Electricity Consumption Forecasting and Theft Detection: Challenges and AI-based Solutions

Abstract

Forecasting the electricity consumption is a very current and at the same time extremely difficult challenge. It is current because there is still no efficient way to store the produced electricity, so the produced amount must not exceed its consumption, in order to avoid the expensive overloading of production plants and the distribution network. It is difficult because although consumption has repetitive seasonal dynamics, it also follows irregular trends, and is subject to random unpredictable variations. An example of this type of variation is the non-technical loss of electricity, i.e. errors in meter reading, errors in accounting, broken or faulty infrastructure and electricity theft. From these, electricity theft is of the greatest interest. Hence, the challenge of accurate prediction of electrical energy consumption is compounded by the challenge of fast detection of these non-technical losses. The response to the mentioned challenges will contribute to a stable power energy system that will deliver high quality electricity. It will have a positive impact on the economy in the country. Most importantly, it will contribute to the preservation of the environment because it will enable optimization of electricity production, reducing the load on production plants and the distribution network.