Supervised machine learning-assisted current transformer waveform reconstruction for fundamental fault current phasor estimation

Deteriorated measurements due to the saturation of current transformers (CTs) are a major challenge in digital power system protection schemes. Failing to resolve this issue effectively can have serious consequences for the accuracy of measured components in phasor-based digital relaying algorithms. Potentially, this may compromise the secure and reliable operation of the protection system and, hence, the entire power system. Accurate fault detection or classification can be achieved by applying an algorithm that can deal with the CT saturation effects. In this paper, a method is presented that can accurately estimate the fundamental current phasor during CT saturation. Reconstruction of the deteriorated measured waveform is accomplished by applying a supervised machine learning algorithm, namely the support vector machine (SVM). The least squares (LS) method is integrated with the SVM-based algorithm to reduce the complexities of waveform reconstruction regressions. A modified discrete Fourier transform (DFT), robust to decaying DC components, is then applied to the reconstructed waveform to extract the required phasor components. The proposed approach is validated by evaluating its classification and phasor estimation performance using standard metrics over numerous simulated cases and field measurements. The results demonstrate the high accuracy of the proposed method to classify different levels of CT saturation and ensure precise estimation of the fundamental phasor component. • Saturated CT signals are reconstructed using supervised time-series prediction. • A combined least squares–SVM model performs fast waveform reconstruction. • An online recursive LSSVM approach overcomes batch-learning limitations. • Modified DFT removes decaying DC and improves phasor estimation accuracy. • Accurate phasor estimation is achieved within one cycle for all CT saturation levels.