Artificial Intelligence-Based Segregation of Power Grid Disturbance Records

Power grids, the lifeline of modern industry and society, often experience disturbances. In response, Disturbance Records (DRs) are generated, which capture fault-related data and are used to identify the type and location of faults. Moreover, an analysis of historical DRs can offer insights into the grid operations, for example, identifying the commonly occurring faults. However, apart from the genuine faults, a real-life substation can have DRs generated because of other reasons as well, such as manual operations and testing. Consequently, in order to gain any meaningful insight, it is imperative that such “real” and “test” DRs are segregated and only the real DRs are used for analysis. In this work, we address this problem by taking an Artificial Intelligence (AI)-based approach. In particular, we consider the DR waveform images and divide them into multiple clusters. Subsequently, we define and compute a score for each cluster by considering a sample of cluster members. Finally, based on the score, we identify the cluster that contains “test DRs,” while the other cluster(s) are inferred to contain “real DR.” We evaluate our prototype by considering DRs collected from multiple real-life substations. The results of the performance evaluation reveal that the real DRs can be identified with up to 97% recall.