Equalização não-linear supervisionada de transformadores de medição para aplicações em qualidade de energia

Abstract

Channel Equalization encompasses a set of advanced signal processing techniques that are essential in systems where the propagation medium interferes with the information transmitted through the channel. Its application aims to mitigate the adverse effects caused by distortions in communication channels, such as attenuation, interference, dispersion, and nonlinearities. Equalization is crucial for Power Quality, considering that the channel does not exhibit a constant response in the frequency domain and therefore distorts the harmonic components and, consequently, the transmitted signal itself. In medium and high voltage networks of the power system, instrumentation transformers play a vital role. They reduce the magnitude of voltage and current signals to levels compatible with measuring equipment. These transformers, which are costly and designed for long service life, are present in all power substations and are designed to have good accuracy for measuring the fundamental component. Therefore, they are not suitable for measuring harmonic components and can present errors in magnitude and phase greater than 100%.One possibility to utilize these transformers for measuring harmonic components or acquiring waveforms with reduced distortion is to use adaptive equalization. This procedure, which is rarely explored in the literature for improving instrumentation transformers (TI), is investigated in this dissertation. Although there are a few works on linear adaptive equalization, this study presents an approach using nonlinear adaptive equalization. Thus, this work presents the study of nonlinear techniques to make the equalization of instrumentation transformers more robust. The techniques investigated include the Volterra Series, Bilinear IIR network in the difference equation form, and neural networks. The motivation for using nonlinear equalization is not to equalize the nonlinear characteristics of the channels, as the channels used in this work are still linear, but rather to overcome some restrictions of linear equalizers, such as the need for a minimum phase channel. It is also expected that using nonlinear equalization will improve the frequency response of the equalized channel. To quantify the stability of the algorithm, the Mean Squared Error (MSE) is observed, while the accuracy of the equalizers is measured by the Total Vector Error (TVE) metric. In the studied scenarios, the Bilinear technique presented the best results on average, keeping the TVE below 3% in practical tests. All algorithms used a supervised process, thus requiring a reference signal for the equalization process.