Deep Learning-Based Risk Prediction Technology for Overhead Lines under Cold Wave Weather Using Line Tension Data

Aiming at the problems of icing breakage and tower collapse of transmission lines under cold wave weather and the lack of historical data and unknown mechanism of icing, this paper proposes a risk prediction model for overhead lines under cold wave weather based on the deep learning of line tension data from the perspective of transmission line mechanics. First, the line tension threshold calculation model for each voltage level is established to estimate the line tension risk threshold under cold wave weather. Aiming at the characteristics of the cold wave weather sample data, the sample extension dataset is trained using an adversarial generative network. Then, based on the Long Short-Term Memory (LSTM) network, a predictive learning model is built with the expanded time-series data samples to study the mapping relationship between the line pull increment of linear towers and meteorological data under cold wave weather. Using an example from a province in China, it is demonstrated that the sample enhancement algorithm can improve the risk prediction performance through sample learning; compared with other common prediction models, the tension prediction model based on multi-model fusion has better prediction stability.