The operational challenges of the integration of electric vehicles (EV), air conditioning and photovoltaic panels (PV) are prompting the upgrade of distribution grids, seen here as cyber-physical infrastructures. An important upgrading feature of the cyber-side is the electrical grid monitoring, which needs to expand both in data coverage and granularity. The challenge is to decide the data strategy, or in other words, which level of granularity is actually needed in active distribution networks. This work proposes a framework to assist grid planners in selecting the level of data expansion needed, by quantifying the impact of extended data granularity on control capabilities, and corresponding grid performance. The framework combines machine learning with AC optimal power flow and state estimation to select incremental upgrades of the cyber-physical infrastructure. Grid planning and operation are simulated and tested for the IEEE 33-bus test system over a 5-year span to assess the role of granularity in grid performance for different cyber-infrastructures. The results show that extending data granularity is critical for mitigating voltage violations under high penetration of EVs, air conditioning and PVs. By modeling the relationships between data, grid planning and operation, and grid performance, the framework supports efficient cyber system upgrades to mitigate operational violations while accounting for budget limitations.

The current power system is a network of electrical components forming a physical system. It is experiencing changes, such as the deployment of electric vehicles and distributed energy sources. Meanwhile, cybernetworks are becoming coupled into the physical grid to an increasing degree. This transformation of electrical grids into smart grids is often the focus of research literature. It is strongly linked to the concurrent evolution of the cyberinfrastructure, which is the focus of this thesis. It aims to support the incremental upgrade of data systems, taking real-world constraints into account, as well as opportunities offered by sensors and machine learning. With this background, two research questions have been identified: i) How to use available data more efficiently to improve asset management? and ii) How much and which kind of new data are actually needed?

With regard to the first research question, ways to use available data more efficiently are investigated in collaboration with a distribution system operator (DSO). One option is for DSOs to adopt best practices in terms of data management, which include: de-silo data, enhance reporting practices, and automatize tasks. To illustrate how combining available data may deliver additional relevant information, criticality indices have been calculated and assigned to components of a substation, by combining outage data, operation data and network diagram. Another option is to develop machine learning algorithms to perform specific or new tasks. A failure warning system has been developed using machine learning to leverage existing data about power components. It provides component-specific red flags, in situations where the widespread installation of component-specific sensors is unrealistic.

With regard to the second research question, a methodology has been developed to identify which data are actually needed. The approach is scenario-based, and formalizes mathematically the relations between data, grid management and grid performance. It has been applied in three studies. One study uses the methodology to explore impacts of data quality on grid management costs, and to evaluate the most profitable amount of investment needed into data quality improvement. Another study provides a tool to evaluate the profitability of investments in condition-monitoring sensors. A third study investigates how data granularity affects decisions in grid upgrades, and ultimately the quality of power supply, and proposes a way to decide where, how, and how much to upgrade the cyberinfrastructure.

In summary, this thesis: i) shows the importance of data for grid performance; ii) conceptualizes, formalizes mathematically, and quantifies relations between data, grid management, and grid performance; iii) develops new approaches to support the transformation of the cyberinfrastructure needed for a transition to smart grids.

Det nuvarande kraftsystemet är ett nät av elektriska apparater som utgör ett fysiskt system. Det upplever förändringar, till exempel ökande antal elfordon och distribuerade energikällor. Samtidigt kopplas cybernätverk in i det fysiska nätet i allt högre grad. Dessa förändringar förvandlar elnätet till ett smart nät som är ofta i fokus för forskningslitteraturen. Den omvandlingen är starkt kopplat till den samtidiga utvecklingen av cyberinfrastrukturen, som är fokus för denna avhandling. Det syftar till att stödja den stegvisa uppgraderingen av datasystem, med hänsyn till verkliga begränsningar, såväl som möjligheter som sensorer och maskininlärning erbjuder. Med denna bakgrund har två forskningsfrågor identifierats: i) Hur kan man använda tillgängliga data mer effektivt för att förbättra nätförvaltning? och ii) Hur mycket och vilken typ av nya data behövs egentligen?

När det gäller den första forskningsfrågan undersöks sätt att använda till-gängliga data mer effektivt i samarbete med en elnätsägare. Ett alternativ är att elnätsägare använder erkänt välfungerande metoder när det gäller datahantering, som inkluderar: kombinera data från olika källor, förbättra rapporteringsmetoder och automatisera uppgifter. För att illustrera hur en kombination av tillgängliga data kan ge ytterligare relevant information, har kritikalitetsindex beräknats och tilldelats komponenter i en transformatorstation genom att kombinera avbrottsdata, driftdata och nätstruktur. Ett annat alternativ är att utveckla maskininlärningsalgoritmer för att genomföra specifika eller nya uppgifter. Ett felvarningssystem har utvecklats med hjälp av maskininlärning för att utnyttja befintliga data om elkraftapparater. Det ger apparatspecifika röda flaggor, i situationer  där en utbredda installation av särskilda sensorer är orealistisk.

När det gäller den andra forskningsfrågan har en metodik utvecklats för att identifiera vilka data som faktiskt behövs. Tillvägagångssättet är scenariobaserat och formaliserar matematiska relationer mellan data, näthantering och nätprestanda. Det har tillämpats i tre studier. En studie använder metoden för att undersöka inverkan av datakvalitet på nätförvaltningskostnader och för att identifiera det mest lönsamma investeringsbeloppet för datakvaliteten förbättring. En annan studie föreslår ett verktyg för att utvärdera lönsamheten för investeringar i tillståndsövervakningssensorer. En tredje studie utreder hur datagranularitet påverkar beslut om nätuppgraderingar  , och i slutändan kvaliteten på strömförsörjningen, och föreslår ett sätt att bestämma var, hur och hur mycket cyberinfrastrukturen ska uppgraderas.

Denna avhandling: i) visar betydelsen av data för nätprestanda ; ii) konceptualiserar, formaliserar matematiskt och kvantifierar relationer mellan data, näthantering samt nätprestanda; iii) utvecklar nya metoder för att stödja omvandlingen av den cyberinfrastruktur som behövs för en övergång till smarta nät.

The deployment of sensors enables the development of condition-based maintenance, as opposed to the traditional time-based and corrective maintenance. This work explores the conditions under which the use of sensors to improve maintenance scheduling on overhead lines is economically profitable. We propose a novel methodology that converts sensor measurements into an asset condition assessment, and then into a maintenance decision. The cost of predictive maintenance is then compared to the cost of corrective maintenance over several decades, ultimately allowing to evaluate the profitability of investing into sensors. This work enables to identify the parameter values that result in profitable investments in sensors. The results show that the use of sensors is particularly justified for short-lived assets, supplying many clients.

With an aging infrastructure and complexifying grid, sensors become essential to monitor the state of power system components. They are part of the evolution of power grids into smarter grids, as well as of the development of digital twins. Currently, sensors are largely absent from distribution grids. While much research literature exists in optimal sensor placement with cost minimization under performance constraints, few publications quantify the profitability of such sensors for distribution grid operators (DSOs). This work aims to bridge this gap and to offer a novel methodology that enables DSOs to assess if and where sensors would be profitable in the context of asset management. The methodology is based on modeling the effects of sensors on maintenance activities and then on grid reliability. We formulate a binary non linear optimization problem, and apply it to study the case of sensors that monitor the condition of power lines. Results show that sensor profitability depends on several factors including i) the component age, ii) the current replacement practice, and iii) the component importance. Thus, this work gives a tool for DSOs to decide whether and where to invest into sensors.

Monitoring the condition of components helps taking preventive actions to avoid failures, and increases reliability. However, performing such monitoring for all components of the distribution grid is prohibitively expensive. Instead, distribution system operators could focus efforts only on the most critical components. In particular, importance indices enable to prioritise components according to a chosen criterion, and to adapt monitoring strategies. This study presents methods to rank grid components using outage data. The importance indices are based on: 1) de-energisation time; 2) frequency of failures; 3) disconnected power; 4) energy not supplied and 5) customer outage time. The results depend largely on the time period of the outage data considered for analysis. Some components' rank varies with the chosen criterion. This indicates that they are critical with respect to a specific criterion. Other components are ranked high with all the methods, which means that they are critical, and need focused monitoring.

Physical asset management in the electric power sector encompasses the scheduling of the maintenance and replacement of grid components, as well as decisions about investments in new components. Data plays a crucial role in these decisions. The importance of data is increasing with the transformation of the power system and its evolution toward smart grids. This thesis deals with questions related to data management as a way to improve the performance of asset management decisions. Data management is defined as the collection, processing, and storage of data. Here, the focus is on the collection and processing of data.

First, the influence of data on the decisions related to assets is explored. In particular, the impacts of data quality on the replacement time of a generic component (a line for example) are quantified using a scenario approach, and failure modeling. In fact, decisions based on data of poor quality are most likely not optimal. In this case, faulty data related to the age of the component leads to a non-optimal scheduling of component replacement. The corresponding costs are calculated for different levels of data quality. A framework has been developed to evaluate the amount of investment needed into data quality improvement, and its profitability.

Then, the ways to use available data efficiently are investigated. Especially, the possibility to use machine learning algorithms on real-world datasets is examined. New approaches are developed to use only available data for component ranking and failure prediction, which are two important concepts often used to prioritize components and schedule maintenance and replacement.

A large part of the scientific literature assumes that the future of smart grids lies in big data collection, and in developing algorithms to process huge amounts of data. On the contrary, this work contributes to show how automatization and machine learning techniques can actually be used to reduce the need to collect huge amount of data, by using the available data more efficiently. One major challenge is the trade-offs needed between precision of modeling results, and costs of data management.

Data play an essential role in asset management decisions. The amount of data is increasing through accumu-lating historical data records, new measuring devices, and communication technology, notably with the evolution toward smart grids. Consequently, the management of data quantity and quality is becoming even more relevant for asset managers to meet efficiency and reliability requirements for power grids. In this work, we propose an innovative data quality management framework enabling asset managers (i) to quantify the impact of poor data quality, and (ii) to determine the conditions under which an investment in data quality improvement is required. To this end, an algorithm is used to determine the optimal year for component replacement based on three scenarios, a Reference scenario, an Imperfect information scenario, and an Investment in higher data quality scenario. Our results indicate that (i) the impact on the optimal year of replacement is the highest for middleaged components; (ii) the profitability of investments in data quality improvement depends on various factors, including data quality, and the cost of investment in data quality improvement. Finally, we discuss the implementation of the proposed models to control data quality in practice, while taking into account real-world technological and economic limitations.

Power grid operators use data to guide their asset management decisions. However, as the complexity of collected data increases with time and amount of sensors, it becomes more difficult to extract relevant information. Therefore, methods that perform detection tasks need to be developed, especially in distribution systems, which are impacted by distributed generation and smart appliances. Until now, methods employed in local power systems for detection purposes using data with low sampling rate, have not been reviewed. This paper provides a literature review focused on anomaly detection, fault location, and load disaggregation. We analyze the methods in terms of their type, data requirements and ways they are implemented. Many belong to the machine learning field. We find that some methods are typically combined with others and perform specific tasks, while other methods are more ubiquitous and often used alone. Continued research is needed to identify how to guide the choice of methods, and to investigate combinations of methods that have not been studied yet.

Power systems are becoming more complex, which increases instability issues and outage risks. The development of smart grids could help manage such complex systems. One important pillar in smart grids is big data analytics. In this poster paper, we discuss where and how machine learning could contribute to more efficient asset management. We also identify challenges that stand in the way of the widespread use of big data analytics in smart grids. While the nature of data, as well as data and asset management systems themselves make the use of big data challenging, data analytics could improve the reliability of power supply by providing the functions of detection, prediction, and selection.