No contexto de Smart Grids, determinar a correta localização das fontes causadoras de Variação de Tensão de Curta Duração (VTCD) não é uma tarefa simples, devido à curta duração destes eventos e também, por sua rápida propagação nas redes de distribuição de energia elétrica. Neste sentido, esse trabalho apresentou um método híbrido recursivo baseado em ferramentas da área de aprendizado de máquinas (algoritmo de agrupamento e base de regras), o qual é capaz de localizar as fontes de VTCD, a partir da análise dos das características dos distúrbios disponibilizadas pelos smart meters instalados no sistema. Assim, o trabalho destinouse ao desenvolvimento de uma plataforma em hardware para aquisição, detecção e classificação dos distúrbios, através de um Sistema Operacional de Tempo Real. Em seguida o algoritmo de agrupamento (k-means) agrupou os dados dos medidores de forma a definir dois clusters, onde um deles correspondeu aos medidores que estão longe da região que ocorreu o distúrbio e o outro, correspondeu aos medidores que estavam localizados próximos da região de ocorrência do distúrbio. Na segunda etapa, um sistema baseado em regras determinou qual dos clusters abrangeu o nó de origem. No entanto, quando o algoritmo determinou uma região muito grande, essa região é introduzida recursivamente, como entrada da metodologia desenvolvida, para refinar a região de localização. O sistema resultante foi capaz de estimar a região de localização com uma taxa de acerto acima de 90%. Assim, o método teve sua concepção adequada ao empregado nos centros de controle e operações de concessionárias de energia elétrica, visando apoiar a decisão do corpo técnico para que ações corretivas fossem estabelecidas de forma assertiva. / In the Smart Grids context, the correct location of short-duration voltage variations sources is not a trivial task, because of the short duration of these events and for rapid propagation in the distribution feeder. In this sense, aiming to develop a recursive hybrid method based on machine learning area tools (clustering algorithm and rule base) that is able to locate the sources of short-duration voltage variations, it was used data from smart meters installed along the distribution feeder. The recursive hybrid method, as input, received the disturbance characteristics provided by the meters installed in the system. Thus, this thesis aimed to development of a measurement hardware for signal acquisition, detection, classification through a realtime operating system. Then, k-means clustering algorithm grouped the meters data in order to define two clusters, where one of them corresponded to the meters that were distant from the region that occurred the disturbance and the other one corresponded to the meters, which were located near to the disturbance occurrence region. In a second step, a rule-based system determined which of the clusters corresponded to the source node. When the algorithm determined a very large region, that region was recursively introduced as input of the developed methodology to decrease its size. The resulting system was able to estimate the location region with a accuracy above 90%. Therefore, this method showed a suitable design for employment by operation control centers of power sector concessionaires, aiming to support technical staff decision to stablish assertive corrective actions.

No contexto de Smart Grids, determinar a correta localização das fontes causadoras de Variação de Tensão de Curta Duração (VTCD) não é uma tarefa simples, devido à curta duração destes eventos e também, por sua rápida propagação nas redes de distribuição de energia elétrica. Neste sentido, esse trabalho apresentou um método híbrido recursivo baseado em ferramentas da área de aprendizado de máquinas (algoritmo de agrupamento e base de regras), o qual é capaz de localizar as fontes de VTCD, a partir da análise dos das características dos distúrbios disponibilizadas pelos smart meters instalados no sistema. Assim, o trabalho destinouse ao desenvolvimento de uma plataforma em hardware para aquisição, detecção e classificação dos distúrbios, através de um Sistema Operacional de Tempo Real. Em seguida o algoritmo de agrupamento (k-means) agrupou os dados dos medidores de forma a definir dois clusters, onde um deles correspondeu aos medidores que estão longe da região que ocorreu o distúrbio e o outro, correspondeu aos medidores que estavam localizados próximos da região de ocorrência do distúrbio. Na segunda etapa, um sistema baseado em regras determinou qual dos clusters abrangeu o nó de origem. No entanto, quando o algoritmo determinou uma região muito grande, essa região é introduzida recursivamente, como entrada da metodologia desenvolvida, para refinar a região de localização. O sistema resultante foi capaz de estimar a região de localização com uma taxa de acerto acima de 90%. Assim, o método teve sua concepção adequada ao empregado nos centros de controle e operações de concessionárias de energia elétrica, visando apoiar a decisão do corpo técnico para que ações corretivas fossem estabelecidas de forma assertiva. / In the Smart Grids context, the correct location of short-duration voltage variations sources is not a trivial task, because of the short duration of these events and for rapid propagation in the distribution feeder. In this sense, aiming to develop a recursive hybrid method based on machine learning area tools (clustering algorithm and rule base) that is able to locate the sources of short-duration voltage variations, it was used data from smart meters installed along the distribution feeder. The recursive hybrid method, as input, received the disturbance characteristics provided by the meters installed in the system. Thus, this thesis aimed to development of a measurement hardware for signal acquisition, detection, classification through a realtime operating system. Then, k-means clustering algorithm grouped the meters data in order to define two clusters, where one of them corresponded to the meters that were distant from the region that occurred the disturbance and the other one corresponded to the meters, which were located near to the disturbance occurrence region. In a second step, a rule-based system determined which of the clusters corresponded to the source node. When the algorithm determined a very large region, that region was recursively introduced as input of the developed methodology to decrease its size. The resulting system was able to estimate the location region with a accuracy above 90%. Therefore, this method showed a suitable design for employment by operation control centers of power sector concessionaires, aiming to support technical staff decision to stablish assertive corrective actions.

The deployment of large-scale photovoltaic (PV) systems is rising in the Swedish power system, both in quantity and in system size. However, the intermittent characteristics of the PV production raises questions concerning the stability in the electric power grid, and power output fluctuations from the PV systems can lead to voltage quality issues. Hence, the distribution system operator E.ON Energidistribution and the solar energy developer company Solkompaniet are interested in investigating potential challenges and possibilities related to the integration of large-scale PV systems in the electric power grid. This thesis studies fast voltage variations in the electric power grid due to output fluctuations from large-scale PV systems, and examines the possibility to mitigate the voltage variations by reactive power support strategies in the PV inverters. Four studies are carried out to investigate the prerequisites for establishing large-scale PV systems. Firstly, a worst-case study considering eight existing substations in the electric power grid as well as a new substation is carried out, to examine the impact of different parameters on the voltage variations. Parameters such as transformer operation mode, location of the point of connection, switching mode and load capacity are compared in the study. Further, time series calculations are done to investigate the voltage variations over one year, and a study with an oversized PV system is done to investigate the possibility for increasing the PV capacity without grid reinforcements. Lastly, a study is performed with reactive power compensation from the PV inverters to examine the possibility to maintain a stabilized voltage level at the point of connection. The studies are performed in E.ONs network model in the power system simulator software PSS/E, with data for the transmission grid, the regional grid, and parts of the distribution grid included. PV systems with a rated capacity from 32 MWp and upwards are connected to substations in the regional grid, where fast voltage variations on nominal voltage levels of 20/10 kV are studied and evaluated from the perspective of the power producer. From this thesis, it can be concluded that neither of the implemented studies results in voltage variations that violate E.ONs technical requirements on fast voltage variations in the point of connection. Further, the results from the worst-case study show the importance of analysing the specific system of interest when connecting PV systems, since the properties of the existing system have an impact on the voltage variations. The time series calculations show that the voltage variations over a time period of one year are highly influenced by the PV production and the load capacity in the substation, and the study with an oversized PV system shows the possibility for increasing the PV capacity without curtailing large amounts of active power. Finally, the study with reactive power compensation concludes that grid support strategies in the PV inverters may be a key solution for making optimal use of the existing electric power grid and enabling the continued expansion of large-scale PV systems in the Swedish power system. / Utbyggnaden av storskaliga solcellsanläggningar (PV) ökar i det svenska kraftsystemet, både i kvantitet och i systemstorlek. De intermittenta egenskaperna hos energiproduktionen väcker emellertid frågor angående stabiliteten i elnätet, och effektförändringar från anläggningarna kan leda till spänningskvalitetsproblem. Därför är distributionssystemoperatören E.ON Energidistribution och solenergiföretaget Solkompaniet intresserade av att undersöka potentiella utmaningar och möjligheter relaterade till integrationen av storskaliga solcellsanläggningar i elnätet. Detta examensarbete studerar snabba spänningsvariationer i elnätet till följd av effektförändringar från storskaliga solcellsanläggningar, och undersöker möjligheten att mildra spänningsvariationerna genom strategier för reaktiv effektreglering i växelriktare.  Fyra studier genomförs för att undersöka förutsättningarna för att etablera storskaliga solcellsanläggningar. För det första genomförs en värsta-fallstudie med beaktande av åtta befintliga stationer i elnätet samt en ny station, för att undersöka olika parametrars påverkan på spänningsvariationerna. Parametrar som transformatorns driftläge, plats för anslutningspunkten, omkopplingsläge och lastkapacitet jämförs i studien. Vidare görs tidsserieberäkningar för att undersöka spänningsvariationerna över ett år, och en studie med en överdimensionerad solcellsanläggning görs för att undersöka möjligheten att öka solcellskapaciteten utan elnäts- förstärkningar. Slutligen genomförs en studie med reaktiv effektkompensation från växelriktare för att undersöka möjligheten att upprätthålla en stabiliserad spänningsnivå i anslutningspunkten. Studierna utförs i E.ONs nätverksmodell i programvaran PSS/E för kraftsystemsimuleringar, med data för transmissionsnätet, regionnätet och delar av distributionsnätet inkluderat. Solcellsanläggningar med en nominell kapacitet från 32 MWp och uppåt ansluts till stationer i regionnätet, där snabba spänningsvariationer på nominella spänningsnivåer om 20/10 kV studeras och utvärderas ur kraftproducentens perspektiv. Från resultaten kan man dra slutsatsen att ingen av de genomförda studierna resulterar i spänningsvariationer som överskrider E.ONs tekniska krav på snabba spänningsvariationer i anslutningspunkten. Vidare visar resultaten från värsta-fallstudien vikten av att analysera det specifika systemet vid anslutning av solcellsanläggningar, eftersom egenskaperna hos det befintliga systemet har en inverkan på spänningsvarationerna. Tidsserieberäkningarna visar att spänningsvariationerna över en tidsperiod av ett år påverkas starkt av både energiproduktionen och lastkapaciteten i stationen, och studien med en överdimensionerad solcellsanläggning visar på möjligheten att öka den nominella kapaciteten utan att spilla stora mängder aktiv effekt. Slutligen ger studien med reaktiv effektkompensation slutsatser om att strategier i växelriktare kan vara en möjlig lösning för att utnyttja det befintliga elnät optimalt och möjliggöra en fortsatt expansion av storskaliga solcellsanläggningar i det svenska kraftsystemet.