Spatio-Temporal Processes for Electricity Theft Detection

SCHEME: Industrial Fellowships

CALL: 2016

DOMAIN: IS - Information and Communication Technologies

FIRST NAME: Patrick Oliver

LAST NAME: Glauner



HOST INSTITUTION: University of Luxembourg

KEYWORDS: Anomaly detection,Artificial intelligence,Big Data,Electricity theft detection,Machine learning,Stochastic processes

START: 2016-09-01

END: 2018-11-30


Submitted Abstract

Electricity theft can range up to 40% of the total electricity distributed in some countries. Improving the detection of electricity theft with novel models thrives significant economic value by allowing electricity providers to generate more revenue and to increase the stability of their electricity grid. The novelty of this project is to push the frontiers of modeling electricity theft through stochastic processes. Certain neighborhoods are more likely to steal electricity than others. We therefore derive features about the neighborhood of customers. Since previous inspections focused on certain neighborhoods, we reduce the sampling bias in the training set using spatial point processes for resampling. Next, we model the temporal dynamics of physical inspects that found fraud using Hawkes processes. Features derived from the neighborhood of the customers and these stochastic processes can then be used in machine learning algorithms to better classify customers into fraudulent and normal. Furthermore, the anomaly detection models created in this project will be generalized and applied to other anomaly detection problems in Big Data sets, such as credit card fraud.

This site uses cookies. By continuing to use this site, you agree to the use of cookies for analytics purposes. Find out more in our Privacy Statement