Tuesday, 30 January 2024: 9:45 AM
347/348 (The Baltimore Convention Center)
Electrical demand forecasting (load forecasting) is a necessary contributor to electrical grid planning and operation. Long-term load forecasting (extending over years or decades into the future) is needed to determine the mix of generation and transmission infrastructure that will supply sufficient electrical power to meet the needs of electricity users at all times with minimal risk of outages caused by a lack of capacity. Short-term forecasting (extending over minutes to days into the future) is needed to inform unit dispatch, in which grid operators determine which facilities will contribute power to the grid at each moment of the day, assuring stable frequency and voltage. The enterprise of load forecasting has increasingly porous boundaries, as weather-dependent generation, electrical storage using a range of technologies, and demand management become increasingly prominent on the grid. Load forecasters must now include behind-the-meter renewable generation (mostly solar) and battery storage, and in many cases are called on to produce net-load forecasts that include utility-scale front-of-meter generation in the forecast, so as to determine the amount of power needed from dispatchable sources, whether batteries, hydroelectric facilities or thermal generation plants. As industrial power requirements increase and transportation and heating-related loads increase (from electric vehicles and heat pump conversions), new temporal patterns of load variability are developing that must be understood and anticipated.
In this talk we will review EPRI's recent work improving the accuracy of our simulation of renewable generation and in modeling weather-dependent loads in a physically consistent framework, and as well as in accounting for technological shifts that are generating increasing electrical load over time. We will review research and development needs, including improvements in probabilistic models on both short and long timescales, and improvements in availability of forecast skill diagnosis, defining a path towards a comprehensive system of net-load modeling with built-in confidence measures for timescales from minutes to decades.

