Quality assessment of clusters of electrical disturbances: a case study

Electrical disturbances can have an adverse affect on people, businesses and other systems, and increased understanding of such events has huge potential benefits. Clustering or unsupervised learning is a technique of computational intelligence that can be used to identify natural clusters or groups of disturbances. The understanding of disturbances is important for planning and maintenance, and may lead to fresh insights which prove useful in upgrade of infrastructure. Ideally this should be a fully automated process, so that no bias is introduced by the practitioner. The complete process involves several steps including data cleaning, transformation, feature selection, clustering, evaluation of clusters, cluster description and cluster interpretation. We designate this process as the Clustering Knowledge Chain, beginning with raw data and ending with new knowledge. This case study examines each of these steps, showing how they might be applied in a situation involving real-world data, and illustrates some of the difficulties that a practitioner with domain knowledge may encounter. Results from this case study reveal new knowledge about electrical disturbances, but also show that selection of parameters using a quantitative measure of clustering quality is not enough, by itself, to guarantee clusters that can inform the practitioner