PDF(1999 KB)
基于Wave⁃ViT的改进多通道深度残差网络的电能质量扰动分类
刘大鹏, 罗嘉宾, 刘勇, 穆勇, 董彪, 张淑清
计量学报 ›› 2025, Vol. 46 ›› Issue (5) : 629-637.
PDF(1999 KB)
PDF(1999 KB)
基于Wave⁃ViT的改进多通道深度残差网络的电能质量扰动分类
Power Quality Disturbance Classification Based on Wave⁃ViT Improved Multi⁃channel Depth Residual Network
提出一种基于小波变换视觉自注意力(Wave-ViT)模型的改进多通道深度残差网络的电能质量扰动分类方法。首先将一维时间序列电能质量扰动(PQDs)信号作为通道一的输入;再将一维PQDs信号通过格拉姆角场(GAF)映射成为二维图像作为通道二的输入;利用Wave-ViT模块深层挖掘二维GAF图像信息,并作为通道三的输入。接着分别对3个通道进行深层次的特征提取,构造适用于PQDs分类的多通道网络框架。通过消融实验,证实多通道对网络收敛速度和分类精度有互补作用。进一步的噪声实验和对比试验表明该方法特征提取能力强,所需迭代次数少,且抗噪性能好,对16种扰动在随机噪声和无噪声环境下的识别率分别能达到99.81%和99.19%,为电能质量扰动识别提供了一种新的思路。
A power quality disturbance classification method based on Wave-ViT improved multi-channel depth residual network is proposed. Firstly, the one-dimensional time series power quality disturbances (PQDS) signal is used as the input of channel one. Then, the one-dimensional PQDS signal is mapped into a two-dimensional image through the Gramian angular field (GAF) as the input of channel two, and the two-dimensional GAF image information is deeply mined by using the wave vision transformer (Wave-ViT) module as the input of channel three, and then the deep feature extraction of the three channels is carried out respectively, and a multi-channel network framework suitable for PQDS classification is constructed. Through ablation experiments, it is confirmed that multi-channel has complementary effect on the convergence speed and classification accuracy of the network. Further noise experiments and comparative experiments show that this method has strong feature extraction ability, less iterations, and good anti noise performance. The recognition rates of 16 kinds of disturbances in random noise and no noise environment can reach 99.81% and 99.19% respectively, which provides a new idea for power quality disturbance recognition.
电磁计量 / 电能质量扰动 / Wave-ViT / 深度残差网络 / 消融实验 / 噪声实验 / 扰动识别
electromagnetic metrology / power quality disturbance / Wave-ViT / depth residual network / ablation experiment / noise experiment / disturbance identification
| 1 |
黄雨涵, 丁涛, 李雨婷, 等. 碳中和背景下能源低碳化技术综述及对新型电力系统发展的启示[J]. 中国电机工程学报, 2021, 41(S1): 28-51.
|
| 2 |
文云峰, 杨游航, 邢鹏翔, 等. 多维因素制约下新能源消纳能力评估方法研究综述[J]. 中国电机工程学报, 2024, 44(1): 127-147.
|
| 3 |
汪飞, 全晓庆, 任林涛. 电能质量扰动检测与识别方法研究综述[J]. 中国电机工程学报, 2021, 41(12): 4104-4121.
|
| 4 |
|
| 5 |
|
| 6 |
刘斌, 曲丽萍, 崔文超, 等. 基于改进HHT+WT复合特征提取的电能质量扰动识别方法[J]. 北华大学学报(自然科学版), 2023, 24(5): 678-686.
|
| 7 |
|
| 8 |
张淑清, 张赟, 刘海涛, 等. 多重分形去趋势波动分析及改进决策树在电能质量分析中的应用[J].计量学报, 2021, 42(4): 424-431.
|
| 9 |
刘海涛, 武祥, 张淑清, 等. 基于优化广义S变换和混合输入神经网络的电能质量扰动识别[J]. 计量学报, 2025, 46(1): 53-61.
|
| 10 |
吴怀诚, 刘家强, 岳蕾, 等. 基于多特征融合的卷积神经网络的电能质量扰动识别方法[J]. 电网与清洁能源, 2023, 39(9): 19-23+31.
|
| 11 |
|
| 12 |
|
| 13 |
赵晨. 基于极限学习机的电能质量复合扰动分类方法研究[D]. 武汉:华中科技大学, 2019.
|
| 14 |
|
| 15 |
肖思丽, 胡洁, 李建宁. 基于小波变换的电能质量检测与分析[J]. 电力与能源, 2018, 39(2): 181-185.
|
| 16 |
|
| 17 |
|
| 18 |
郑炜, 林瑞全, 王俊, 等. 基于GAF与卷积神经网络的电能质量扰动分类[J]. 电力系统保护与控制, 2021, 49(11): 97-104.
|
| 19 |
|
| 20 |
贺才郡, 李开成, 杨王旺, 等. 基于双通道GAF和深度残差网络的电能质量复合扰动识别[J]. 电网技术, 2023, 47(1): 369-379.
|
| 21 |
|
| 22 |
|
| 23 |
季佳伸, 吴俊勇, 王彦博, 等. 基于深度残差网络的电力系统暂态电压稳定评估[J]. 电网技术, 2022, 46(7): 2500-2511.
|
| 24 |
罗溢, 李开成, 肖贤贵, 等. 基于马尔可夫转移场和深度残差网络的电能质量复合扰动多标签分类[J]. 中国电机工程学报, 2024, 44(7): 2519-2531.
|
/
| 〈 |
|
〉 |