Metodologia para identificação e clusterização espacial de perdas não-técnicas em sistemas de distribuição de energia elétrica

Abstract

Considering the growing demand for electricity in contrast to supply, the price of electricity keeps on raise, despite the relentless pursuit of the regulatory institution for equality, the costs to final consumers may commit the community's interest, specially in one of most complete and complex tariff methodology of the world and in a country with extreme social inequalities such as Brazil. The main source of revenue of a distributor is the sale of electric energy for their captive market. Thus the increasing level of non-technical losses and default caused the sector distribution of electricity has developed increasingly with management of Recovery Revenue's procedures and control. Despite the technological advances that allowed greater capillarity and efficiency in actions to eliminate this kind of market segment (bad credit), the levels of non-technical losses did not stop on the expected volume. The fact is that the proliferation of points that present irregular energy consumption is similar to a viral and cyclic process, since the "fraud industry" always renews knowledge, even with maturation time longer than the progress technological level. Added to this fact and considering that the tools of Smart Grid are not widespread in the Brazilian market yet, it's observed that most companies operate this process by visiting the consumer units in the field, being detecting irregularities - that generate non-technical losses - either running the collection of debts by disconnection. In this context and expecting comply with the civic duty of improving power quality and equality of treatment, it is interesting the operational management of field teams, vii searching for optimize the action's routes through clustering - allowing minimal effort and maximum revenue recovery. This thesis presents a methodology for identification of non-technical losses in distribution systems of electric power, evaluating the application of spatial and clustering techniques - as the rule of "k-nearest neighbors" aligned to methodologies for graphical and visual analysis such as wavelet transform. In this approach, developed in 11 steps, the behavior of the electrical system front to the element's and cluster's consumption of energy is acquired by classical techniques of circuit analysis and compared to the values measured by periodic reading. Simulations with the proposed methodology have resulted in the identification of irregular consumption points, randomly selected at the beginning of the tests, among the most suitable for regulating, being ranked on the first quartile of the data set, on average. During that time it was also found that the most critical point of installation (or higher steady decrease in consumption) resembles that seen in reality and analyzes of clusters identified have indicated that the development of its energy consumption is presented, in most cases, lower than expected for the consumer class and those verified in the nearest logical neighbors, indicating probable irregular consumption.