PQ classification by way of parallel computing - A sensitivity analysis for a real-time LSTM approach using waveform and RMS data

This paper shows a sensitivity analysis for a long short-term memory approach for the classification of power quality issues. In this investigation, both, high-resolution data and root mean square values are used. These data are preprocessed with the discrete wavelet transform and the discrete short-time Fourier transform. From the results of the transformations, statistical features are extracted, which form the basis of the training of the long short-term memory network. This network is capable of classifying data from different measuring points and phases in real time. The system is able to differentiate between 14 different power quality classes. The result of this paper is a high classification accuracy for the synthetic data test set, as well as a presentation of a comprehensive sensitivity analysis on the impact factors on the classification result, based on a data stream with different feature extraction methods and the long short-term memory network classification.