New adequacy measures for the evaluation of the load profiling process

The process of grouping load curves based on the similarity of their shapes is represented by unsupervised machine learning. Usually, in the load profiling problems, there is no available information about the number of desired clusters. The load data are grouped together and the objective is to minimize various indexes or adequacy measures that are related with the distances between the data within the same cluster. This paper presents all the adequacy measures that have been proposed in the load profiling related literature. Some of these measures show unstable behavior while the number of the output clusters increases. Hence, they are not suitable for defining the optimal number of clusters. Two new adequacy measures, used in other clustering problems are introduced, for easy detection of the appropriate number of clusters. Additionally, two demand pattern representation techniques are compared in terms of minimizing the clustering error.