Advanced Clustering Techniques for Power System Data Analysis: Beyond k-Means

This panel presents advanced clustering techniques for power system data analysis. Clustering techniques play an important role to capture characteristics of data obtained. Clustering methods may be decided into hierarchical and non-hierarchical from a standpoint of the cluster structure. The former tries to construct a dendrogram that explains a data structure by a tree while the latter assigns data to clusters with the predetermined number. In this panel, non-hierarchical clustering methods are mainly discussed due to their widespread use in power systems. As one of the typical methods, k-means has been widely spread because of the straightforward algorithm. The formulation may be expressed as a nonlinear optimization problem that separates obtained data into clusters in a way that a sum of the Euclidean norm between the center and elements is minimized in clusters. However, it is well-known that the solution quality is not so good because the solutions easily get stuck in local minima, which means that the final solutions are affected by the initial ones significantly. As a result, the ingenuity of solution improvement is required to deal with data analysis appropriately. In this panel, non-hierarchical clustering techniques are discussed to improve the solutions of k-means in view of Trust-Tech of nonlinear optimization theory, Random Forest of machine learning, Fuzzy Clustering of Fuzzy Logic, and Evolutionary Computation. A comparison is made between k-means and advanced clustering techniques to understand the characteristics of the clustering methods. Presentations in this panel session: - Decision Support for Updating Electrical Components in Distribution Networks Using Intelligent and Clustering Techniques (23PESGM4113) - An Evolutionary Computation Technique for Data Clustering of Electricity Prices (23PESGM4114)