Time Detection and Recognition of Power Quality Events Based on WED-EMD and LSTM-FCNN

Aiming at the problem of power quality time detection and identification in the new power system, power quality time detection method based on Wavelet Entropy Denoising-Empirical Mode Decomposition (WED-EMD) and identification method based on Long Short Term Memory-Fully Connected Neural Network (LSTM-FCNN) are proposed. To solve the unbalanced nature and high-frequency noise interference of power quality signals, multiple layers of high-frequency and low-frequency components of the power quality signal are obtained by wavelet decomposition, and the high-frequency components are filtered using wavelet entropy method to realize signal denoising. Afterwards, the signal is decomposed into multiple Intrinsic Mode Function (IMF) components by EMD. The modulus maximum detection method is used to find the extreme point of IMF component to realize the time detection of power quality. On this basis, the decomposed IMF component is used as the feature vector to train the established LSTM-FCNN model, so as to realize power quality recognition. Finally, the case study shows that the wavelet entropy method can effectively remove the Gaussian white noise of power quality signal, and the IMF component can accurately detect the start and end time of power quality problems. The established neural network can accurately identify various power quality issues, and the average recognition accuracy is 99.2%.