Application of competitive learning clustering in the load time series segmentation

Load time series segmentation can serve as the basis for the implementation of variety of applications that have the potential to modify the demand patterns. The scope of this study is three-fold. Firstly, a novel modeling technique of the metered load data of a high voltage industrial consumer is introduced. Instead of representing the daily load curve with a vector with T elements, where T is the time interval of the metering, it is proposed to represent the demand with six indicators that are related with the shape of the curve. Secondly, a new clustering algorithm is introduced in the load time series segmentation field of research. Lastly, a new clustering validity indicator is proposed that can provide an accurate evidence on the optimal number of clusters. The data under study are the active and reactive metered load of a full year.