Institutional Repository
Technical University of Crete
Technical University of Crete
EN
|
EL
| URI: | http://purl.tuc.gr/dl/dias/E31AA7B7-5167-437F-A74B-6E58A1546C00 | ||
| Year | 2019 | ||
| Type of Item | Diploma Work | ||
| License |
|
||
| Bibliographic Citation | Georgios Titakis, "Implementation of power theft detection algorithms based on neural networks and fuzzy logic techniques", Diploma Work, School of Electrical and Computer Engineering, Technical University of Crete, Chania, Greece, 2019 https://doi.org/10.26233/heallink.tuc.83842 | ||
| Appears in Collections |
The main issue for which we have written this thesis is non-technical losses detection in a smart electrical system. With the implementation of data mining techniques for analyzing the electricity consumption patterns of consumers, we identify the illegal consumers based on irregularities in their consumption data.Distribution of electricity involves significant Technical Losses (TL) as well as Non-Technical Losses (NTL). Illegal consumption of electricity or electricity theft constitutes a major part of NTL.This thesis presents four different algorithmic approaches which are used to detect abnormalities in the received energy consumption readings.First of all, we gathered the energy consumption data, taken from a real smart grid network in Ireland [50] and prepared them for each method. Applying features, normalization techniques and principal component analysis for the dimensional reduction of smart meter readings, we managed to adjust the energy consumption data to a better form to work with. Then we tested the five algorithmic methods in order to classify the consumers to legal or illegal.The tools that used in this thesis in order to create neural networks for the classification of illegal and legal consumers, were Fast Artificial Neural Network Library (FANN) and Deep Learning Toolbox (or Neural Network Toolbox). For comparison, we implemented three more methods, which were not based on neural networks. The first used the LibOPF library for the design of optimum-pathforest classifier, the second Support Vector Machines (SVM) and the third Neuro-Fuzzy Inference Systems.Finally, we discussed the results of the above electricity theft detection techniques we experimented with. We observed that almost all the algorithmic methods which were used, gave us very satisfactory results for the detection of irregularities in energy consumption data.
Copyright © DIAS 2013
