Different ice types of overhead transmission lines have different harm to the power grid, which effect the decision-making of anti-icing and de-icing. Aiming at the problems of poor image quality, low utilization rate and weak generalization ability in ice type image recognition, this paper studies ice types identification and prediction of overhead transmission lines driven by micro-meteorological data of three consecutive days icing. Based on the shooting time and terminal number data of the icing images, the icing image, micro-meteorological, terminal and tower data from the transmission line icing monitoring system of China Southern Power Grid during 2014 to 2018 are merged, so that a micro-meteorological dataset of ice types is constructed. The micro-meteorological characteristics and geographical distribution characteristics of ice types such as glaze, rime, mixed rime and wet snow are statistically analyzed. The k-nearest neighbors (KNN) classification method is proposed to identify and predict the ice types of 1 269 samples in test sets by using 5 075 samples training sets of 9 micro-meteorological parameters and ice types in three consecutive days. The identification accuracy can reach more than 80%, and the percision(P), racall(R)and accuracy(A) are 86.7%, 86.6% and 92.2% respectively. The results show that the proposed method can greatly improve the utilization rate, analysis efficiency and generalization ability of overhead transmission line icing data.