Traditional icing load prediction models exist many shortcomings, such as forecasting inaccuracy, casualness in choosing model parameters, and low prediction efficiency. Thus, an online prediction model based on the field micrometeorological data is proposed to predict the icing load of power transmission line. Firstly, this paper extracts effective information from micrometeorological data based on Principal Component Analysis (PCA), and optimizes the regression parameters by Genetic Algorithm (GA), and builds and trains offline LS-SVM training model. Secondly, online updating of regression function and prediction model is realized based on Karush-Kuhn-Tucker conditions and incremental online learning algorithm. Finally, the validity of the model is evaluated by related transmission lines of Yunnan Power Grid. Experimental results indicate that this method could predict the real-time icing load on overhead power lines, obtaining better performance in single-step and multi-step forecast than traditional icing load prediction models , which could serve for deicing and maintenance decision for power transmission and distribution system. This work is supported by National Natural Science Foundation of China (No. 61763049) and Science and Technology Plan of Applied Basic Research Programs Key Foundation of Yunnan Province (No. 2018FA032).