Power systems have been historically designed at a time when the production was<p>centralized and the electricity had to be transmitted to the loads from the closest power<p>plant. Nowadays, there is an increasing integration of decentralized and intermittent pro-<p>duction. Moreover, the energy market coupling has enabled the transfer of electric power<p>for economical purposes. Also, former isolated power systems are now interconnected for<p>reliability and financial reasons.<p>All of these changes make difficult to predict the future behavior of the grid. Studies<p>are done in order to plan for the future needs of the system. However, building new in-<p>frastructures takes time and it is expected that these needs will not be completely fulfilled<p>in all the parts of the grid. Therefore, transmission of active power could be limited by<p>the existing infrastructure. For example, the presence of inter-area oscillations is often<p>the limiting factor when a high active power is transmitted on a long transmission line<p>between two groups of generators. Since higher levels of active power are exchanged on<p>longer distances, problems of inter-area oscillations may arise in power systems previously<p>not affected by this phenomenon.<p>In this work, a measurement-based approach, able to predict in the short-term the<p>future behavior of oscillations, is presented. This approach is complementary to the<p>long-term planning of the grid.<p>The mandatory first step towards a measurement-based approach is to have the ability<p>to extract useful information among a huge quantity of data. To face this issue, some<p>comparisons of data mining algorithms are performed. The proposed method combines<p>two decision tree algorithms to obtain both prediction accuracy and comprehensibility.<p>The second required step for building a measurement-based model is to take into<p>account the limitations of the measurements. Two types of wide area measurements are<p>used, synchronized measurements from PMUs and traditional unsynchronized data from<p>the SCADA/EMS system. Oscillation monitoring using PMUs is especially of interest<p>and an approach is presented to post-process damping estimates. This post-processing<p>method consists in a noise reduction technique followed by a damping change detection<p>algorithm.<p>Finally, the method, combining these two steps, is implemented to analyze the Con-<p>tinental European grid. This implementation takes place in the context of the European<p>project Twenties. The results, using several months of measurements, are described in<p>detail before being discussed. / Doctorat en Sciences de l'ingénieur / info:eu-repo/semantics/nonPublished

Power systems must face new challenges in the current environment. The energy market liberalization and the increase in the loading level make the occurrence of instability phenomena leading to large blackouts more likely. Existing tools must be improved and new tools must be developed to avoid them.<p><p>The aim of this thesis is the improvement of the voltage stability assessment efficiency. Two orientations are studied: the monitoring methods and the load modeling.<p><p>The purpose of the monitoring methods is to evaluate the voltage stability using only measurements and without running simulations. <p><p>The first approach considered is local. The parameters of the Thevenin equivalent seen from a load bus are assessed thanks to a stream of local voltage and current measurements. Several issues are investigated using measurements coming from complete time-domain simulations. The applicability of this approach is questioned.<p><p>The second approach is global and uses measurements acquired by a Wide-Area Measurement System (WAMS). An original approach with a certain prediction capability is proposed, along with intuitive visualizations that allow to understand the deterioration process leading to the collapse.<p><p>The load modeling quality is certainly the weak point of the voltage security assessment tools which run simulations to predict the stability of the power system depending on different evolutions. Appropriate load models with accurate parameters lead to a direct improvement of the prediction precision.<p><p>An innovative procedure starting from data of long measurement campaigns is proposed to automatically evaluate the parameters of static and dynamic load models. Real measurements taken in the Belgian power system are used to validate this approach.<p> / Doctorat en Sciences de l'ingénieur / info:eu-repo/semantics/nonPublished