Research Information System
Thesis Type: Doctorate
Institution Of The Thesis: Gazi University, Fen Bilimleri Enstitüsü, Turkey
Approval Date: 2023
Thesis Language: Turkish
Student: Görkem GÖK
Principal Supervisor (For Co-Supervisor Theses): Müslüm Cengiz Taplamacıoğlu
Co-Supervisor: Özgül Salor Durna
Open Archive Collection: AVESIS Open Access Collection
Abstract:
This thesis work provides solutions to two different issues using the amplitude and phase distributions of the discrete Fourier transforms of the sampled power system data. First part of the thesis is focused on generating synthetic data to enrich the training set of a deep learning (DL) based system to classify power system transients using the Phasor Measurement Unit (PMU) frequency measurements. It has been shown that the synthetically enhanced training set improves classification performance compared to the case where only the data collected in the field is used for training set. The proposed classification system has helped to extract high-frequency transient information out of PMU measurements, which are collected at a relatively much lower rate, especially when a small training data set is available. Synthetic PMU frequency data has been generated based on Discrete Fourier Transform (DFT) analysis statistics on limited size PMU frequency data. The generation of synthetic data has been achieved by resynthesizing the PMU frequency data using inverse DFT, which separately mimics the DFT frequency and phase behavior for each event type. The deep learning framework then has been trained to classify power system transients using a synthetically enriched trainset. The second focus of the thesis work is to model the behavior of Electric Arc Furnace (EAF) currents depending on tap-to-tap time with a method based on DFT amplitude histograms of EAF current waves. The proposed method models the EAF current behavior separately for each phase of the EAF (boring, melting and refining). The model is validated by comparing the Total Harmonic Distortion (THD) histograms and flicker measurements of the original and modeled EAF current waves. The proposed model can be used as an EAF model in the simulation environment for various purposes prior to the installation of an EAF. It has been shown that this method has a low computational load compared to other techniques, as it uses the amplitude distributions for the first 13 harmonic components while using a one-cycle noise signal representing the higher order harmonics.
Key Words : Power quality, power system events, power system event classification, deep
learning, Gramian Angular Field, synthetic data generation, electric arc
furnace (EAF), EAF modeling