A Unified Approach to Predicting Electricity Demand and Its Performance During Heatwaves

Goals:

This project aims to develop a unified approach for predicting electricity demand at both hourly and daily timescales, with forecast lead times ranging from hours to days. Building on previous research linking extreme heat metrics to electricity demand, the project will improve predictions during heatwaves and expand the approach so it can be applied across different regions of the United States. More accurate electricity demand forecasts can support grid resilience, improve load management, and help maintain a more stable electricity market during periods of extreme heat.

Objectives:

The project will extend previous daily electricity demand analyses to the hourly level and develop calibrated statistical models for predicting both hourly and daily demand. Researchers will also test Transformer-based time series foundation models, including Chronos-2 and Sundial, to evaluate their ability to predict electricity demand across different regions without extensive location-specific training. These models will be tested during major heatwave events, with heat metrics incorporated as additional predictors to determine whether they improve forecast accuracy. If necessary, the foundation models will be fine-tuned to further improve their performance and transferability across the United States.