Artificial intelligence helps combat non-technical power losses

A software program that uses an artificial neural network system (artificial intelligence) to identify commercial losses and thus help guide and plan on-site inspection actions is the result of the project developed by Companhia Energética de Pernambuco (CELPE) and researchers from Universidade Federal Fluminense (UFF) and the Federal University of Rio de Janeiro (UFRJ). Initially, clusters (groups) of customers with similar load characteristics, connected to a given processor, were formed. Next, Social Possession and Consumption Habits Survey (PSPH) was conducted, in order to calculate the estimated load curves for each group and to feed the software with information. Therefore, the difference between the values ​​estimated by the tool for each transformer and the effectively measured values results in the commercial loss calculated by the system. With this information, a three-step procedure is set up: treatment of any registration errors, correction of these deviations, and installation of comparative meters in the low-voltage network, where the load curve greatly differs from the corrected estimated load or if it presents oscillations compared to the projected curve. These discrepancies occur, among other reasons, due to errors in the billing process, power theft, or clandestine or irregular connections. During the survey, 30 transformers were measured, resulting in the appointment of 170 consumer units for inspection, with 15 commercial units, 133 residential units, and 32 low-income units. In 57% of cases of the low-income consumer units, the situation was normal and, in the remaining cases, there were issues related to the loss. Among residential customers, normal behavior was observed in 56% of cases. The researchers concluded that the proposed method is useful to indicate transformers that require inspections, reducing the high cost of checking all transformers or installing comparative meters throughout the system.