针对电能质量扰动(power quality disturbance,PQD)分类问题,文中使用一维卷积神经网络(convolutional neural network, CNN)分别构建得到时域(原始信号)、频域(傅里叶变换)、时频域(小波变换)三个CNN 分类模型,建立基于BP(Back Propagation)神经网络的分类结果融合模型实现PQD 分类。仿真实验及对比分析证明,该分类方法在较低运算量的前提下,实现了较高的准确率和良好抗噪性—信噪比不低于30 dB时分类准确率保持在99.7 %以上,在20 dB、15 dB 及10 dB 信噪比时依然可以保持99.58%、99.33 %和98.91 %的准确率,具有一定的实际应用价值,为基于融合方法的PQD 分类问题深入研究提供了参考。
英文摘要:
Aiming at the problem of power quality disturbance (PQD) classification, classification sub-models based on one-dimension CNN in the time domain (original signal), frequency domain (Fourier transform) and time-frequency domain (wavelet transform) are modified. And then, a fusion model of classification result based onback propagation (BP) neural network is constructed to realize PQD classification. By simulation experiment and comparative analysis, the fusion method is proved to have high accuracy and good robustness under the premise of low computational complexity. The classification accuracy is higher than 99. 7% with SNR not less than 30 dB, while the accuracy of 99. 58% , 99. 33% and 98. 91% can still be maintained at 20 dB, 15 dB and 10 dB SNR. The method is proved valuable in practical application, and provides a reference for the further study of fusion method-based PQD classification.