This paper aims to investigate the impact of wire de-icing on the dynamics of transmission lines, response of conductor icing and de-icing and prediction model of de-icing height. Initially, a transmission tower-line system model incorporating wind loads is developed using Abaqus finite element analysis software. Subsequently, experiments under five different conditions are conducted to analyze the effects of critical parameters such as span length, height difference, ice thickness, and de-icing rate on the maximum de-icing height. A convolutional neural network and multi-head self-attention (CNN-MHAM)-based model is then constructed to predict the jumping height of conductors following ice shedding, with the model trained using both laboratory experimental data and field-collected data. Simulation validation demonstrates that the proposed de-icing height prediction model achieves an average relative error of less than 3% when compared with actual data. This study provides valuable insights for designing the insulation clearance of transmission lines in icing-prone areas.