Non-technical losses in the distribution of electricity cause major financial losses, both to the utilities and their customers, as well as directly affecting the quality of energy reaching regular consumers. Because of the size of the electricity market there is a greater complexity of combating losses, making the resolution of this issue of utmost importance. That said, this work presents the development of an embedded system with the ESP32 microprocessor capable of sending every minute to a database active, reactive power and power factor information of electric consumers. Through the raised consumption pattern of consumers, a database of residential load curves was formed, built exactly for the application of database knowledge discovery (KDD), as a way to improve the fraud and theft identification process. in the electricity distribution network. For this, ten residential daily load curves were collected, divided into three data sets according to their social standard, and the scikit-learn module of Python programming language was used for data mining. To classify the consumer as regular or irregular, a supervised classification task with decision tree algorithms was applied to each data set. Three classifier models were generated which obtained accuracy rates of 86, 90 and 86%, respectively.
|
