Deep-Learning-Based Harmonics and Interharmonics Predetection Designed for Compensating Significantly Time-Varying EAF Currents

In this article, a new approach to compensate both the response and reaction times of active power filters (APF) for special cases of highly time-varying harmonics and interharmonics of electric arc furnace (EAF) currents is proposed. Instead of using the classical approach of taking a window of past current samples and analyzing the data, future samples of EAF currents are predetected using a deep learning (DL)-based method and then analyzed, which provides the opportunity to make real-time analysis. This can also serve the needs of other possible APF applications. Two different methods for prediction of future samples of harmonics and interharmonics have been proposed: predetection of harmonics and interharmonics in the time domain (TD) and in the frequency domain (FD). To obtain the best possible accuracy for both methods, grid search has been employed for parameter optimization of the DL structure. Both TD and FD approaches have been tested on field data, which had been obtained from transformer substations supplying EAF plants. It is shown that the response time of the APF algorithms can be compensated using the TD-based approach, while it is possible to compensate both the response and reaction times of APFs using the proposed FD-based approach. The developed method can be considered to be a feasible candidate solution for generating reference signals for the controllers of new generation of compensation devices, which can be referred to as predictive APFs.