摘要
针对传统方法对通信信道特征分类的精准度差、分类召回率较高,提出一种低压配电网HPLC通信信道特征分类方法。从位置、用户密度、电线长度与网络设计等方面,分析客户侧低压配电网HPLC通信信道特征;根据HPLC通信信道传递函数得到相关衰减因子数值,构建HPLC通信信道模型,获取最佳传输频率;利用K-L变换算法将复数信号分割为实部与虚部,建立信号的协方差矩阵,凭借实数变换操作获取降维后信道特征数据;将平均时延、角度扩展作为特征提取目标,通过支持向量机算法训练样本数据,得出超平面划分公式,计算最大分类间隔,完成特征分类。设备级运维应完善监测及诊断后的优化方案、云端部署方案,并考虑和现场作业终端的融合。建立一套标准的HPLC在线监测与故障诊断系统,该系统包括云端系统和手持式测量装置。仿真实验表明,所提方法的信道特征分类准度最高可达99.06%,召回率最高可达97.88%、整体精度最高可达99.54%,说明该方法分类效果较好。
As the traditional method for the communication channel feature classification has poor accuracy and high recall rate,this paper proposes an HPLC communication channel feature classification method for low-voltage distribution networks. The communication channel characteristics of the customer side lowvoltage distribution network are analyzed from the aspects of location,user density,wire length and network design;the relevant attenuation factor is obtained according to the transfer function of HPLC communication channel,and the HPLC communication channel model is constructed to obtain the best transmission frequency. The complex signal is divided into real part and imaginary part by K-L transform algorithm,the signal covariance matrix is established,and the channel characteristic data after dimensionality reduction is obtained by real transform operation.Taking the average delay and angle spread as the feature extraction target,the sample data are trained by support vector machine algorithm,the hyperplane division formula is obtained,the maximum classification interval is calculated,and the feature classification is completed. Equipment level operation and maintenance shall improve the optimization scheme and cloud deployment scheme after monitoring and diagnosis,and consider the integration with field operation terminal. A set of standard HPLC on-line monitoring and fault diagnosis system is established,which includes cloud system and handheld measurement device.Simulation results show that the channel feature classification accuracy of this method is up to 99.06%. The recall rate of channel feature classification is up to 97.88%,and the overall accuracy of channel feature classification is up to 99.54%.
作者
黄瑞
肖宇
曾伟杰
刘小平
HUANG Rui;XIAO Yu;ZENG Weijie;LIU Xiaoping(School of Electrical and Information Engineering,Hunan University,Changsha 410000,Hunan,China;State Grid Hunan Electric Power Co.,Ltd.,Changsha 410000,Hunan,China;Hunan Key Laboratory of Intelligent Electrical Measurement and Application Technology,Changsha 410000,Hunan,China)
出处
《电网与清洁能源》
北大核心
2022年第11期1-7,共7页
Power System and Clean Energy
基金
国家电网公司总部科技项目(5700-202027173A-0-0-00)。
关键词
机器学习
低压配电网
高速电力线载波
通信信道
支持向量机
machine learning
low voltage distribution network
high speed power line carrier
communication channel
support vector machine
