Statistical Models of EAF Harmonics Developed for Harmonic Estimation Directly from Waveform Samples Using Deep Learning Framework

In this paper, a method to generate large amounts of Electric Arc Furnace (EAF) currents with harmonics simulating the actual EAF operation characteristics to be used with deep learning (DL) applications of harmonic estimation is investigated. For this purpose, the behavior of the EAF current harmonics is examined in statistical terms using the field data collected at a transformer substation supplying an EAF plant. Then, a significantly larger amount of EAF current data is generated using the statistics mimicking the real EAF behavior to train the DL-based harmonic estimator. The outcomes of the research work presented in this paper are two-fold: (i) DL-based method is used to extract phase and amplitude information of the harmonics of the EAF currents using the waveform directly, without computing any time- or frequency-domain features during the estimation process, which helps reduce the processing time, (ii) EAF current data with realistic amounts of time-varying harmonics based on the statistics obtained from a tap-to-tap time of the EAF currents is generated, hence a detailed statistical analysis of the EAF current spectrum is achieved. The method proposed can be used to eliminate the uncharacteristic harmonics of the EAF currents, since it can provide fast and accurate phase and amplitude estimates of the harmonics, serving the need for active power filters in the electricity system.