A Quantile Regression Model for Electricity Peak Demand Forecasting: An Approach to Avoiding Power Blackouts

Electricity peak demand forecasting is a key exercise undertaken to avoid power blackouts and system failure. In this paper, the next day's load peak demand is estimated and forecasted. The challenge is to generate a peak demand forecast that is capable of avoiding the risk of a power blackout. We take an empirical approach to the question of estimating quantiles to indicate forecast uncertainty. Point forecasts generated from quantile regression are compared with the prediction intervals of linear regression. In addition, and to justify the result, their out-of-sample forecasting performance is compared. Distinctively from previous studies on load forecasting, models are evaluated based on their ability to avoid under-prediction i.e. avoid the risk of power blackouts. The analysis shows that quantile regression tends to under predict less than linear regression. Thus quantile regression is more appropriate for avoiding power blackouts.