To further improve the accuracy of transmission line icing prediction, a prediction model based on an improved Harris hawks optimization(IHHO) algorithm optimizing hybrid kernel extreme learning machine(HKELM) is proposed. The hybrid kernel function is introduced into the kernel extreme learning machine to form HKELM. The IHHO algorithm is improved by strategies such as golden sine, nonlinear decreasing inertia weight, and Gaussian random walk. The IHHO algorithm is then utilized to optimize the weight vector and kernel parameters of HKELM, establishing a transmission line icing prediction model based on IHHO-HKELM. The input variables of the icing prediction model are determined by calculating the grey relational grade between meteorological factors and icing thickness. The results of case studies show that the mean square error, maximum error, and average relative error of the IHHO-HKELM model are 0.285, 0.860 mm, and 2.83%, respectively. The prediction effect is better than other models. Applying the icing prediction model in this paper to other icing lines can achieve good application effects and verify the superiority and practicality of the model.