Electricity demand forecasting of single residential units

Abstract

The introduction of demand side Advanced Metering Infrastructures in power distribution grids, allows the collection of huge amount of valuable information about energy usage. Utilities are already exploiting such information through Demand Side Management and Forecasting Algorithms that have been proved to help reducing the overall electricity demand. To push further this “green” trend toward the realization of Smart Grid, we propose to apply the forecasting techniques also to the residential users electricity demand. Exponential smoothing forecasting has been demonstrated to be effective to analyze and to provide trends for higher scale (National or Regional level) of the demand. We tested and moved the approach to residential users and assessed the performance when data have high time variability. Two different datasets have been used and the accuracy of the forecasting has been compared with the performance of the same predictors when national level data are used. Our tests show encouraging results, even if the prediction's accuracy is much lower when dealing with single users and the importance of the pre-filtering of the collected data is fundamental.