Icing Growth Model of Overhead Transmission Line on Multiple Machine Learning Algorithms

As the key equipment connecting regional power stations, substations and load, overhead transmission lines can be easily affected by ice disaster, which cause huge loss to the power system. Therefore, this paper establishes an icing growth model to predict the development trend of ice. First of all, the original data is preprocessed, and the feature variable with the most important rank is selected as the input. Secondly, nine machine learning algorithms are used to construct the prediction model, including three linear regression models (ridge regression, lasso regression, and elastic net regression), three single algorithm models (decision tree, K-nearest neighbors, and support vector regression), and three ensemble learning algorithms (gradient boosting regression, random forest, and adaptive boosting). Finally, the analysis is concluded that the decision tree, gradient lifting regression, random forest and adaptive lifting algorithm have demonstrated excellent performance on the test set. Furthermore, the predictive capability of these four optimal models is further validated by predicting new datasets.