Modifications of the clustering validity indicators for the assessment of the load profiling procedure

Load profiling aims at tracking exploitable information about the demand patterns of various consumer categories. The segmentation of the load curves corresponds to a clustering procedure, i.e. the grouping of load curves that are more similar to each other than those of another group. The interpretation of the clustering effectualness is done with a set of validity indicators or adequacy measures. The present study deals with the presentation and the examination of all clustering validity indicators that have been proposed in the load profiling literature. These indicators utilize the Euclidean distance as metric to address the similarity of the load curves that belong to the same cluster. The authors examine the behavior of the validity indicators by taking into account other metrics like the Manhattan and the Chebychev distances. This leads to a new definition of the adequacy measures that have been proposed in the literature.