Streaming Data Oriented Galloping Prediction

The galloping of ice-coated transmission lines keeps posing challenges to the reliable operation of power systems. Hence, it is crucial for the system operator to be able to accurately predict such events. However, this is no easy task due to the complex physical dynamics and the limited collected data. With the advance of metering technology, many advanced data-driven predictors have been proposed with remarkable performance. However, these predictors are mostly designed for offline applications. In this paper, we propose to use the Hoeffding tree based random forest to predict conductor galloping. To tackle the challenge brought by imbalanced data sets, we introduce the notion of the reinforcement training process. The resulting predictor enjoys the nice property from the Hoeffding tree: the tree structure is not sensitive to new data with a light computational burden. Hence, our proposed predictor is ideal to process streaming data, in contrast to those in the literature.