电力系统自动化2025，Vol.49Issue(23)：89-98,10.DOI:10.7500/AEPS20240417004
台风灾害下基于非平衡样本数据的配电网杆塔受损预测
Tower Damage Prediction for Distribution Network Based on Unbalanced Sample Data Under Typhoon Disasters
摘要
Abstract
Given that extreme weather events with low probability but high impact often exhibit the characteristic of imbalanced sample data,a Wasserstein generative adversarial network(WGAN)based deep ensemble learning(DEL)model is proposed to achieve the accurate prediction of damaged towers in distribution network in scenarios with imbalanced sample data.Firstly,using preprocessed meteorological,geographical,and electrical data as inputs,the WGAN is employed to generate samples of damaged towers.Subsequently,a DEL model is constructed through Stacking,utilizing Bagging-integrated random forest and boosting-integrated gradient boosting decision tree as base learners,and a neural network as a meta-learner.Finally,taking the distribution network towers in Zhanjiang City of China during Typhoon Talim(No.4 in 2023)as an example,the predicted distribution of damaged towers is visualized.The model is evaluated for interpretability using the SHAP method to analyze the influence mechanism of feature variables on the prediction results.The results demonstrate that,compared to traditional data augmentation and machine learning algorithms,the proposed model exhibits higher prediction accuracy and superior data information mining capabilities,enabling effective tower damage prediction in distribution network even when data samples are insufficient.关键词
极端天气/台风/配电网/杆塔/受损预测/生成对抗网络/Wasserstein距离/深度集成学习Key words
extreme weather/typhoon/distribution network/tower/damage prediction/generative adversarial network/Wasserstein distance/deep ensemble learning引用本文复制引用
高富,侯慧,魏瑞增,王磊,李正天,林湘宁..台风灾害下基于非平衡样本数据的配电网杆塔受损预测[J].电力系统自动化,2025,49(23):89-98,10.基金项目
国家自然科学基金资助项目(U22B20106) （U22B20106）
中国南方电网有限责任公司科技项目(GDKJXM20231426). This work is supported by National Natural Science Foundation of China(No.U22B20106)and China Southern Power Grid Company Limited(No.GDKJXM20231426). （GDKJXM20231426）
