Machine Learning-Based Study of Wire Icing Growth under Coexisting Freezing Rain and Supercooled Fog

To investigate the mechanism for wire icing growth and simulate the icing growth rate, this study analyzed two wire icing cases observed in Lushan, Jiangxi Province of South China during winter 2016 under conditions of coexisting freezing rain and supercooled fog weather. By combining meteorological elements with microphysical parameters of mixed freezing rain and supercooled fog, the correlation between the icing growth rate and these factors was examined. Four widely used machine learning models: Random Forest (RF), Support Vector Machine (SVM), Convolutional Neural Network (CNN), and Extreme Learning Machine (ELM) were employed to simulate the icing growth process. The results indicate that in both cases, rain rate, temperature, wind speed (≥ 1 m s^-1), and wind direction exhibit a significant positive correlation with the icing growth rate. In the stronger wind scenario (Case 1), the icing growth rate was negatively correlated with the number concentration, liquid water content, mean diameter, and mean volumetric diameter of hydrometeor particles, with wind speed being the most important factor in the RF model. The icing growth was primarily driven by the increased rain rate, which subsequently led to higher liquid water content. In the weaker wind scenario (Case 2), the icing growth rate exhibited a positive correlation with number concentration, liquid water content, and mean diameter, with the mean volumetric diameter of supercooled fog droplets being the most important factor in the RF model. The growth of icing was primarily due to an increase in the number of water-phase particles and an overall increase in particle size, leading to an increase in liquid water content. All four machine learning models successfully simulated the icing growth process, yielding results that outperformed those from traditional empirical formulas and numerical simulations. Among them, the RF, SVM and CNN models demonstrated particularly strong performance.