Frequency monitoring network (FNET) is a GPS-synchronized distribution-level phasor measurement system. It is a powerful synchronized monitoring network for large-area power systems that provides significant information and data for power system situational awareness, real time and post-event analysis, and other important aspects of bulk systems. This work explored FNET measurements and utilized them for different applications and power system analysis. An island system was built and validated with FNET measurements to study the stability of the OTEC integration. FNET measurements were also used to validate a large system model like the U.S. Eastern Interconnection. It tries to match the simulation result and frequency measurement of a real event by adjusting the simulation model. The system model is tuned with the combination of different impact factors for different confirmed actual events, and some general rules and specific tuning quantities were concluded from the model validation process. This work also investigated the behavior of the power system frequency during large-scale, synchronous societal events, like the World Cup, Super Bowl and Royal Wedding. It is apparent that large groups of people engaging in the same event at roughly the same time can have significant impacts on the power grid frequency. The systematic analysis of the accumulating and statistical FNET frequency data presents an incisive point of view on the power grid frequency behavior during such events. To better understanding of system events recorded by FNET, a visualization tool was developed to visualize major events that occurred in the North American power grid. The measurement plot combined with the geographical contour map provides intuitive visualization of the event. Finally, the EI system was simplified and clustered into four groups based on FNET measurements and simulation results of generator trip cases. The generation and load capacity of each cluster was calculated based on the clustering result and simulation model, and a flow diagram of this simplified EI system was demonstrated with clusters and power flow between them.

abstract: Unsupervised learning of time series data, also known as temporal clustering, is a challenging problem in machine learning. This thesis presents a novel algorithm, Deep Temporal Clustering (DTC), to naturally integrate dimensionality reduction and temporal clustering into a single end-to-end learning framework, fully unsupervised. The algorithm utilizes an autoencoder for temporal dimensionality reduction and a novel temporal clustering layer for cluster assignment. Then it jointly optimizes the clustering objective and the dimensionality reduction objective. Based on requirement and application, the temporal clustering layer can be customized with any temporal similarity metric. Several similarity metrics and state-of-the-art algorithms are considered and compared. To gain insight into temporal features that the network has learned for its clustering, a visualization method is applied that generates a region of interest heatmap for the time series. The viability of the algorithm is demonstrated using time series data from diverse domains, ranging from earthquakes to spacecraft sensor data. In each case, the proposed algorithm outperforms traditional methods. The superior performance is attributed to the fully integrated temporal dimensionality reduction and clustering criterion. / Dissertation/Thesis / Masters Thesis Computer Engineering 2018

La population des villes devrait doubler d'ici le milieu du siècle, selon les estimations de l’OMS. Cette augmentation rapide de la population a un impact sur les transports et la croissance économique, et accroîtra les responsabilités des autorités de gestion locales. Nous vivons une transformation des villes en villes intelligentes offrant de nouveaux services à la population, en optimisant l’utilisation des ressources disponibles. Qu'il s'agisse de données provenant des citoyens, de données gouvernementales ouvertes ou d'autres sources en ligne, une pluralité de sources de données peut permettre la création d’outils intelligents pour gérer efficacement les activités quotidiennes. De plus, grâce au progrès d'Internet et des technologies mobiles, les plateformes de réseaux sociaux (Twitter) sont devenues des modes de communication populaires. Elles permettent aux utilisateurs de partager un large éventail d'informations, y compris des données spatio-temporelles. Ainsi, Il est aisé d'accéder, en temps réel, à des connaissances provenant de différents types de données disponibles, riches, géo-référencées et issues de sources multiples et de les intégrer sur une carte. Il s'agit d'une réelle opportunité d'enrichir les cartes traditionnelles.Dans cette thèse, nous proposons d'abord un système de recommandation d'itinéraires, tenant compte des contraintes temps réel, en l'absence d'infrastructure physique ; en exploitant les données géolocalisées issues de réseaux sociaux (twitter) pour identifier les contraintes de trafic temps réel et, par conséquent, recommander un chemin optimisé. Nous avons mis en œuvre un système d'indexation à base de grille spatiale pour notre modèle de prédiction en quasi-temps réel. Ensuite, nous avons introduit le concept de "cartes intelligentes " intégrant la représentation visuelle de couches de « connaissances pertinentes » par le biais de la collecte, la gestion et l'intégration de sources de données hétérogènes. Contrairement aux cartes conventionnelles, les cartes intelligentes extraient des informations à partir des événements annoncés et découverts en temps réel (concerts, incidents, ...), les offres en ligne et les analyses statistiques (zones dangereuses, …) en encapsulant les données entrantes semi-structurées et non structurées dans des paquets génériques structurés.Cette méthodologie ouvre la voie à la fourniture de services et applications intelligents. De plus, le développement de ‘’cartes intelligentes’’ nécessite un traitement efficace et évolutif et la visualisation de couches basées sur les connaissances à plusieurs échelles cartographiques, permettant ainsi une navigation fluide et sans encombre. Enfin, nous présentons Hadath, un système évolutif et efficace qui extrait les événements sociaux d'une multitude de flux de données non structurés. Nous utilisons le traitement du langage naturel et les techniques de regroupement multidimensionnel pour extraire les ‘’événements pertinents’’ à différentes échelles cartographiques et pour déduire l'étendue spatio-temporelle des événements détectés.Le système comprend un composant de gestion et de prétraitement des différents types de sources de données et génère des paquets de données structurés à partir de flux non structurés. Notre système comprend également un schéma d'indexation spatio-temporelle hiérarchique en mémoire pour permettre un accès efficace et évolutif aux données brutes, ainsi qu'aux groupes d'événements extraits. Dans un premier temps, les paquets de données sont traités pour la découverte d’événements à l'échelle locale, puis l'étendue spatio-temporelle appropriée. Par conséquent, les événements détectés sont affichés à différentes résolutions spatio-temporelles, ce qui permet une navigation fluide. Enfin, pour valider notre approche, nous avons mené des expériences sur des flux de données réelles. Le résultat final du système proposé, nommé Hadath crée une expérience unique et dynamique de navigation cartographique. / The population in cities is slated to double by mid-century according to estimates prepared by the World Health Organization. This rapid increase in population will impact transportation and economic growth, and will increase responsibilities of local managing authorities and different stakeholders. It is a need of the hour to convert cities into smart cities in order to provide new service to the public, by using available resources in an optimum manner. From crowd-sourced data and open governmental data to other online sources, a variety of data sources can provide users with smart tools to efficiently manage their daily activities. Moreover, with the advancement in Internet and mobile technologies, social networking platforms such as Facebook and Twitter have become popular modes of communication. They allow users to share a spectrum of information, including spatio-temporal data, both publicly and within their community of interest in real-time. Scrutinizing knowledge from different types of available, rich, geo-tagged, and crowd-sourced data and incorporating it on a map has become more feasible. This presents a real opportunity to enrich traditional maps and enhance conventional spatio-temporal queries with the help of different types of data extracted from a variety of available data sources. In this thesis, we first propose a constraint-aware route recommendation system in lack of physical infrastructure environment that leverages geo-tagged data in social media and user-generated content to identify upcoming traffic constraints and, thus, recommend an optimized path. We have designed and developed a system using a spatial grid index to inform users about upcoming constraints and calculate a new, optimized path in minimal response time. Later, the concept of “smart maps” will be introduced by collecting, managing, and integrating heterogeneous data sources in order to infer relevant knowledge-based layers. Unlike conventional maps, smart maps extract information about live events (e.g., concert, competition, incidents, etc.), online offers, and statistical analysis (e.g., dangerous areas) by encapsulating incoming semi- and un-structured data into structured generic packets. This methodology sets the ground for providing different intelligent services and applications. Moreover, developing smart maps requires an efficient and scalable processing and the visualization of knowledge-based layers at multiple map scales, thus allowing a smooth and clutter-free browsing experience. Finally, we introduce Hadath, a scalable and efficient system that extracts social events from a variety of unstructured data streams. Hadath applies natural language processing and multi-dimensional clustering techniques to extract relevant events of interest at different map scales, and to infer the spatio-temporal extent of detected events. The system comprises a data wrapping component which digests different types of data sources, and prepossesses data to generate structured data packets out of unstructured streams. Hadath also implements a hierarchical in-memory spatio-temporal indexing scheme to allow efficient and scalable access to raw data, as well as to extracted clusters of events. Initially, data packets are processed to discover events at a local scale, then, the proper spatio-temporal extent and the significance of detected events at a global scale is determined. As a result, live events can be displayed at different spatio-temporal resolutions, thus allowing a smooth and unique browsing experience. Finally, to validate our proposed system, we conducted experiments on real-world data streams. The final output of our system named Hadath creates a unique and dynamic map browsing experience