Nowadays, the use of non-linear loads and power electronics-based equipment in residential, commercial and industrial facilities are contributing to the significant increase of current harmonic distortion levels and, consequently voltage harmonic distortions, as noted in the Brazilian distribution systems. Increasing levels of harmonic distortion in electrical distribution networks is a concern to electric utilities and customers, because the presence of these harmonic sources causes, among others, loss of quality in the energy supply. With a focus on this problem, this thesis proposes the development of non-parametric regression models to identify and quantify what non-linear loads can be considered main sources of voltage harmonic distortion at a point of interest in the electric network. The proposed methodology is based on data correlation analysis, using non-parametric regression statistical models to establish the correlation among the non-linear loads harmonic currents and harmonic voltage at a point of interest. This model is built from harmonic voltage and currents measurements, obtained in measuring campaigns using power quality analyzers installed at the points of interest. In addition, it should be pointed out that these harmonic voltages and currents must be express in base units, rather than percent values in relation to the fundamental component, in order to prevent the influence in the creation of the regression model. An important aspect in this methodology is the use of techniques based on Kernel local polynomial regression, for the estimation of the regression model between harmonic voltage and current. To validate the models it is introduced the determination coefficient R2, which can be obtained from the Pearson correlation coefficient, to measure the accuracy degree of the developed models. The non-parametric regression procedure provides a great flexibility in the estimation of regression models, since it makes possible to carry out a more effective model fit to the data samples, and therefore it is able to characterize the influence of each harmonic source in more detail for the entire measurement period. This technique presented more reliable results and overcomes the shortcomings of the linear regression model, which requires the harmonic currents of other sources, called background, not to vary when analyzing a particular load current. The linear and non-parametric regression models were simulated using the program R, which is a language and environment for statistical calculations and graphs, and as test system it was used the Federal University of Pará electric distribution network, consisting of 84 (84) load busbars in 13.8 kV. The results so obtained are compared to those obtained with the linear regression models, and presented good performance, allowing its application for electric power distribution companies.
|
