Detección de pérdidas no técnicas en clientes especiales con telemedición, basada en inteligencia artificial con aplicación en la empresa eléctrica Ambato.

Abstract

The EEASA, as a distribution company, stores a large amount of data generated by smart meters placed in special customers; This historical information is analyzed to detect anomalous consumption that is not easily recognized and is a significant part of the non-technical energy losses in the accounting. The execution of adequate clustering techniques with their Machine Learning validation indices and the use of Deep Learning Neural Networks, work hand in hand with the objective of detecting misleading curves that register less than real energy consumption. Given this context, a methodology is developed that consists of classifying energy consumption based on weekdays, weekends and holidays, which present similar behaviors in energy consumption, to then be grouped and finally taken as data sets to the learning, testing and validation of a classifying neural network that is densely connected and being able to identify the daily curves described by these clients; obtaining as a result, the detection of abnormal patterns of energy consumption.