基于卷积神经网络的电流互感器畸变信号识别方法  

A Distortion Signal Recognition Method for Current Transformer Based on Convolutional Neural Network

作　　者：徐敏锐 卢树峰 李志新 欧阳曾恺 陈刚 XU Minrui;LU Shufeng;LI Zhixin;OUYANG Zengkai;CHEN Gang(State Grid Jiangsu Electric Power Co.,Ltd.Marketing Service Center,NanJing 210019,China)

机构地区：[1]国网江苏省电力有限公司营销服务中心,南京210019

出　　处：《自动化与仪器仪表》2023年第10期288-291,共4页Automation & Instrumentation

基　　金：国网江苏省电力有限公司科技项目:宽量程计量用电流互感器及其校验装置研发及应用(2021209)。

摘　　要：在识别电流互感器畸变信号时,主要通过基础的神经网络提取信号特征,只能得到低层特征,使得畸变信号识别结果而F1分数较低。因此,应用卷积神经网络,设计一种新型电流互感器畸变信号识别方法。明确电流互感器的工作原理,绘制整体高频等效电路图,并基于此建立畸变信号模型。应用软阈值去噪原理,对采集的电流互感器信号进行去噪处理。再依托于多通道卷积神经网络,设计信号特征提取模型结构,将去噪后的信号输入其中进行深度学习,组合每个通道输出的低层特征,输出更加抽象的高层信号特征。最后针对特征提取进一步计算,构建特征空间,以此来实现畸变信号的准确识别。实验结果表明:所提方法识别结果的F1分数保持在0.97以上,展现出极好的信号识别效果。When identifying current transformer distorted signals,the basic neural network is mainly used to extract signal features,which can only obtain low-level features,resulting in a low F1 score for distorted signal recognition.Therefore,using convolutional neural networks,a novel method for identifying current transformer distortion signals is designed.Define the working principle of the current transformer,draw the overall high-frequency equivalent circuit diagram,and establish a distortion signal model based on this.The principle of soft threshold denoising is applied to denoise the collected current transformer signal.Relying on multi-channel convolutional neural networks,a signal feature extraction model structure is designed.The denoised signal is inputted into it for deep learning,and the low-level features outputted from each channel are combined to output more abstract high-level signal features.Finally,further calculation is conducted for feature extraction to construct a feature space to achieve accurate recognition of distorted signals.Experimental results show that the F1 score of the proposed method’s recognition results remains above 0.97,exhibiting excellent signal recognition effects.

关 键 词：卷积神经网络 电流互感器 畸变信号 特征提取 多通道 识别方法 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]