| 基于小波特征提取与深度学习的微电网故障诊断与分类方法 |
| |
| 引用本文： | 姚欣,邢砾云,辛平. 基于小波特征提取与深度学习的微电网故障诊断与分类方法[J]. 陕西电力, 2021, 0(12): 17-24 |
| |
| 作者姓名： | 姚欣 邢砾云 辛平 |
| |
| 作者单位： | （北华大学电气与信息工程学院，吉林吉林 132000） |
| |
| 摘 要： | 针对现有微电网（MG）故障诊断准确率不高，分类精度不理想等问题，提出了一种基于小波特征提取与深度学习的微电网故障诊断与分类方法。首先，采用最大重叠离散小波变换（MODWT）和母小波提取MG电力信号特征，并进行三级分解，以获得高精度的信号特征提取。然后，利用长短期记忆网络优化深度Q网络，构建深度循环Q网络（DRQN），更好地分析复杂数据且克服噪声的干扰。最后，将MODWT每个分解层次上的信号分量能量输入DRQN，实现故障的诊断和分类。基于MATLAB环境搭建MG系统仿真模型对所提方法进行实验论证，结果表明使用高采样频率和电流、电压信号时，诊断性能最佳，分类准确率超过91%。同时，所提方法在11种故障类型和4种场景下的分类准确率均超过90%，优于其他对比方法。
|
| 关 键 词： | 微电网 故障诊断 故障分类 最大重叠离散小波变换 深度循环Q网络 长短期记忆网络 特征提取 |
Fault Diagnosis and Classification of Microgrid Based on Wavelet Feature Extraction and Deep Learning |
| |
| YAO Xin,XING Liyun,XIN Ping. Fault Diagnosis and Classification of Microgrid Based on Wavelet Feature Extraction and Deep Learning[J]. Shanxi Electric Power, 2021, 0(12): 17-24 |
| |
| Authors: | YAO Xin XING Liyun XIN Ping |
| |
| Affiliation: | （School of Electrical and Information Engineering， Beihua University， Jilin 132000， China） |
| |
| Abstract: | Targeting the problem of low accuracy and poor classification accuracy of fault diagnosis in existing microgrid （MG）， this paper proposes a method of the fault diagnosis and classification of the microgrid based on wavelet feature extraction and deep learning. Firstly， maximum overlap discrete wavelet transform （MODWT） and mother wavelet are used to extract the features of power signal from MG， and three-level decomposition is carried out to obtain high-precision signal feature extraction. Then deep Q-network is optimized by long-term and short-term memory network， and deep recurrent Q-network （DRQN） is constructed to better analyze complex data and overcome the interference of noise. Finally， the energy of signal components at each decomposition level of MODWT is input into DRQN to realize the fault diagnosis and classification. The simulation model of MG system based on MATLAB is built to demonstrate the proposed method. The results show that the diagnosis performance of the method is the best when using high sampling frequency and current and voltage signals， and the classification accuracy is more than 91%. At the same time， the classification accuracy of the proposed method considering eleven fault types and four scenarios is more than 90%， which is better than other comparison methods. |
| |
| Keywords: | microgrid fault diagnosis fault classification MODWT DRQN long-term and short-term memory network feature extraction |
|
| 点击此处可从《陕西电力》浏览原始摘要信息 |
|
点击此处可从《陕西电力》下载免费的PDF全文 |
