Demand flexibility to support probabilistic load forecasting concerns in residential scenarios

Load Forecasting usually deals with the point prediction of electricity consumption at different aggregation levels. Although the literature in this field is broad, it is quite small if we refer to the probabilistic prediction of demand, which might be particularly useful in operation and planning. In this paper, authors show some applications of probabilistic prediction in Demand Response. Quantile Regression Forest, a machine learning technique, is applied to obtain the quantiles of the conditional distribution function giving a set of probabilistic forecasts. From these predictions, demand flexibility strategies are proposed to reduce the incertitude of forecasts. Possibilities are analyzed considering Physical-Based Load Model to evaluate the potential of demand flexibility of two end-uses: Water Heating and Electric Heating. The proposal is illustrated with hourly demand data from a small Spanish city (around 5 MW of peak demand), assuming the partial control of both loads.