Classification of electricity consumers using artificial neural networks

Abstract

This paper explains the process of using neural networks, as one of numerous data mining techniques, for the classification of electricity consumers. The processed data comprised more than a million recordings of electricity consumption for 21,643 consumers over the period of four years and eight months. Using a data subset (70% of the entire dataset), the network was trained for the classification of consumers according to the type of the electric meter they possess (single-rate or dual-rate) and the zone they live in (city or village). The network input data in both cases included: consumer code, reading period from-to, current and previous meter reading for both low and high tariff, dual and single rate tariff consumption for that period and their total amount, as independent variables, whereas the network output comprised dependent variable classes (zone or type of electric meter). The results show that a network created in this way can be trained so well that it achieves high precision when evaluated using the test dataset. Using the available recordings about electricity consumption, the type of the electric meter consumers possess and the zone they live in can be predicted with the accuracy of 77% and 82%, respectively. These findings can provide the basis for further research using other data mining techniques.