Abstract

The forecasting of energy consumption is one of the key business motive for Phinergy srl, an analytic company, which utilises energy consumption of the aggregate customers smart meter ‘pods’ data to make forecast of future consumption during the hours of the day. A challenging issue is that the stochastic behaviour of some portions of customers cannot be relatively accurately predicted, and this high unpredictability yields high forecasting errors and causes any forecasting algorithm to struggle to gain high-precision results. The motivation for this work is to investigate the impact of clustering on improving the forecast accuracy of future energy consumption, by building forecasting models using a cluster of data samples which have distinct patterns in energy consumption. Results show that the addition of clustering as a pre-processing step is valuable in improving the accuracy of the forecast of electricity demand consumption.