Hierarchical Clustering for Anomalous Traffic Conditions Detection in Power Substations

Abstract

The IEC 61850 standard has contributed significantly to the substation management and automation process by incorporating the advantages of communications networks into the operation of power substations. However, this modernization process also involves new challenges in other areas. For example, in the field of security, several academic works have shown that the same attacks used in computer networks (DoS, Sniffing, Tampering, Spoffing among others), can also compromise the operation of a substation. This article evaluates the applicability of hierarchical clustering algorithms and statistical type descriptors (averages), in the identification of anomalous patterns of traffic in communication networks for power substations based on the IEC 61850 standard. The results obtained show that, using a hierarchical algorithm with Euclidean distance proximity criterion and simple link grouping method, a correct classification is achieved in the following operation scenarios: 1) Normal traffic, 2) IED disconnection, 3) Network discovery attack, 4) DoS attack, 5) IED spoofing attack and 6) Failure on the high voltage line. In addition, the descriptors used for the classification proved equally effective with other unsupervised clustering techniques such as K-means (partitional-type clustering), or LAMDA (diffuse-type clustering).