基于深度卷积神经网络下单相智能电能表故障检测系统  

Single Phase Intelligent Energy Meter Fault Detection System Based on Deep Convolutional Neural Network

作　　者：吴泽新 WU Zexin(Shenzhen Power Supply Bureau Co.,Ltd.Luohu Power Supply Bureau,Shenzhen 518000,China)

机构地区：[1]深圳供电局有限公司罗湖供电局,深圳518000

出　　处：《自动化与仪表》2025年第2期106-110,共5页Automation & Instrumentation

摘　　要：考虑到智能电表故障的突发性和复杂性以及传统检测方法所面临的检测精度不足等问题,研究借助深度卷积神经网络进行单相智能电能表故障检测设计。该方法首先对神经网络进行拓扑结构优化和分类代价函数改进,以提高故障信息分类精度,随后对电能表故障模式及关联情况进行分析。结果表明,该方法在数据集上的诊断准确率超过90%,高于其他比较算法,且其在占比类型最多的2种故障上的关联准确率均超过了0.95,检测时间均小于25 s。研究提出的改进方法能有效评估电能表可靠性,保障电力安全稳定。Considering the suddenness and complexity of smart meter faults,as well as the problem of insufficient detection accuracy faced by traditional detection methods,this study uses the idea of deep convolutional neural networks to design single-phase smart energy meter fault detection.Firstly,this method optimizes the topology structure of the neural network and improves the classification cost function to improve the accuracy of fault information classification.Then,it analyzes the fault modes and correlation of the electric energy meter.The results show that the diagnostic accuracy of this method on the dataset exceeds 90%,which is higher than other comparison algorithms.Moreover,its correlation accuracy on the two types of faults with the highest proportion exceeds 0.95,and the detection time is less than 25 s.The proposed improvement method can effectively evaluate the reliability of electric energy meters and ensure the safety and stability of electricity.

关 键 词：深度卷积神经网络 单相智能电能表 超参数 FMEA 关联分析 

分 类 号：TM933[电气工程—电力电子与电力传动] TP183[自动化与计算机技术—控制理论与控制工程]