摘要
电能质量扰动(PQDs)可能会引发电力设备故障,导致能源浪费。使用传统的机器学习方法识别多种类型的PQDs往往需要人工设计不同的特征提取器或分类器,耗时费力。而采用深度学习模型能够同时处理多种类型的扰动,具有较强的适应性。文章将深度残差神经网络ResNet18与坐标注意力(CA)模块相结合,构建了一个用于PQDs分类任务的Res-CA模型。首先,通过ResNet18骨干网络提取时间嵌入特征,然后采用CA模块进一步捕获重要性更高的深层时间特征,最后通过Softmax分类器实现PQDs信号类型的识别。在20 dB和30 dB两种信噪比条件下对16种PQDs信号进行实验,实验结果表明:Res-CA网络可以有效地对PQDs信号进行分类,在两种信噪比下的识别准确率分别为99.41%和99.78%。
Power quality perturbations(PQDs)can cause power equipment failures,resulting in wasted energy.Using traditional machine learning methods to identify various types of PQDs,it is often necessary to manually design different feature extractors or classifiers,which is time-consuming and laborious.The deep learning model can deal with multiple types of perturbations at the same time and has strong adaptability.In this paper,a Res-CA model for PQDs classification task is constructed by combining deep residu⁃al neural network ResNet18 with coordinate attention(CA)module.Firstly,the time embedding feature is extracted through ResNet18 backbone network.Then,CA module is used to capture more important deep temporal features.Finally,the PQDs signal type is recognized by Softmax classifier.The experimental results show that the Res-CA network can effectively classify PQDs signals under two SNR conditions of 20 dB and 30 dB,and the recognition accuracy is 99.41%and 99.78%respectively.
出处
《木工机床》
2024年第2期6-12,41,共8页
Woodworking Machinery
关键词
电能质量扰动
注意力机制
深度残差神经网络
power quality perturbations
attention mechanism
deep residual neural networks
