Noisy and non-stationary power quality disturbance classification based on adaptive segmentation empirical wavelet transform and support vector machine

The empirical wavelet transform (EWT) has demonstrated better performance in signal noise removal compared to other threshold techniques based on the conventional wavelet transform (WT), achieved by generating an adaptive filter bank. However, the enhanced EWT (EEWT), the most advanced form of EWT, has limited practical applications since it requires previous knowledge of the spectral components present in the superposed signal. This work introduces a novel adaptive empirical wavelet transform (AEWT) technique designed to segment the signal spectrum and adaptively find out the count of signal components, eliminating the need for manual intervention as required by the traditional EEWT technique. The proposed AEWT technique is particularly well-suited for practical applications, including power quality disturbance (PQD) classification. It is observed that the proposed AEWT technique offers better border detection and denoising performance for PQD signal. In this paper, AEWT is proposed to extract robust features from noisy and non-stationary combined PQD signals, resulting in high accuracy in PQD classification based on these robust features. The proposed AEWT method utilizes a smaller dataset for feature extraction compared to Deep Learning (DL)-based methods, yet achieves higher accuracy than traditional signal processing techniques.