基于HWST时频分析和DNN的电能质量监测框架

A Power Quality Monitoring Framework Based on HWST Time-Frequency Analysis and DNNs

随着电网的快速发展,电气设备的使用大量增加,电能质量干扰监测对于电网的可靠和平稳运行变得十分重要。为此,提出一种基于堆叠自编码器(stackedautoencoder,SAE)的深度神经网络(deep neural network,DNN)模型,用于提取单一和组合电能质量干扰的时频谱特征。首先,通过采用随机搜索优化技术调整超参数,利用超曲线s变换(hyperbolic window stockwell transform,HWST)分析电能质量信号的时频特征。然后将HWST时频矩阵输入到3层SAE网络,自动学习50维深度特征。最后,将提取的深度特征输入到多种机器学习分类器进行识别。实验结果表明,采用XGBoost分类器可以以99.86%的准确率识别18种单一和组合电能质量事件。该框架在噪声环境和频率变化条件下也表现出良好的鲁棒性,并成功应用于实际电网数据。

With the rapid development of power grids and the increased use of electrical equipment,monitoring power quality disturbances has become critical for reliable and stable grid operation.A deep neural network(DNN)model based on stacked autoencoders(SAE)is proposed to extract timefrequency features of both single and combined power quality disturbances.First,the hyperparameters are tuned using a random search optimization technique,and the hyperbolic windowed Stockwell transform(HwST)is employed to analyze the time-frequency characteristics of the power quality signals.Then,the HWST time-frequency matrix is input into a 3-layer SAE network to automatically learn 50-dimensional deep features.Finally,the extracted deep features are fed into various machine learning classifiers for identification.Experimental results show that the XGBoost classifier achieves a recognition accuracy of 99.86%for 18 types of single and combined power quality events.The framework also demonstrates robustness in noisy environments and under frequency variations,and has been successfully applied to real power grid data.