http://lattes.cnpq.br/6066634529532605; BARROS, Rafael Mendonça Rocha.
Abstract:
A new methodology for improving the management of non-technical loss in electrical distribution power systems is presented in this work. The solution consists of using techniques based on Advanced Analytics to build an automated and adaptive process capable of identifying risk and protection factors for non-technical loss’ occurrence; likelihood of loss existing; an estimate of energy not measured in the system and a potential financial return for on-site inspections of consumers. To perform the research, a database with real information of 261,489 consumers from a Brazilian utility was used. Causal Inference was applied to identify the degree
of association of several features with the occurrence of non-technical loss and, thus, to identify the risk and protection factors. The variables associated with the occurrence of loss were used as input into Machine Learning models in order to identify the probability of loss occurrence, as well as providing an estimate of the value of unmeasured energy in the system. In total, 23 classification models and seven regression models were evaluated with different algorithms and data approaches. The results of the predictive models were used to calculate the potential of financial return from field inspections. Subsequently, a model was proposed to determine the optimal infrastructure for loss management in a utility, which results in the maximization of the
financial return of field inspections. All stages of the methodology were validated through new field inspections carried out on 1,417 consumers. The main results achieved in the work were the identification of 76 risk or protection factors for the occurrence of non-technical loss; the identification of the Rotation Forest algorithm as the most suitable for loss identification, which presented an precision of 66.5% in the field inspections carried out; the identification of the XGBT algorithm as the most suitable for prediction of unmeasured energy values; which showed a deviation of 3.02% in estimate of the total value of energy recovered in a group of
consumers; maximizing the financial return of field inspections, which was able to increase the return on field inspections by up to 11.5 times in the best case. From the achieved results, it is possible to conclude that the new methodology proposed in this work represents an valuable improvement compared to other works in available bibliography, since it provides a solution for the characterization of non-technical loss; identifies a new algorithm with superior performance for classifying consumers; presents an approach to estimate unmeasured energy in systems; provides a strategy to estimate the financial return of on-site inspections and, finally, presents a method for maximizing the return of actions in non-technical loss management. So that all these contributions fill gaps in the existing bibliography. Finally, it should be noted that this work’s results can be used by utilities to improve their non-technical loss management strategies, providing an increase in revenue and a reduction in their operational costs. Which, in turn, will be reflected in a reduction in electricity tariff for the benefit of the whole society.