首页|期刊导航|电气技术|融合高斯-拉普拉斯金字塔多尺度格拉姆角场表征与双路径网络的电能质量复合扰动分类

融合高斯-拉普拉斯金字塔多尺度格拉姆角场表征与双路径网络的电能质量复合扰动分类OA

Power quality composite disturbance classification based on Gaussian-Laplacian pyramid multi-scale Gramian angular field representation and a dual-path network

中文摘要英文摘要

针对强噪声和复合扰动条件下电能质量扰动识别精度下降的问题,本文面向 23 类扰动信号与正常信号构成的 24 类分类任务,提出一种融合高斯-拉普拉斯(G-L)金字塔多尺度格拉姆角场(GAF)表征与双路径网络的识别方法.首先,将一维扰动波形编码为 GAF 图像,并构建高斯-拉普拉斯混合金字塔,以联合表征全局幅值趋势与局部突变纹理.随后,采用 EfficientNetV2-B0与 ResNet18-small 组成双路径网络,分别提取图像特征和时域特征,并通过门控机制实现融合,同时结合渐进式解冻训练策略和类别结构正则提升训练稳定性.实验结果表明,该方法在 10~30 dB 信噪比条件下的平均分类准确率超过 99%,在硬件实测数据上的综合准确率为 97.2%.所提方法能够有效提升复杂电能质量扰动场景下的识别精度、噪声鲁棒性和泛化能力.

To address the decline in recognition accuracy of power quality disturbances under strong noise and composite disturbance conditions,this paper proposes a recognition method that integrates a multi-scale Gramian angular field(GAF)representation based on a Gaussian-Laplacian(G-L)pyramid with a dual-path network for a 24-class classification task consisting of 23 disturbance types and one normal class.First,one-dimensional disturbance waveforms are encoded into GAF images,and a hybrid Gaussian-Laplacian(G-L)pyramid is constructed to jointly characterize global amplitude trends and local transient textures.Then,a dual-path network composed of EfficientNetV2-B0 and ResNet18-small is employed to extract image-domain features and time-domain features,respectively,which are fused through a gating mechanism.In addition,a progressive unfreezing training strategy and class-structure regularization are introduced to improve training stability.Experimental results show that the proposed method achieves an average classification accuracy of over 99%under signal-to-noise ratios from 10 to 30 dB,and attains an accuracy of 97.2%on hardware-measured data.These results demonstrate that the proposed method can effectively improve recognition accuracy,noise robustness,and generalization performance in complex power quality disturbance scenarios.

钟文轩;王文波;何陈程

武汉科技大学数学与系统科学学院,武汉 430070武汉科技大学数学与系统科学学院,武汉 430070武汉科技大学数学与系统科学学院,武汉 430070

电能质量扰动格拉姆角场(GAF)高斯-拉普拉斯(G-L)金字塔双路径网络门控融合

power quality disturbanceGramian angular field(GAF)Gaussian-Laplacian(G-L)pyramiddual-path networkgated fusion

《电气技术》 2026 (10)

1-11,20,12

冶金工业过程系统科学湖北省重点实验室开放基金(No.Z202302)

评论