基于CNN-SVM的三相四线接线异常诊断算法
Research on Three-Phase Four-Conductor Wiring Anomaly Diagnosis Algorithm Based on CNN-SVM
闫浩 1张恒超 2李敏 2王巨灏 2王琨 1徐磊 1王纲 1贺昊辰1
作者信息
- 1. 江南大学机械工程学院,江苏无锡 214122
- 2. 国网江苏省电力有限公司苏州供电分公司,江苏苏州 215004
- 折叠
摘要
三相四线电能装置错接线类型多,情况复杂,给电能计量带来了很大误差,影响了供电公司的经济效益.传统判别方法依靠人员经验,速度慢、效率低.结合深度学习,提出一种基于卷积神经网络(convolutional neural network,CNN)和支持向量机(support vector machine,SVM)的接线异常诊断算法.该诊断算法将典型错接线类型重新划分为 5 类.使用CNN对电压、电流、有功功率和无功功率等 11 个特征进行自适应特征提取,利用改进后的混沌自适应灰狼算法对SVM的参数进行优化选择,并利用优化后的SVM对提取后的数据进行分类.运用实际数据对所提方法进行验证,并与另外3 种模型的结果进行对比,结果表明,经过混沌自适应灰狼优化后的CNN-SVM收敛效果更好,且准确率比传统CNN提升 9.6%,验证了所提算法具有较好的判别精度以及稳定性.
Abstract
Three-phase four-conductor electric energy devices are prone to have many types of miswiring and complicated situations,which bring great errors to electric energy metering and affect the economic benefits of power supply companies.The traditional discrimination method mainly relies on personnel experience,so it is slow and inefficient.In this context,combined with deep learning,this paper proposes a wiring abnormality diagnosis algorithm based on convolutional neural network(CNN)and support vector machine(SVM).The diagnostic algorithm first reclassifies the typical miswiring types into five categories.Second,11 features such as voltage,current,active power,reactive power and other features are adaptively feature extracted using CNN,after which the parameters of SVM are optimally selected using the improved chaotic adaptive gray wolf algorithm,and then the optimized SVM is used to classify the extracted data.Finally,the actual data are used for verification,and the results are compared with those of the other three models,which show that the CNN-SVM optimized by Chaotic Adaptive Gray Wolf has a better convergence effect,and the accuracy is improved by 9.6%compared with that of the traditional CNN,which indicates that the proposed algorithm has a better discriminative accuracy as well as stability.
关键词
接线异常/CNN/SVM/混沌自适应灰狼算法Key words
wiring abnormality/CNN/SVM/chaotic adaptive gray wolf algorithm引用本文复制引用
基金项目
江苏省科技支撑计划(工业)项目-重点项目(BE2020006-5)
出版年
2024
NETL
NSTL
万方数据