National Technical University of Athens (NTUA)
Αξιοποίηση δεδομένων έξυπνων μετρητών στη διαχείριση δικτύων διανομής ηλεκτρικής ενέργειας
Abstract
dc:descriptionModern power systems, and especially distribution networks, are now facing a set of promiscuous challenges, originating from technological developments and changes in human needs. One of these technological developments is related to the advent of smart meters, which are probably the most important interfaces between humans and the power system. The data acquired by these devices encapsulate a large portion of human behavior itself.Electricity theft is also a result of specific human behaviors taking place inside different social and economic spheres. This doctoral thesis aims in the development of novel concepts and algorithms for detecting electricity theft using smart meter data in order to reduce non-technical losses in distribution networks. A detailed review of non-technical loss (NTL) detection methods is first performed. A new taxonomy is proposed, where electricity theft detection methods are categorized according to the type of algorithm utilized. Additionally, analysis of parameters such as data requirements, extracted features, performance metrics, response time etc. is performed in order to better understand the different issues regarding NTL detection systems. Most NTL detection systems make use of data analytics and machine learning methods as their core operation algorithm. Therefore, part of this thesis is dedicated in the analysis of such methods and the problems rising during their design and implementation. Attention is given to extracting features with strong discriminative capabilities, the use of which will enhance the relative classifiers’ performance. One of these classifiers is the Support Vector Machine (SVM). This thesis provides a detailed description of the utilization of SVM for detecting NTL, but also explores the possibility of using unsupervised methods. Unsupervised methods do not require labeled samples, and thus their application can be easier, compared to supervised methods such as the SVM. An important disadvantage of data oriented methods is that they don’t take into account the physical laws governing power systems. The focus is thus moved to network oriented methods, which make use of voltage and power measurements as well as power system analysis in order to detect electricity theft. The first algorithm developed bases its operation on the calculation of distribution network voltage sensitivities. Voltage sensitivities express the relationship between voltage and power which is disrupted by electricity theft behaviors. The second algorithm utilizes optimization techniques and is inspired by the solution of traditional optimal power flow problems. Furthermore, these algorithms are combined with the SVM previously described, thus forming a hybrid NTL detection system. The aforementioned NTL detection methods exhibit high performance but require smart metering data in order to achieve it. In reality, such data may not be available. This is true for Greece and other countries, who still mostly rely on electromechanical meters for measuring energy consumption. In fact, in the case of Greece, energy is measured every four months, making the application of most NTL detection methods extremely difficult. This thesis tackles exactly this problem, by designing and implementing a rule-based algorithm for detecting electricity frauds. In addition, a software tool is designed which acts as a decision support system aiding the distribution network operator on choosing which consumers to inspect. The final part of this thesis tackles the problem of topology identification in distribution grids, by proposing a new methodology to accurately detect the tree structure and line resistances/reactances of radial networks. The algorithm first estimates the network voltage sensitivity matrices from active power, reactive power and voltage measurements. Smart meters are assumed to be installed only on the leaf nodes, and not on each network bus. The calculated sensitivity matrices are fed to a clustering algorithm, inspired by phylogenetics, which extracts the final network topology. The proposed methodology is assessed over numerous random networks and also by applying it on the optimal battery energy storage placement and sizing problem when network topology is not available.
Degree
thesis:*- Grantor dc:publisher
- National Technical University of Athens (NTUA)
- Year dc:date
- 2020
Author and committee
dc:creator, dc:contributor.*- Authors dc:creator
-
- Messinis, Georgios
- Μεσσήνης, Γεώργιος
Subjects
dc:subject × 18- Ευφυή δίκτυα
- Δίκτυο διανομής ενέργειας
- Έξυπνοι μετρητές
- Μη-τεχνικές απώλειες
- Ρευματοκλοπές
- Ανίχνευση απάτης
- Ανίχνευση τοπολογίας δικτύου
- Smart grids
- Distribution grid
- Smart meters
- Non-technical losses
- Electricity theft
- Fraud detection
- Topology identification
- Επιστήμες Μηχανικού και Τεχνολογία
- Επιστήμη Ηλεκτρολόγου Μηχανικού, Ηλεκτρονικού Μηχανικού, Μηχανικού Η/Υ
- Engineering and Technology
- Electrical Engineering, Electronic Engineering, Information Engineering
Rights
dc:rights- Statement dc:rights
-
- BY_NC_ND
- Language dc:language
- gre
Identifiers
dc:identifier.*- Identifier
- 10.12681/eadd/48727