SOURCE:

Faculty: Engineering
Department: Electronic And Computer Engineering

CONTRIBUTORS:

Aniedu, A. N.
Inyiama, H. C.

ABSTRACT:

Non-technical loss (NTL) defined as any consumed energy or service which is not
billed because of measurement equipment failure or ill-intentioned and
fraudulent manipulation of said equipment and which therefore results in
inconsistencies in consumption profile of the consumers is a major problem for
electricity utility companies. This dissertation presents an enhanced electricity
consumption monitoring algorithm for non-technical loss detection in Advanced
Metering Infrastructure (AMI) based on the analysis of consumers’ consumption
pattern leveraging Machine learning (ML) techniques. Support Vector Machines
(SVM) was selected, modelled, trained and applied towards classifying
consumer’s electric energy usage readings (after filtering and formatting), and
also for performing predictive analysis for the dataset after a careful survey of a
number of machine learning classifiers and a methodical selection of the two
main SVM parameters (ie Cost parameter, C, and kernel function, gamma). A
novel pre-classifier was designed and developed which resulted in better
prediction outcome with the SVM classifier. Classification accuracy (and
subsequently, class prediction) of 99.2% was achieved with the developed preclassifier
as against 79.46% obtained without pre-classification (although there
is a trade-off with processing time when the pre-classification time is taken into
consideration). This implies that utility workers can predict the occurrence of
NTL in electricity usage with about 99.2% accuracy using the developed model.
This will enable them take intelligent decisions and promptly disconnect any
fraudulent user remotely, using facilities embedded in AMI and smart meters. It
was also observed from analysis of the energy usage dataset that there is
normally a heightened use of electric energy during the winter period (especially
among residential consumers) in contrast to other seasons of the year, which is
a critical information for balancing the load in energy distribution. It has been
shown, through this research, that fraud detection in electricity consumption,
and hence a solution to non-technical losses can be achieved using the right
combinations of Machine Learning techniques in conjunction with AMI
technology.