数据驱动的台风灾害下配网10k V杆塔受损空间分布预测  

Data-driven Prediction of Spatial Distribution of Damage to 10kV Towers in Distribution Network Under Typhoon Disaster

作　　者：侯慧[1] 俞菊芳 谢宇风 申原 黄勇 朱韶华 HOU Hui;YU Jufang;XIE Yufeng;SHEN Yuan;HUANG Yong;ZHU Shaohua(School of Automation,Wuhan University of Technology,Wuhan 430070,Hubei Province,China;Guangdong Power Grid Co.,Ltd.,Guangzhou 510000,Guangdong Province,China;Electric Power Research Institute of Guangdong Power Grid Co.,Ltd.,Guangzhou 510080,Guangdong Province,China;Power Remote Sensing Technology Joint Laboratory of China Southern Power Grid,Guangzhou 510080,Guangdong Province,China)

机构地区：[1]武汉理工大学自动化学院,湖北省武汉市430070 [2]广东电网有限责任公司,广东省广州市510000 [3]广东电网有限责任公司电力科学研究院,广东省广州市510080 [4]南方电网电力遥感技术联合实验室,广东省广州市510080

出　　处：《电网技术》2021年第9期3681-3689,共9页Power System Technology

基　　金：教育部产学合作协同育人项目(201902056044);中国南方电网有限责任公司科技项目(GDKJXM20198382)。

摘　　要：相比于主网,配网杆塔因设计规范相对较低且数量庞大,在台风灾害下更易发生大规模断杆、倒杆等事故。为此,文章结合气象、电网和地理信息,提出一种数据驱动的台风灾害下10kV杆塔受损空间分布预测方法。首先,对收集和提取的数据进行标准化、分类变量独热编码、样本均衡等处理以提高数据质量,并基于皮尔逊相关系数绘制相关性热力图,根据相关性分析结果选择合适的变量作为最终输入数据;其次,利用Ada Boost回归、梯度提升回归、K近邻回归、随机森林、支持向量回归等5种机器学习算法建立配网10kV杆塔受损空间分布预测模型;再次,基于层次分析法和熵权法进行指标综合赋权,并对各预测模型进行综合打分,实现最优模型选取;最后,以台风"山竹"下江门市10kV杆塔为例,对杆塔受损空间分布预测结果进行可视化,对比实际杆塔受损分布情况,验证了所提方法的科学性和可行性。Compared with the main network,the distribution network towers are more prone to large-scale tower breakage and collapse accidents due to relatively low design specifications and a large number of towers under typhoon disasters.To this end,combined with meteorological,power grid and geographic information,a data-driven prediction of spatial distribution of damage to 10 k V towers in distribution network under typhoon disaster is proposed.First,the collected and extracted data are standardized,classified variables are encoded by One-Hot,and the sample is balanced to improve the data quality.Based on the Pearson correlation coefficient,the correlation thermal diagram is drawn,and the appropriate variables are selected as the final input data according to the correlation analysis results.Secondly,five machine learning algorithms including Ada Boost regression,gradient boosting regression,K nearest neighbor regression,random forest and support vector regression are used to establish the spatial distribution prediction model of 10 kV tower damage in distribution network.Thirdly,the comprehensive weighting of indicators is carried out based on the analytic hierarchy process and the entropy weight method.Each model is scored to select the optimal model selection.Finally,taking the 10 kV towers in Jiangmen City under typhoon"Mangkhut"as an example,the prediction results of the damage spatial distribution of the towers were visualized,and the actual damage distribution of the towers was compared to verify the scientificity and feasibility of the proposed method.

关 键 词：台风灾害 杆塔受损 数据驱动 综合赋权 空间分布预测 

分 类 号：TM721[电气工程—电力系统及自动化]