Prediction of galloping accidents in power lines using statistical analysis

Abstract

leaving only seven physical variables, two categorical, and five continuous. Before forming estimation model, with 84 provided galloping cases, 840 non-galloped cases were chosen out of 13 billion cases, and with backward elimination, number of variables were again reduced to 10. By using logistic regression model and support vector machine, estimation model for galloping phenomenon has been formed. Validation has been conducted with 4-fold validation method and corresponding AUC value of ROC curve has been used to assess the discrimination level of estimation models. As the result, both logistic regression analysis and SVM methods effectively discriminated the power lines that experienced galloping from those that did not.

Galloping is one of the most serious vibration problems in transmission lines. Transmission power lines can be extensively damaged owing to aerodynamic instabilities caused by ice and snow accretion and also galloping can lead to abrupt voltage drops and interrupt power transmission. Thus, a better understanding of this phenomenon is necessary to predict and prevent galloping in transmission lines. In this study, the conditions under which galloping occurs in transmission lines were analyzed using logistic regression and support vector machine (SVM) methods. Since there were limited number of weather observatories compared to transmission towers, 676 observatories while 50,000 transmission towers exist, lying in the country land, interpolations of weather factors were conducted in prior to forming galloping estimation model. For physical factors, KEPCO has provided 122 variables for transmission towers. Because of the large number of explanatory variables, histogram analyses and logistic regression have been conducted to sort out the only meaningful variables for further use