覆冰严重威胁架空输电线路安全运行，由于雨凇、混合凇、雾凇、湿雪等覆冰类型对架空输电线路危害程度不同，采取的除冰措施也不同。当架空输电线路覆冰达到一定程度时会根据覆冰类型做出启动融冰、采用机械除冰、调整运行方式等不同的的运维决策，预测覆冰类型可以了解未来覆冰风险。在图像识别绝缘子覆冰类型研究基础上，提出了监测图像融合前3天微气象时间序列的数据驱动覆冰类型预测模型。基于南方电网2014—2021年覆冰监测数据，以图像拍摄时刻搜索所在终端最临近微气象监测时刻，再把该时刻及其前3天微气象时间序列与该图像融合成一个样本，构建用于覆冰类型数据驱动模型的图像与微气象融合数据集。以网格搜索法-极致梯度提升树 （grid search method-extreme gradient boosting，GSM-XGBoost）为模型算法，以前3天间隔6 h微气象时间序列为输入，以图像识别的覆冰类型为输出，训练集为4 503个融合样本、测试集为1 931个融合样本时，数据驱动覆冰类型预测模型的宏精确率P_m、宏召回率R_m和宏平均F1分数分别为95.0 %、96.3 %和95.6 %，实现了架空输电线路覆冰类型准确预测。

The safety of overhead transmission lines is threatened by icing. Due to the different degrees of damage caused by ice types including glaze, mixed rime, rime, and wet snow to overhead transmission lines, the de-icing measures taken are also different. When the icing of overhead transmission lines reaches a certain level, appropriate operation and maintenance decisions will be made according to different icing type, including melting the ice, using mechanical removal and adjusting the operation mode. Icing type prediction can provide insights into future icing risks. On the basis of study on identifying insulator icing types through images, a data-driven ice type prediction model is proposed, which fuses monitoring images with micrometeorological data from the past three days. Based on the icing monitoring data of China Southern Power Grid from 2014 to 2021, the nearest micrometeorological monitoring time from the same terminal is searched according to the image capture time. Micrometeorological time series from this time and the past three days is combined with the image to form a sample, constructing a fused dataset of images and micrometeorological data for the data-driven ice type prediction model. The Grid Search Method-eXtreme Gradient Boosting （GSM-XGBoost） is used as the model algorithm, and the micrometeorological time series at 6h intervals from the past three days are used as inputs. The icing types identified from images are used as outputs. With 4 503 fused samples of the training set and 1 931 fused samples of the test set, the macro precision （P_m）, macro recall （R_m）, and macro F1 score of the data-driven icing type prediction model are 95.0 %, 96.3 %, and 95.6 %, respectively. The icing type prediction for overhead transmission lines is achieved accurately.