Abstract

This paper illustrates how Big Data may be used to predict lightning outages in the transmission system. A comprehensive database that contains the necessary information about the historical lightning related events is developed and utilized for training of the prediction algorithm. A Mixture-of-Experts model incorporating multiple spatially aware logistic regression models is utilized to calculate highly accurate short-term predictions 1-3 hours in advance. Lightning strikes to the transmission lines were considered in this study. Different durations of lightning related failures were taken into account, including both temporary and permanent faults. The predictions allow a smart decision-making approach to implementing the proposed mitigation techniques. The model was tested using real utility data. The results demonstrate the capability of the algorithm to predict lightning outage probability with high accuracy for a specific location. The prediction accuracy of the developed algorithm is 0.9370, with the Area Under the Curve being 0.7576. This suggests that the algorithm is good at both predicting high probability for cases with outages, and low probability for cases without outages. The outage probabilities are calculated in real time for every substation and transmission line in the network. Thus, this research provides a significant improvement over the existing studies that aggregate the lightning outage expectancy over a larger geographical area.