摘要
随着新能源汽车数量的激增,电动汽车充电设施的稳定运行对电网安全和用户权益保障变得尤为关键。本文针对电动汽车充电桩的故障预测进行了深入分析,首先基于核密度估计研究用户的充电行为,探讨了充电起始、持续时间和结束时刻的时序相关性,并据此提出了非欧几里得域数据建模方法。进一步,研究引入图卷积神经网络(GCN)和卷积神经网络(CNN),搭建了一个GCN-CNN深度学习联合模型,有效捕捉故障分类与数据特征间的复杂非线性关系。通过在真实数据集上进行消融和算法对比实验,本模型在验证集上取得了F_(1score)和G-mean均为0.844的优越性能,较其他模型平均性能分别提升了6.28%和6.04%。该研究为充电桩故障预测提供了创新解决方案,有助于降低运维成本并提升检测效率。
With the surge in the number of new energy vehicles,the stable operation of electric vehicle charging facilities has become particularly critical for grid security and user rights protection.In this study,an in-depth analysis is conducted for the fault prediction of EV charging piles.Firstly,the user’s charging behavior is studied based on kernel density estimation,and the temporal correlations of charging onset,duration,and end moments are explored,and a non-Euclidean domain data modeling method is proposed accordingly.Further,the study introduces Graph Convolutional Neural Network(GCN)and Convolutional Neural Network(CCN),develops a GCN-CNN joint deep learning model to effectively capture the complex nonlinear relationship between fault classification and data features.Through ablation and algorithm comparison experiments on real datasets,this model achieves a superior performance of 0.844 for both F_(1score)and G-mean on the validation set,which improves the average performance over other models by 6.28%and 6.04%,respectively.This study provides an innovative solution for charging pile fault prediction,which helps to reduce O&M costs and improve detection efficiency.
作者
王亚超
党兆帅
李学超
韩迪
戚成飞
毕超然
杨挺
WANG Yachao;DANG Zhaoshuai;LI Xuechao;HAN Di;QI Chengfei;BI Chaoran;YANG Ting(State Grid Jibei Electric Power Co.,Ltd.Metrology Center,Beijing 100032,China;School of Electrical and Information Engineering,Tianjin University,Tianjin 300072,China;State Grid Jibei Electric Power Co.,Ltd.Tangshan Power Supply Company,Tangshan 063099,China)
出处
《电工电能新技术》
北大核心
2025年第8期119-128,共10页
Advanced Technology of Electrical Engineering and Energy
基金
国网冀北电力有限公司科技项目“电动汽车非车载充电机智能化现场检定关键技术研究”(520185220006)。
关键词
电动汽车充电桩
充电桩故障预测
图卷积神经网络
联合网络模型
深度学习
electric vehicle charging station
charging pile failure prediction
graph convolutional neural network
joint network model
deep learning
