Utilização de técnicas de similaridade dinâmica para detecção de novidades em sinais de sistemas elétricos de potência

Abstract

With the technological evolution of Electric Power Systems (EPS), making them more intelligent and robust, the subject of Power Quality (PQ) has become extremely important. Non-linear loads are the main responsible for inserting electrical disturbances in power systems, consequently, it becomes indispensable to search for techniques that detect these disturbances, known as novelty detectors. This dissertation presents the use of three dynamic similarity techniques: Dynamic Time Warping (DTW), Edit Distance on Real signal (EDR) and, Time Warp Edit Distance (TWED) for detecting electrical disturbances (novelties) in PQ signals. These techniques are mathematical tools for comparing two signals and generating a measure of similarity. Novelty detection occurs by comparing a reference frame (signal window) with the frame under analysis. If the distance between the two frames exceeds a threshold value, a novelty is detected. To evaluate the performance of the methods used, statistical tests were performed and some metrics were evaluated from the ROC curve (Receiver Operating Characteristic), and the obtaining the value of the area under the curve (AUC - Area Under Curve). The performance obtained by the proposed techniques was also compared with two other techniques, Energy Difference Between Frames (DEF) and Ruzicka, present in the literature. The tests were performed in a Matlab® environment. The three similarity techniques showed good performance for novelty detection, especially DTW, which besides the simplicity of the algorithm, presented AUC values close to one (ideal), for signals corrupted by noise, one with SNR = 60dB, and one with SNR = 30dB, during novelty detection. Therefore, the applicability of dynamic similarity techniques for detecting disturbances in of disturbances in SEP signals.