In this presentation, we discuss methodologies for forecasting peak demand. Annual peak demand is anomalous, and the forecasting methodology must take this into account.  Current approaches largely rely on OLS, which is best suited to forecasting average consumption and not peak demand:

  • OLS with one data point for each annual peak
  • OLS on a subset of relatively extreme days

We propose Quantile Regression (QR) as a superior solution.  We compare QR and OLS methods of the same functional form using 32 “utilities” in a meta-study.

View Presentation

Experts Involved
Other Contributing Authors
Related files
Quantile Regression for Peak Demand Forecasting