基于卷积神经网络的配电网馈线故障预测方法研究  

Research on Feeder Fault Prediction Method of Distribution Network Based on Convolutional Neural Network

作　　者：龚杰 GONG Jie(State Grid Jiangsu Electric Power Co.,Ltd.,Huai’an Hongyang District Power Supply Branch,Huai’an 223100,China)

机构地区：[1]国网江苏省电力有限公司淮安市洪泽区供电分公司,江苏淮安223100

出　　处：《通信电源技术》2024年第8期4-6,共3页Telecom Power Technology

摘　　要：针对配电网馈线故障预测中,均方误差(Mean Square Error,MSE)较高和召回率较低导致无法精准预测馈线故障的问题,文章提出基于卷积神经网络的配电网馈线故障预测方法。利用无人机搭载图像传感器采样馈线图像数据,通过卡尔曼滤波算法降噪处理采集的数据,并且结合箱型图法处理单维属性异常值,引入卷积神经网络,将预处理后的数据作为输入,提取配电网馈线故障特征,并且引入误差补充项和L1正则化方法进行优化,预测配电网馈线健康度,识别配电网馈线故障,从而实现配电网馈线故障预测。经实验证明,该方法预测结果的MSE低于0.1,召回率高于98%,其预测结果具备较好的可靠性。Aiming at the problem of high Mean Square Error(MSE)and low recall rate in the prediction of feeder faults in distribution networks,the article proposes a method based on convolutional neural network for the prediction of feeder faults in distribution networks.The feeder image data are sampled using UAV-mounted image sensors,the collected data are processed by Kalman filter algorithm for noise reduction,and combined with the box plot method to deal with the unidimensional attribute outliers,a convolutional neural network is introduced,and the preprocessed data are used as the inputs to extract the distribution grid feeder fault features,and the error complementary term and L1 regularization method are introduced for optimization,to predict the health degree of the distribution grid feeder and to identify the distribution grid feeder faults,so as to realize the prediction of distribution network feeder faults.It is proved that the MSE of this method is lower than 0.1,the recall rate is higher than 98%,and the prediction results have good reliability.

关 键 词：卷积神经网络 配电网 馈线 故障预测 

分 类 号：TM715[电气工程—电力系统及自动化] TP183[自动化与计算机技术—控制理论与控制工程]