Narrow your search

Library

KU Leuven (9)


Resource type

dissertation (9)


Language

English (7)

Dutch (2)


Year
From To Submit

2022 (2)

2021 (2)

2020 (3)

2016 (1)

2014 (1)

Listing 1 - 9 of 9
Sort by

Dissertation
Communication-less Protection Algorithms for Meshed VSC HVDC Cable Grids
Authors: --- ---
Year: 2016 Publisher: Leuven KU Leuven. Faculty of Engineering Science

Loading...
Export citation

Choose an application

Bookmark

Abstract

For the large-scale integration of renewable energy sources into the power system, transmission corridors with power ratings and lengths greatly exceeding those in the existing power system will be needed. To realize these corridors, Voltage Source Converter High Voltage Direct Current (VSC HVDC) offers several advantages over the currently widely used ac technology. The use of VSC HVDC in a large-scale meshed grid can provide the major reinforcements to the power system needed for the integration of massive amounts of renewable energy sources.Selective protection against dc side faults is essential to safely and reliably operate meshed HVDC grids. Since required operating times for HVDC grid protection are ten to hundred times faster than existing ac protection, HVDC grid protection algorithms are fundamentally different from those used in ac systems. Furthermore, the limited number of HVDC grid protection algorithms reported in the recent literature were only tested in specific small-scale test systems. For a generally applicable and reliable HVDC grid protection, a more fundamental approach towards the development of protection algorithms is needed.This work provides the necessary concepts to develop communication-less protection algorithms for meshed HVDC grids. A detailed overview of dc fault phenomena is provided and fault clearing strategies proposed in the literature are discussed and classified. The fault current contribution of the half-bridge modular multilevel converter is characterized and a reduced converter model for dc fault studies, is proposed. Guidelines for the design of fault detection methods, based on fundamental traveling wave theory, are provided. Furthermore, signal processing requirements for protection algorithms, in particular required sampling frequency and digital filtering, are investigated. Finally, fast and selective HVDC grid protection algorithms for primary and backup protection are developed. These algorithms are tailored for selective fault clearing in VSC HVDC cable grids with inductive cable termination.

Keywords


Dissertation
Continuous Operation of Meshed HVDC Grids During DC Faults : HVDC Circuit Breaker Integration & Design from a System Perspective
Authors: --- --- ---
Year: 2021 Publisher: Leuven KU Leuven. Faculty of Engineering Science

Loading...
Export citation

Choose an application

Bookmark

Abstract

Developments in Voltage-Source Converter (VSC) technology has enabled the massive integration of renewable energy resources using High Voltage Direct Current (HVDC) technology. Large-scale HVDC grids provide the required flexibility and reliability to future transmission systems with high penetration of intermittent renewable energy sources. However, in case of faults such as DC-side short-circuits, fast rising currents with high steady-state value appear, resulting in a widespread disturbance to the system. To avoid these adverse effects, the DC-side fault current should be cleared in the millisecond time range, which is at least ten times faster than in an AC protection system.HVDC circuit breakers (DCCBs) are expected to be widely installed in large-scale meshed HVDC grids. However, the selection of the most appropriate DCCB parameters (i.e. DCCB design), such as DCCB operating time, line inductor value and interruption capabilities, is complex and challenging for several reasons. First, many DCCB technologies exist with a variety of operating times, but no choice has been made yet on which technology is optimal for large-scale meshed HVDC grids. Second, the DCCB design influences the operation of the grid during DC-side faults and in post-fault conditions. Third, no efficient means are available for calculating DC-side fault currents. At present, there are still no complete methods exist for determining all DCCB parameters.This thesis provides the fundamental concepts, approaches and tools to develop, investigate and evaluate the selective protection scheme using DCCBs for large-scale meshed HVDC grids, considering various DCCB technologies. First, three scenarios are proposed, describing the high-level performance of the HVDC grid during DC-side faults, ranging from the most to the least strict performance. Second, a visual approach is developed characterizing the DCCB parameters. This visual tool provides an insight into the interdependence and trade-offs to be made on the DCCB parameters, thereby allowing for an optimal selection. It is shown that moving from the most to the least strict scenario leads to a relaxation in the DCCB parameter selection. However, relaxing DCCB constraints may lead to more complex operation of converters in post-DC-side fault conditions. Third, the post fault recovery is investigated, by providing an in-depth analysis and insight into the interactions between the converters and the grid during the recovery process, in addition to developing new methods to enhance the post-fault recovery performance. These methods show an effective mitigation of unwanted oscillations during the recovery process. Fourth, necessary approaches are developed to efficiently estimate the DCCB parameters. These approaches help in facilitating the investigation of DCCB requirements for multiple scenarios in future large-scale meshed HVDC grids in an automated way with acceptable computational effort and sufficient accuracy. Fifth, an enhanced active-resonance DCCB topology is proposed to improve the interruption performance of the mechanical interrupter. This allows for an increase in the DCCB breaking capability, a shorter interruption time and consequently reduced DCCB requirements. In summary, this thesis opens a wider view of the selective protection scheme design, which allows for the integration and design of DCCBs with a wide range of operating times.

Keywords


Dissertation
De impact van het proximity effect op de golfpropagatie in ondergrondse kabels

Loading...
Export citation

Choose an application

Bookmark

Abstract

Het doel van deze thesis is het onderzoeken van de impact van het proximity effect op de golfpropagatie in ondergrondse kabels. Door de elektromagnetische transiënten die ontstaan na hoogfrequent gedrag zijn er vele effecten die een rol spelen in de evolutie van de golfsnelheid en de verzwakking in functie van de frequentie. Elke kabelmodellering neemt echter verschillende effecten in rekening, wat leidt tot veranderingen in het golfgedrag tussen de geleidende kabel-delen. In het eerste deel worden ondergrondse kabels besproken en krijgen allerhande effecten die een invloed uitoefenen op de impedantie een plaats. De momentenmethode met oppervlakte operator blijkt een goede methode te zijn om de impact op de golfpropagatie te visualiseren en te analyseren. Nadien bekijkt deze thesis vier cases met stijgende complexiteit: 'Twee Parallelle Geleiders', 'Drie Parallelle Geleiders', 'Eénaderige Volle Kabel' en 'Driefasige Kabel'. Voor elke configuratie wordt de impedantie in functie van de frequentie bepaald en worden er enkele sensitiviteitsanalyses toegepast. De laatste twee opstellingen krijgen een verdere uitwerking om ook de invloed van het proximity effect op de golfpropagatie aan te tonen. Eindigen doet dit verslag met een validatie, een opsomming van enkele uiteenlopende mogelijkheden om dit onderzoek in de toekomst uit te breiden en een toelichting van de bekomen resultaten.

Keywords


Dissertation
Machine Learning to avoid power outages in a renewable dominated system
Authors: --- --- ---
Year: 2020 Publisher: Leuven KU Leuven. Faculteit Ingenieurswetenschappen

Loading...
Export citation

Choose an application

Bookmark

Abstract

The transition towards more renewable energy challenges the current way of securely operating the power system. In particular, the power system must be protected against faults on its lines. To avoid black-outs in a renewable dominated system, faults must be detected and identified correctly about ten times faster compared to the existing way of protection. In this thesis, we implement machine learning techniques in order to evaluate the performance of travelling wave based methods. Those methods rely on higher frequency components of the transient fault signals and therefore can potentially be much faster than traditional methods based on power-frequency components. We resort to machine learning in order to explore this problem in a novel and more extensive way. The aim is to get a better insight into the relevant features at play. Therefore interpretability is an important factor for the selection of the techniques used. In the PART 1, we review the requirements for online and offline protection algorithms and introduce the theoretical background of the travelling waves. We also introduce the different machine learning algorithms that will be used: decision tree, multinomial logistic regression, M5 and multilayer perceptron. In the PART 2, we first introduce how data has been generated, preprocessed and checked. Then, we sequentially review how we implemented the different algorithms for fault type classification, internal/external fault classification and fault localization. Our results and findings are presented for each problem. Some of the novel features of our thesis are the use of Haar wavelet and a comparison of its performance against the more common db4 wavelet, as well as the successful use of M5 algorithm for fault localization. Additionally, we have established a robust process which can be used to test other network configurations and intuitions about possible new relevant features. In the conclusion, we summarize our results and findings, discuss limitations and possible improvements and finally propose possible developments.

Keywords


Dissertation
Physics informed neural networks for efficient power system analysis
Authors: --- --- ---
Year: 2022 Publisher: Leuven KU Leuven. Faculteit Ingenieurswetenschappen

Loading...
Export citation

Choose an application

Bookmark

Abstract

To mitigate climate change, the increase in renewable energy is of vital importance. There exist several types of renewable energy sources, such as solar energy, wind energy and hydropower. The countries that are members of the European Union need to contribute to the growth of such renewable energy sources. That is why Belgium needs to increase its current offshore wind farm capacity. These wind farms will be connected to the existing grid, which can lead to potential overvoltages if the wrong electrical components are chosen. These overvoltages can be damaging to the components of the circuit, and hence they need to be avoided. In this thesis the potential of physics informed neural networks (PINNs) for detecting overvoltages was researched, since they may be more efficient than current simulation techniques like Electromagnetic Transients (EMT) simulations. It is believed that PINNs use less data and generalize better than regular feedforward neural networks. To train the network and determine the performance on unseen examples, data from EMT simulations is provided. The data contains the voltage across a capacitor in an RLC circuit containing a transformer for different transformer switching times and values of the resistor. To start off, a regular neural network was made that predicts only the maximum voltage for different switching times and resistor values, as a baseline model. Next, some experiments were performed to see how neural networks best learn a periodic function. The periodicity that is present in the data can be put into the activation function of the hidden layers using a snake function, or it can be put into the loss function, where the form of the final solution is assumed and the coefficients thereof are determined. It was shown that both methods can be used to learn periodic functions. Having obtained this knowledge, PINNs were constructed that were increasingly complex, with the goal of ultimately being able to create a PINN to solve the problem with the provided data. First, a successful PINN was created that determined the current in an RL circuit for a direct current (DC) voltage source. Next, a PINN was created that does the same, but for an alternating current (AC) voltage source. It was shown that both the snake activation function and a decomposition of the signal could be used to learn a highly oscillatory time-varying periodic function in the framework of PINNs. Following this, a PINN was created that determined the current of an RLC circuit with a DC voltage source. The performance of a regular neural network and PINN were compared. When the components were chosen such that current oscillated with a low frequency, the (physics informed) neural network performed well. However, when a frequency of 50 Hz was used, only the regular neural network performed well.

Keywords


Dissertation
Control Design and Fault Behavior of Modular Multilevel Converters

Loading...
Export citation

Choose an application

Bookmark

Abstract

High voltage direct current (HVDC) technology is set to play a key role in integrating renewable energy sources in the electric power system and increasing interconnection capacity. Connecting HVDC links to an alternating current (AC) power system requires an AC/DC converter. The modular multilevel converter (MMC) is a promising AC/DC converter topology for HVDC. The investment cost for MMCs is mostly determined by the numerous capacitors and semiconductor components in the converter. These components must be able to withstand any overvoltages and overcurrents arising during contingencies. As the current and voltage rating of the MMC components has a large influence on cost, it is desirable to limit overvoltages and overcurrents as much as possible. This work investigates the MMC behavior during AC faults, with a focus on MMC internal stresses. Control of the MMC is complex: the controller consists of many subcontrollers for which multiple possible implementations of exist. The interactions between controllers are highly nonlinear, and seemingly equivalent subcontroller implementations can result in different behavior during faults. To compare these controllers, commonly used implementations in literature are discussed from a theoretical perspective and implemented in Electromagnetic Transient (EMT) simulation software. Fault behavior is then compared through simulation. Recent regulatory developments require HVDC converters to support the grid voltage during faults. This requires additional control loops which further complicates fault behavior. To investigate the impact of voltage support requirements on the component voltage and current rating requirements, control loops from literature were implemented in EMT software and compared through simulation. It was found that voltage support leads to increased voltages across the MMC components during faults, but decreases the current rating requirements.

Keywords


Dissertation
Analysis of DC breaker requirements for different HVDC grid protection methodologies
Authors: --- --- ---
Year: 2014 Publisher: Leuven : KU Leuven. Faculteit Ingenieurswetenschappen

Loading...
Export citation

Choose an application

Bookmark

Abstract

HVDC grids are getting an increased interest for the transmission of large powers over long distances. A possible protection method for HVDC grids is the use of DC breakers to interrupt DC faults. Different DC breaker designs have been developed to protect the grid from DC faults, of which a hybrid breaker (combining a mechanical and power electronic switch) is the most prominent one. The different designs are challenged by the requirement for a high breaking current capacity, energy dissipation capacity and fast commutation time. Current developed DC breakers do not meet all these requirements for high transmitted powers. This thesis provides a solutions to decrease the breaker requirements by examining the influence of different network parameters such as inductors in series with the breaker, IGBT blocking of the converter and switching time on the DC breaker requirements. Secondly, the DC breaker requirements are compared for two different protection schemes which are currently researched for HVDC applications. The first protection method is a full selective DC protection similar to AC protection. Only the breakers at the faulted line trip, hence they must absorb all energy stored in the circuit. A second option, referred to as open grid protection, is by opening each breaker independently in case of a fault, and reclose the breakers when the fault is cleared. Starting from the advantages and drawbacks of these two protection schemes, a new protection method is proposed, which is a hybrid solution between the first and second protection scheme. The new protection scheme combines the advantages of both previous schemes; breaker requirements are reduced and fault clearance becomes more predictable. The potential of reducing the breaker requirements for the different protection schemes is simulated by applying a pole-to-ground DC fault in a four-terminal network with modular multi-level converters. The required energy for the different protection schemes is investigated for different current ratings and switching times. All simulations have been performed using PSCAD/EMTDC software.

Keywords


Dissertation
Machine Learning to avoid power outages in a renewable dominated system

Loading...
Export citation

Choose an application

Bookmark

Abstract

The transition towards more renewable energy challenges the current way of securely operating the power system. In particular, the power system must be protected against faults on its lines. To avoid black-outs in a renewable dominated system, faults must be detected and identified correctly about ten times faster compared to the existing way of protection. In this thesis, we implement machine learning techniques in order to evaluate the performance of travelling wave based methods. Those methods rely on higherfrequency components of the transient fault signals and therefore can potentially be much faster than traditional methods based on power-frequency components. We resort to machine learning in order to explore this problem in a novel and more extensive way. The aim is to get a better insight into the relevant features at play. Therefore interpretability is an important factor for the selection of the techniques used. In the PART 1, we review the requirements for online and offline protection algorithms and introduce the theoretical background of the travelling waves. We also introduce the different machine learning algorithms that will be used: decision tree, multinomial logistic regression, M5 and multilayer perceptron. In the PART 2, we first introduce how data has been generated, preprocessed and checked. Then, we sequentially review how we implemented the different algorithms for fault type classification, internal/external fault classification and fault localization. Our results and findings are presented for each problem. Some of the novel features of our thesis are the use of Haar wavelet and a comparison of its performance against the more common db4 wavelet, as well as the successful use of M5 algorithm for fault localization. Additionally, we have established a robust process which can be used to test other network configurations and intuitions about possible new relevant features. In the conclusion, we summarize our results and findings, discuss limitations and possible improvements and finally propose possible developments.

Keywords


Dissertation
Geschiktheid van de norm IEC 61850 voor DC-vermogensschakelaars in HVDC-netten

Loading...
Export citation

Choose an application

Bookmark

Abstract

Door meer gebruik te maken van hoogspanningsgelijkstroom- (HVDC) verbindingen ontstaat op lange termijn een HVDC-‘supernet’ dat de ruggengraat vormt voor een volledig geïntegreerd net. HVDC-netten laten namelijk een grotere maximale lengte toe in tegenstelling tot wisselspannings- (AC) netten. De uitdaging daarbij bestaat uit het standaardiseren van het HVDC-net. Voor de beveiliging van het HVDC-net zijn DC-vermogensschakelaars essentiële componenten. Ze hebben echter geen natuurlijke nuldoorgangen. Daarbovenop worden de stromen bij een fout in het HVDC-net groter en hebben deze een grotere stijgtijd door het gebrek aan inductantie. Dat impliceert dat DC-vermogensschakelaars sneller moeten werken om de fout te onderbreken. Deze thesis onderzoekt de geschiktheid van de internationale communicatiestandaard IEC 61850, die ontwikkeld is voor AC-systemen, voor de hybride DC-vermogensschakelaar. Enerzijds moet gekeken worden of de opbouw van het datamodel van IEC 61850 voldoet voor de hybride DC-vermogensschakelaars en voor zijn bijbehorende functionaliteiten. Anderzijds moet gecontroleerd worden hoe groot de communicatiesnelheid tussen een “Intelligent Electronic Device” (IED) en de hybride DC-vermogensschakelaar mag zijn om de DC-vermogensschakelaar op tijd aan te sturen om de fout op tijd te kunnen onderbreken. Het IED zorgt voor de beveiliging van het HVDC-net door een tripsignaal naar de DC-vermogensschakelaar te sturen bij het optreden van een fout. De maximale toegelaten communicatiesnelheid van 3 ms voor een tripsignaal (IEC 61850-5) kan een probleem worden omdat de totale onderbrekingstijd (detectietijd: 0,300 ms, communicatietijd: maximaal 3 ms en openingstijd van de DC-vermogensschakelaar: minimaal 2 ms) onder 5 ms moet blijven. Uit het onderzoek van het gebruikte datamodel van IEC 61850 blijkt dat geen “Logical Node” (LN) voor de hybride DC-vermogensschakelaar gedefinieerd is in IEC 61850. Afhankelijk van de verschillende functionaliteiten (bv. openen-sluiten, proactief openen) wordt de LN uitgebreid met een aantal instantienummers. Bij de resultaten van de communicatietijden moet rekening gehouden worden met het feit dat in de opstelling van het netwerk slechts één Ethernetswitch aanwezig is. In een echt HVDC-net moet rekening gehouden worden met de gebruikte netwerktopologie en het aantal gebruikte switches, wat kan zorgen voor extra vertragingen van de communicatiesnelheid. Uit de resultaten van de communicatiesnelheid blijkt dat zowel de minimale als de maximale communicatietijd onder de maximaal toegelaten 3 ms van IEC 61850 blijft voor een tripsignaal, namelijk steeds onder 0,750 ms. De beveiliging van het HVDC-net wordt niet gewaarborgd door gebruik te maken van IEC 61850 met het GOOSE-protocol. Daarbovenop is het minimale tijdsinterval van het herhalingsmechanisme te groot om de DC-vermogensschakelaar nog op tijd te onderbreken als de controller van de DC-vermogensschakelaar de commando’s bij de eerste verzending niet ontvangt. Het GOOSE-protocol van IEC 61850 is niet geschikt voor de vereiste snelle communicatie bij DC. Om IEC 61850 geschikt te maken voor hybride DC-vermogensschakelaars in een HVDC-net, moet een nieuwe LN gedefinieerd worden in het datamodel. Daarnaast zal het tripcommando een maximale toelaatbare communicatietijd van 0,750 ms moeten vereisen en het minimale tijdsinterval van het herhalingsmechanisme zal zeker viermaal kleiner moeten zijn.

Keywords

Listing 1 - 9 of 9
Sort by