Detecção de outliers em dados de medição fasorial baseado em tipicidade e excentricidade

Abstract

Currently, phasor measurement units (PMU) are essential equipment for real-time operation of the electrical system. However, due to complex factors, these data can easily be compromised by interference, synchronization failure or even a failure of some equipment. This will raise several levels of data quality problems, potentially affecting applications that are based on data quality or at the same time threatening to secure two power systems. This work aims to detect these anomalies based only on the data of the PMU, without the need for knowledge of electrical parameters of the network. For this purpose, an online algorithm was proposed based on the eccentricity and typicality of the sample in relation to its set of previous samples, or TEDA. It was proposed a modification in TEDA, or TEDA Janelado, where its main objective was to deixate the method more sensitive to variations in media and variation over time, maximizing the eccentricity of a sample and favoring the detection of outliers more difficult . A third strategy to improve the detection of TEDA was also presented in the work, or sketch factor, whose role is to establish a more fair weight between the new samples and the average/variance of its set of data, this method was called TEDA Sketch. Finally, these three methods were validated according to performance metrics used in the literature. Two test scenarios were used considering given phasor measurement areas, being used at frequency, phase voltage and current, totaling 36000 samples for each electrical magnitude. For all the scenarios analyzed, the algorithms used have a satisfactory performance, with emphasis on the TEDA Janelado and the TEDA Esquecimento that obtain the best results in the most challenging scenarios. It has also been certified to be capable of two methods to detect abnormalities during electrical disturbances. Finally, it is hoped that this research will contribute to this topic, presenting a new approach for detecting outliers in PMU data and contributing to better quality of data, guaranteeing that its applications in energy systems are used more safely.