable way. TELL takes as input gridded hourly time-series of meteorology and uses the temporal variations in weather to predict hourly profiles of total electricity demand for every county in the lower 48 United States using a multilayer perceptron (MLP) approach. Hourly predictions from TELL are then scaled to match the annual state-level total electricity loads predicted by the U.S. version of the Global Change Analysis Model (GCAM-USA). GCAM-USA is designed to capture the long-term co-evolution of the human-Earth system. Using this unique approach allows TELL to reflect both changes in the shape of the load profile due to variations in weather and climate and the long-term evolution of energy demand due to changes in population, technology, and economics. TELL is unique from other probabilistic load forecasting models in that it features an explicit spatial component that allows us to relate predicted loads to where they would occur spatially within a grid operations model. Now in the final stages of development and evaluation, TELL will soon be available to the community as an open-source Python package.

Funding Program Area(s)