Electromagnetic transient events (EMTE) classification in transmission grids

Electromagnetic transient events (EMTE) are short duration events with high frequency contents that may remain hidden from grid operators. EMTE lead to insulator failure mode in transmission system equipment (e.g., transformers) that develops over weeks, or even months. Therefore, early detection and classification of different categories of EMTE in transmission grids prioritize the targeted and preventive maintenance scheduling. This also prevents an eventual failure with an expensive replacement cost and reduced power grid reliability over time. This paper proposes a data-driven classification algorithm to identify potentially disruptive EMTE such as lightning versus de-energized transmission line switching, and energized transmission line switching. The EMTE classification is developed based on integration of an autoencoder with logistics function, known as 'softmax' regression. The input data to the classification algorithm is the transient voltage measurements. The performance of the proposed framework is validated through EMTP simulation of a simple transmission grid. The classification accuracy is evaluated with respect to several scenarios such as gradual increase of training data set and boosting.