Meter tampering detection through short-lived patterns clustering
Machine learning algorithms applied towards the detection of non-technical losses are increasingly becoming a go-to solution. Tools that help detect losses are being developed by power utilities of all sizes across the globe. Non-technical losses represent the most significant fraction of distributi...
| Autor: | |
|---|---|
| Tipo de recurso: | tesis de maestría |
| Fecha de publicación: | 2021 |
| País: | España |
| Institución: | Universitat Politècnica de Catalunya (UPC) |
| Repositorio: | UPCommons. Portal del coneixement obert de la UPC |
| Idioma: | inglés |
| OAI Identifier: | oai:upcommons.upc.edu:2117/354617 |
| Acceso en línea: | https://hdl.handle.net/2117/354617 |
| Access Level: | acceso abierto |
| Palabra clave: | Electric meters -- Energy consumption -- Software Cluster analysis -- Industrial applications -- Software Comptadors elèctrics -- Consum d'energia -- Programari Anàlisi de conglomerats -- Aplicacions industrials -- Programari Àrees temàtiques de la UPC::Enginyeria electrònica |
| Sumario: | Machine learning algorithms applied towards the detection of non-technical losses are increasingly becoming a go-to solution. Tools that help detect losses are being developed by power utilities of all sizes across the globe. Non-technical losses represent the most significant fraction of distribution losses in electric power systems of all countries, making utilities experience significant revenue losses. These losses are primarily represented by fraudulent activities, which lead to additional losses, damaging the infrastructure of networks and deteriorating the grid’s safety. As lately many Smart Meters have been installed, meter tampering and false data injection are becoming a serious threat. This project, which is part of a larger project called BD4OPEM, aims to propose a preliminary solution to detect Smart Meters tempering through machine learning algorithms. The first part of the thesis gives a theoretical overview of losses in the distribution grid, particularly it provides an extensive review of the non-technical ones, from the causes to the reasons why it is essential to reduce them. Then an explanation about the current use of Big Data and Machine Learning in the field of fraud detection is given. In particular, the different methodologies for non-technical losses detection are reported. The experimental section of the thesis is divided into three parts. In the first one, an algorithm that an algorithm that creates synthetics frauds from real load profiles is implemented. In the second section, three threat models are developed and effectively employed according to a methodology presented in the literature. The clustering algorithm k-means and fuzzy c-means, and the classification algorithm SVM are implemented and tested on different fraudulent data sets. The results are reported, and their performance is evaluated through the use of a confusion matrix. Lastly, a synthetic grid with nine regular Smart Meters and a fraudulent one is created, and the algorithms are tested on it. This example is used to see if the implemented methodology would work in a realistic scenario. In the "Ethical aspects" and "Environmental assessment" the social and sustainable advantages a well functioning, fraud-less grid can provide. Finally, the conclusions are drawn suggesting the scope of future work followed by a brief cost-benefit analysis designed for utilities considering investing in the fraud detection tool development |
|---|