Icing image of Baoji power transmission line in January 2025

Collected by the "Ice Observing Elf" image sensor installed on the research area line for real-time monitoring of ice cover thickness. The data processing and analysis methods are as follows: meteorological data is located and matched with stations using the Haversine formula, evaluated for reliability using weighted methods, and finely processed using interpolation methods. It is then input into the WRF numerical weather model to simulate key meteorological elements; The icing image is preprocessed, wire pixel extraction, and width measurement are completed through machine vision technology to calculate the icing thickness; After feature engineering processing, multi-source data is used to train the XGBoost risk assessment model. The final results show that after fine processing of meteorological data, the relative error is ≤ 10%, the monitoring error of ice cover thickness is ≤ 3%, and the prediction error is ≤ 5%. The visualization system based on the three types of data can accurately predict and grade the probability of ice cover tripping for 1-3 hours and 3-6 hours, providing reliable data support for intelligent operation and maintenance of the power grid.

本数据集由搭载于研究区域线路的“冰情观测精灵”图像传感器采集，用于实时监测覆冰厚度。其数据处理与分析流程如下：首先利用哈弗辛公式（Haversine formula）对气象数据进行站点定位匹配，采用加权法评估其可靠性，并通过插值方法完成精细化处理；随后将处理后的气象数据输入WRF数值天气预报模型，以模拟关键气象要素。针对覆冰图像，则通过机器视觉（Machine Vision）技术完成预处理、导线像素提取与宽度测量，进而计算覆冰厚度。经特征工程处理后，基于多源数据训练得到XGBoost风险评估模型。最终实验结果表明，经精细化处理的气象数据相对误差≤10%，覆冰厚度监测误差≤3%，预测误差≤5%。基于三类数据构建的可视化系统可精准预测并分级1-3小时、3-6小时区间内的覆冰跳闸概率，为电网智能运维提供可靠的数据支撑。