Transformation in the South African power sector and new load patterns necessitated a review of load models used for financial, technical and tariff analysis. This pilot study took advantage of available data of customer measurements on medium Voltage (MV) feeders in Eskom's database. Load models with distinct profiles for typical days were developed for non-overlapping customer classes using a set of coherent parameters derived from MV chronological load profiles and the k-means algorithm. The

more »

... results suggested that two profiles can be used to for summer and two profiles can be used for winter instead of using 365 hourly profiles for simulations. The results also reveal that load classification can be improved when the parameters are directed towards specific objectives, and also when the k-means algorithm is supervised using exogenous (external) parameters of loads. A comparison of the results to the economic activity class suggests that there are sub-clusters identifiable within the economic classes. The proposed process is practical, implementable with available data and suitable for various studies on MV networks.