Transmission Lines Fault Identification Based on Cluster Tendency Visualization Techniques

In the present investigation, a clustering identification method to identify transmission lines faults is proposed. An important feature of this contribution relies in using real measurements stored in COMTRADE files of a three-phase power system, where the type of fault is not known. The VAT and iVAT algorithms are used as methods to determine if there is any pattern within the set of measurements that show if these have a clustering tendency. To this end, a strategy based on pre-processing and feature engineering is proposed to improve the clustering of faults using k-Means. Finally, a labeling of the dataset based on seven well-known types of electrical faults is proposed. The implemented methodology has allowed to cluster and classify the fault events using only the variances of the currents in each phase.