A distributed gaussian-means clustering algorithm for forecasting domestic energy usage

The adaptation of new technologies into the electrical energy infrastructure enables development of novel energy efficiency services. Introduction of smart meters into residential households allows collection of granular energy usage measures at frequent intervals. Analysis of such data could bring ample and detailed insights into the consumption behavior of households, allowing more accurate prediction of future loads. With the data intensive nature of these technologies, recent big data solutions allows harnessing of the enormous amounts of data being generated. We present a novel, scalable, distributed gaussian mean clustering algorithm for analyzing the energy consumption behavior of households in relation to different contributing factors such as weather conditions, type of day and time of the day. Based on forecasts of such contributing factors, we were able to predict a household's future energy usage much more accurately than other standard regression methods used for load forecasting.