基于多特征融合卷积神经网络结合Transformer的电能质量扰动分类方法  

Classification Method for Power Quality Disturbances Based on Multi-feature Fusion Convolutional Neural Networks Combined with Transformer

作　　者：王高峰 张卓石 高蔓 钱云[1] WANG Gaofeng;ZHANG Zhuoshi;GAO Man;QIAN Yun(Electrical and Information Engineering College of Beihua University,Jilin 132021,China)

机构地区：[1]北华大学电气与信息工程学院,吉林吉林132021

出　　处：《北华大学学报（自然科学版）》2025年第1期115-124,共10页Journal of Beihua University（Natural Science）

基　　金：吉林省科技发展计划项目(20210203169SF)。

摘　　要：随着可再生能源发电技术的发展,越来越多的可再生能源和设备应用到电力系统中,使电能质量扰动(Power Quality Disturbances,PQDs)发生频率显著增加。PQDs的准确分类对于研究PQDs发生原因和预防至关重要。提出基于多特征融合的卷积神经网络(CNN)结合Transformer模型(CNN-Transformer)对PQDs进行分类。利用快速傅里叶变换(fast Fourier transform,FFT)从PQDs时间序列中提取频域信息,使用CNN-Transformer模型分别对PQDs的时域和频域信息进行特征提取,实现PQDs识别分类。使用该模型对16种合成PQDs数据进行仿真,结果显示:该模型在无噪声条件下的分类准确率为99.88%,在噪声条件下准确率在98.00%以上,且拥有良好的抗噪性和泛化性能。与现有部分分类模型比较显示,本文模型在对比的模型中性能最优。With the development of renewable power generation technology,more and more renewable energy sources and equipment are applied to the power system,resulting in a significant increase in the frequency of power quality disturbances(PQDs).Accurate categorization of PQDs is essential to studying the causes and prevention of PQDs.We propose a convolutional neural network(CNN) based on multi-feature fusion combined with a Transformer model(CNN Transformer) for classifying PQDs.Fast Fourier transform(FFT) is used to extract frequency domain information from PQDs time series,and the CNN-Transformer model is used to extract features from time domain and frequency domain information of PQDs respectively to realize PQDs identification and classification.The model was used to simulate 16 types of synthesized PQDs data,and the results showed that the classification accuracy of this model is 99.88%under noiseless conditions and above 98.00% under noisy conditions,and it has good noise resistance and generalization performance.Comparison with some existing classification models further verifies that the model in this paper has the best performance among the compared models.

关 键 词：电能质量 扰动分类 时频分析 卷积神经网络 多头注意力机制 

分 类 号：TM711[电气工程—电力系统及自动化]