Nesta tese são apresentados um novo modelo estatístico para cálculo de probabilidade de paternidade e sua implementação em software. O modelo proposto utiliza o genótipo como informação básica, em contraste com outros modelos que usam alelos. Por esta diferença, o modelo proposto resulta mais abrangente, mas que, sob certas restrições, reproduz os resultados dos modelos que usam alelos. Este modelo foi implementado em um software que recebe descrições da genealogia e dos marcadores em uma linguagem dedicada a isso e constrói uma rede bayesiana para cada marcador. O usuário pode definir livremente a genealogia e os marcadores. O cálculo da probabilidade de paternidade é feito, sobre as redes construídas, por um software para inferência em redes bayesianas e a probabilidade de paternidade combinada considerando todos os marcadores é calculada, resultando em um \"índice de paternidade. / This thesis presents a novel statistical model for calculation of the probability of paternity and its implementation as a software. The proposed model uses genotype as basic information. Other models use alleles as basic information. As a result the proposed model is broader, in the sense that, under certain constraints the results from the other models are reproduced. The software implementation receives pedigree and markers data, in a specifically designed language, as input and builds one bayesian network for each marker. The user can freely define any pedigree and any marker. Paternity probabilities for each locus are calculated, from the built networks, by a software for inference on Bayesian Networks and these probabilities are combined into a single \"paternity index\".

In this dissertation, several approaches I have taken to build upon the student learning model are described. There are two focuses of this dissertation. The first focus is on improving the accuracy with which future student knowledge and performance can be predicted by individualizing the model to each student. The second focus is to predict how different educational content and tutorial strategies will influence student learning. The two focuses are complimentary but are approached from slightly different directions. I have found that Bayesian Networks, based on belief propagation, are strong at achieving the goals of both focuses. In prediction, they excel at capturing the temporal nature of data produced where student knowledge is changing over time. This concept of state change over time is very difficult to capture with classical machine learning approaches. Interpretability is also hard to come by with classical machine learning approaches; however, it is one of the strengths of Bayesian models and aids in studying the direct influence of various factors on learning. The domain in which these models are being studied is the domain of computer tutoring systems, software which uses artificial intelligence to enhance computer based tutorial instruction. These systems are growing in relevance. At their best they have been shown to achieve the same educational gain as one on one human interaction. Computer tutors have also received the attention of White House, which mentioned an tutoring platform called ASSISTments in its National Educational Technology Plan. With the fast paced adoption of these data driven systems it is important to learn how to improve the educational effectiveness of these systems by making sense of the data that is being generated from them. The studies in this proposal use data from these educational systems which primarily teach topics of Geometry and Algebra but can be applied to any domain with clearly defined sub-skills and dichotomous student response data. One of the intended impacts of this work is for these knowledge modeling contributions to facilitate the move towards computer adaptive learning in much the same way that Item Response Theory models facilitated the move towards computer adaptive testing.

"A common problem shared amongst online tutoring systems is the time-consuming nature of content creation. It has been estimated that an hour of online instruction can take up to 100-300 hours to create. Several systems have created tools to expedite content creation, such as the Cognitive Tutors Authoring Tool (CTAT) and the ASSISTments builder. Although these tools make content creation more efficient, they all still depend on the efforts of a content creator and/or past historical. These tools do not take full advantage of the power of the crowd. These issues and challenges faced by online tutoring systems provide an ideal environment to implement a solution using crowdsourcing. I created the PeerASSIST system to provide a solution to the challenges faced with tutoring content creation. PeerASSIST crowdsources the work students have done on problems inside the ASSISTments online tutoring system and redistributes that work as a form of tutoring to their peers, who are in need of assistance. Multi-objective multi-armed bandit algorithms are used to distribute student work, which balance exploring which work is good and exploiting the best currently known work. These policies are customized to run in a real-world environment with multiple asynchronous reward functions and an infinite number of actions. Inspired by major companies such as Google, Facebook, and Bing, PeerASSIST is also designed as a platform for simultaneous online experimentation in real-time and at scale. Currently over 600 teachers (grades K-12) are requiring students to show their work. Over 300,000 instances of student work have been collected from over 18,000 students across 28,000 problems. From the student work collected, 2,000 instances have been redistributed to over 550 students who needed help over the past few months. I conducted a randomized controlled experiment to evaluate the effectiveness of PeerASSIST on student performance. Other contributions include representing learning maps as Bayesian networks to model student performance, creating a machine-learning algorithm to derive student incorrect processes from their incorrect answer and the inputs of the problem, and applying Bayesian hypothesis testing to A/B experiments. We showed that learning maps can be simplified without practical loss of accuracy and that time series data is necessary to simplify learning maps if the static data is highly correlated. I also created several interventions to evaluate the effectiveness of the buggy messages generated from the machine-learned incorrect processes. The null results of these experiments demonstrate the difficulty of creating a successful tutoring and suggest that other methods of tutoring content creation (i.e. PeerASSIST) should be explored."

Submitted by Maicon Juliano Schmidt (maicons) on 2015-06-09T17:46:58Z No. of bitstreams: 1 Henrique Manfron Seffrin_.pdf: 4996070 bytes, checksum: facf64690edf2c78dfd329c9ec67d18c (MD5) / Made available in DSpace on 2015-06-09T17:46:58Z (GMT). No. of bitstreams: 1 Henrique Manfron Seffrin_.pdf: 4996070 bytes, checksum: facf64690edf2c78dfd329c9ec67d18c (MD5) Previous issue date: 2015-02-20 / CAPES - Coordenação de Aperfeiçoamento de Pessoal de Nível Superior / CNPQ – Conselho Nacional de Desenvolvimento Científico e Tecnológico / Pesquisas têm mostrado que os alunos apresentam ganhos mais significativos de aprendizagem através do ensino individualizado, pois o professor pode se focar nas dificuldades de cada um. Por ser uma estratégia de custo elevado, os Sistemas Tutores Inteligentes (STI) oferecem uma alternativa mais viável. Esses sistemas, através de técnicas de Inteligência Artificial, são capazes de se adaptar às características de cada aluno, provendo assistência individualizada. Esta adaptação personalizada é fornecida pelo componente Modelo de Aluno, que é capaz de avaliar e mapear o conhecimento de cada estudante. Na literatura, são encontrados diversos trabalhos que lidam com a questão de avaliação de conhecimento do aluno, dentre os quais encontram-se alguns trabalhos relacionados ao domínio de álgebra. Estes trabalhos, geralmente, apresentam modelagens com redes Bayesianas, que são estruturas probabilísticas amplamente utilizadas por apresentarem resultados muito interessantes no que se refere à avaliação do conhecimento dos estudantes. No entanto, nestes trabalhos, estas estruturas relacionam apenas os conceitos algébricos, ou modelam relações entre operações algébricas, com suas principais propriedades e falsas concepções. Esses trabalhos não buscam definir as relações entre os conceitos algébricos e as respectivas operações, e como os primeiros podem estar interferindo, positivo ou negativamente, na aprendizagem dos segundos. Por exemplo, na álgebra, há conceitos chave, como incógnita e a igualdade entre os lados da equação, que interferem diretamente na compreensão de certas operações algébricas. Se um estudante não os compreende, dificilmente ele será capaz de aplicar corretamente as operações relacionadas em todas as situações. Desse modo, é desejável que os modelos de inferência sejam capazes de identificar se o estudante compreende tais conceitos. Além disso, outra limitação dos trabalhos relacionados de modelos de alunos voltados para a álgebra se refere a como eles tratam as evidências. Como estes trabalhos utilizam os itens de avaliação para isto, a cada novo exercício, é necessário inserir um novo nodo na rede, e estabelecer as relações com cada conceito abordado por este item. Isso torna o projeto da rede trabalhoso e dependente de cada exercício aplicado no STI. Nesse contexto, este trabalho propõe um modelo de aluno algébrico que além de inferir o conhecimento algébrico dos estudantes de conceitos (como incógnita, igualdades, operações inversas), habilidades (operações algébricas) e falsas concepções, busca definir as relações entre conceitos e habilidades. Como foco inicial deste trabalho serão utilizadas as equações de 1o grau. Para a inferência, será empregada a estrutura de Redes Bayesianas Dinâmicas (RBD), usando como evidência a operação aplicada pelo aluno em cada passo da resolução de uma equação. Nesta estrutura de RBD, cada time slice corresponde à resolução de um passo, o que torna o modelo proposto independente dos exercícios aplicados pelo STI. Dessa forma, o modelo de inferência proposto pode ser utilizado em qualquer equação algébrica, sem a necessidade de qualquer alteração na rede, como ocorre nos outros trabalhos relacionados. Visando verificar a capacidade de inferência desta rede, foram conduzidas avaliações. A partir dos históricos dos alunos, que utilizaram o PAT2Math, foram obtidas as evidências para a rede; e a partir dos dados dos pós-testes, realizados pelos mesmos alunos, formam obtidos os percentuais a serem comparados com a inferência da rede. Como os resultados não foram satisfatórios, empregou-se a regra do limiar, instanciando toda a variável que o ultrapassasse. Avaliada sob os limiares de 96% e 98%, a rede demostrou resultados mais precisos com o limiar de 96%, no qual as diferenças entre os resultados da rede e os percentuais dos pós-testes permaneceram, em sua maioria, em até 5%. / Students learn more through personalized instruction, because the teacher can focus on each learner. Being a impracticable strategy in terms of cost, Intelligent Tutoring Systems (ITS) offers a feasible alternative. By using Artificial Intelligence techniques, these systems are able to adapt themselves to the students, providing individualized instruction. Such adaptation is provided by the Student Model, which is able to assess and map the knowledge of each student. In the literature there are several studies that deal with knowledge evaluation in ITS, some of them are related to algebra. These studies present a Bayesian Network modeling, probabilistic structures that are widely used because of their interesting results concerning the evaluation of the student knowledge. However, in this studies, the network structure only models algebraic concepts, or only model a relationship between algebraic operations and its main properties and common misconceptions. These studies do not aim to represent the relationship between concepts and algebraic operations and how the former can be interfering, in a positive or negative way, on the learning of the second one. For example, in algebra, there are key concepts, such as the unknown and equality among sides of the equation, which directly interferes with the understanding of some algebraic operations. If a student does not understand these concepts, he would hardly be able to apply correctly the related operations in every situation. Thus, it is desirable that the inference model be able to identify if the student understands such concepts. In addition, another limitation of the related work of algebraic student models refers to how they deal with the evidence. As these studies use the assessment items for evidence, for each new exercise, it is necessary to insert a new node in the network, and establish relationships with each concept addressed by this item. This makes the network design laborious and dependent on each ITS exercise. In this context, this work proposes an algebraic student model that, in addition to infer the student knowledge of algebraic concepts (as unknown, equality, inverse operation), skills (algebraic operations) and common misconceptions, defines the relationship between concepts and skill. An initial focus of this study will be the 1st degree equations. For the inference model we use the Dynamic Bayesian Networks (DBN), in which the evidences are the operations applied by the student to solve each equation step. In this structure of DBN, each time slice corresponds to a resolution step, which makes the proposed model independent of the ITS exercises. Thus, the proposed inference model can be used in every algebraic equation, without need to make changes in the network, as occurs with other works.In order to verify the inference capacity of the network, evaluations were conducted. From the resolution history of the students, that interact with PAT2Math, the evidences for the network were obtained; and from the post-test data, solved by the same students, the percentages to compare with the results of the network were obtained. As the results aren’t very satisfactory, we applied the threshold rule, every variable that exceeded this value are instantiated. The network were evaluated under the threshold of 96% and 98%. The proposed DBN has shown more accurate inference with the 96% threshold, in which the differences between the results of the network and the percentages of the post-test remained mostly with ceiling of 5%.

Submitted by Silvana Teresinha Dornelles Studzinski (sstudzinski) on 2016-03-17T16:13:01Z No. of bitstreams: 1 Leonardo Ribeiro Machado_.pdf: 1406241 bytes, checksum: d0229eb3cc9a08809b94e758fa60d7e6 (MD5) / Made available in DSpace on 2016-03-17T16:13:01Z (GMT). No. of bitstreams: 1 Leonardo Ribeiro Machado_.pdf: 1406241 bytes, checksum: d0229eb3cc9a08809b94e758fa60d7e6 (MD5) Previous issue date: 2011-03-31 / CAPES - Coordenação de Aperfeiçoamento de Pessoal de Nível Superior / O desempenho de um SGBD é um fator crítico a ser considerado durante a sua utilização. Diversas técnicas são atualmente empregadas na tentativa de aumentar o desempenho de um SGBD. Esta pesquisa integra tecnologias de agentes e de mineração de dados para a criação de modelos probabilísticos (bayesianos) de decisão aptos a auxiliar no processo de melhoria de desempenho de um SGBD. Este modelo é usado, então, como base da ferramenta ATTuneDB de sintonia de SGBD. A partir da carga real de operação de um SGBD PostgreSQL, a ferramenta utiliza este modelo para identificar o regime de trabalho do SGBD e encontrar o melhor conjunto de valores para os parâmetros deste SGBD, apoiando o administrador do SGBD na tarefa de otimizar o desempenho deste. / The performance of a DBMS is a critical factor to be considered while using it. Several techniques are currently employed in an attempt to increase the performance of a DBMS. This research integrates agent technologies and data mining for building probabilistic decision models (Bayesian) able to assist the performance improvement process of a DBMS. This model is used to build the ATTuneDB DBMS fine-tuning tool. Receiving information about the real workload being submitted to a PostgreSQL DBMS, and using the probabilistic model, the tool is able to identify the type of the workload, and find the best set of value for the parameters of this DBMS, thus, supporting the DBA on the task of optimizing the DBMS performance.

La prédiction et l'inférence sur des données temporelles sont très souvent effectuées en utilisant uniquement les séries temporelles. Nous sommes convaincus que ces tâches pourraient tirer parti de l'utilisation des métadonnées contextuelles associées aux séries temporelles, telles que l'emplacement, le type, etc. Réciproquement, les tâches de prédiction et d'inférence sur les métadonnées pourraient bénéficier des informations contenues dans les séries temporelles. Cependant, il n'existe pas de méthode standard pour modéliser conjointement les données de séries temporelles et les métadonnées descriptives. De plus, les métadonnées contiennent fréquemment des informations hautement corrélées ou redondantes et peuvent contenir des erreurs et des valeurs manquantes.Nous examinons d’abord le problème de l’apprentissage de la structure graphique probabiliste inhérente aux métadonnées en tant que réseau Bayésien. Ceci présente deux avantages principaux: (i) une fois structurées en tant que modèle graphique, les métadonnées sont plus faciles à utiliser pour améliorer les tâches sur les données temporelles et (ii) le modèle appris permet des tâches d'inférence sur les métadonnées uniquement, telles que l'imputation de données manquantes. Cependant, l'apprentissage de la structure de réseau Bayésien est un défi mathématique conséquent, impliquant un problème d'optimisation NP-difficile. Pour faire face à ce problème, nous présentons un algorithme d'apprentissage de structure sur mesure, inspiré de nouveaux résultats théoriques, qui exploite les dépendances (quasi)-déterministes généralement présentes dans les métadonnées descriptives. Cet algorithme est testé sur de nombreux jeux de données de référence et sur certains jeux de métadonnées industriels contenant des relations déterministes. Dans les deux cas, il s'est avéré nettement plus rapide que l'état de la l'art, et a même trouvé des structures plus performantes sur des données industrielles. De plus, les réseaux Bayésiens appris sont toujours plus parcimonieux et donc plus lisibles.Nous nous intéressons ensuite à la conception d'un modèle qui inclut à la fois des (méta)données statiques et des données temporelles. En nous inspirant des modèles graphiques probabilistes pour les données temporelles (réseaux Bayésiens dynamiques) et de notre approche pour la modélisation des métadonnées, nous présentons une méthodologie générale pour modéliser conjointement les métadonnées et les données temporelles sous forme de réseaux Bayésiens hybrides statiques-dynamiques. Nous proposons deux algorithmes principaux associés à cette représentation: (i) un algorithme d'apprentissage qui, bien qu'optimisé pour les données industrielles, reste généralisable à toute tâche de fusion de données statiques et dynamiques, et (ii) un algorithme d'inférence permettant les d'effectuer à la fois des requêtes sur des données temporelles ou statiques uniquement, et des requêtes utilisant ces deux types de données.%Nous fournissons ensuite des résultats sur diverses applications inter-domaines telles que les prévisions, le réapprovisionnement en métadonnées à partir de séries chronologiques et l’analyse de dépendance d’alarmes en utilisant les données de certains cas d’utilisation difficiles de Schneider Electric.Enfin, nous approfondissons certaines des notions introduites au cours de la thèse, et notamment la façon de mesurer la performance en généralisation d’un réseau Bayésien par un score inspiré de la procédure de validation croisée provenant de l’apprentissage automatique supervisé. Nous proposons également des extensions diverses aux algorithmes et aux résultats théoriques présentés dans les chapitres précédents, et formulons quelques perspectives de recherche. / Prediction and inference on temporal data is very frequently performed using timeseries data alone. We believe that these tasks could benefit from leveraging the contextual metadata associated to timeseries - such as location, type, etc. Conversely, tasks involving prediction and inference on metadata could benefit from information held within timeseries. However, there exists no standard way of jointly modeling both timeseries data and descriptive metadata. Moreover, metadata frequently contains highly correlated or redundant information, and may contain errors and missing values.We first consider the problem of learning the inherent probabilistic graphical structure of metadata as a Bayesian Network. This has two main benefits: (i) once structured as a graphical model, metadata is easier to use in order to improve tasks on temporal data and (ii) the learned model enables inference tasks on metadata alone, such as missing data imputation. However, Bayesian network structure learning is a tremendous mathematical challenge, that involves a NP-Hard optimization problem. We present a tailor-made structure learning algorithm, inspired from novel theoretical results, that exploits (quasi)-determinist dependencies that are typically present in descriptive metadata. This algorithm is tested on numerous benchmark datasets and some industrial metadatasets containing deterministic relationships. In both cases it proved to be significantly faster than state of the art, and even found more performant structures on industrial data. Moreover, learned Bayesian networks are consistently sparser and therefore more readable.We then focus on designing a model that includes both static (meta)data and dynamic data. Taking inspiration from state of the art probabilistic graphical models for temporal data (Dynamic Bayesian Networks) and from our previously described approach for metadata modeling, we present a general methodology to jointly model metadata and temporal data as a hybrid static-dynamic Bayesian network. We propose two main algorithms associated to this representation: (i) a learning algorithm, which while being optimized for industrial data, is still generalizable to any task of static and dynamic data fusion, and (ii) an inference algorithm, enabling both usual tasks on temporal or static data alone, and tasks using the two types of data.%We then provide results on diverse cross-field applications such as forecasting, metadata replenishment from timeseries and alarms dependency analysis using data from some of Schneider Electric’s challenging use-cases.Finally, we discuss some of the notions introduced during the thesis, including ways to measure the generalization performance of a Bayesian network by a score inspired from the cross-validation procedure from supervised machine learning. We also propose various extensions to the algorithms and theoretical results presented in the previous chapters, and formulate some research perspectives.

Este estudo visa compreender a adequação dos rendimentos às necessidades e condições de vida dos brasileiros. Observando os dados da Pesquisa de Orçamentos Familiares (POF) realizada pelo IBGE (Instituto Brasileiro de Geografia e Estatística) para o período: 2008 e 2009, o estudo identifica um modelo que se concentra na investigação sobre o fato de 75% dos domicílios brasileiros declararem dificuldades orçamentárias. Para desenvolver um modelo, foi utilizada a percepção declarada e subjetiva de adequação da renda, informada pelo chefe de família ou pessoa de referência no domicílio. O referencial teórico baseia-se no comportamento do consumidor e foca nos recursos econômicos. O método quantitativo foi desenvolvido com Inteligência Artificial, mais especificamente Redes Bayesianas. Redes Bayesianas são estruturas em forma de grafos onde as distribuições de probabilidade são representadas por nós ligados por arcos acíclicos, que podem representar ou não relações causais entre as variáveis. No final pretende-se contribuir para o conhecimento e melhoria no desenho de políticas públicas e para as empresas em geral, dando um panorama sobre o que afeta as dificuldades das famílias, proporcionando uma visão que vai além da tradicional divisão de classes econômicas. / This study aims to understand the adequacy of Brazilians´ income to their needs and living conditions. According to the data from the Household Budget Survey (POF) conducted by IBGE (Brazilian Institute of Geography and Statistics) for the years of 2008 - 2009, the study identifies a model which focuses on the investigations about the fact that 75% of Brazilian households reported budgetary difficulties. To develop a model, was used the perceived adequacy of income declared by the householder or reference person in the household. The theoretical framework was based on consumer behavior and focuses on economic resources. The quantitative method was developed by Artificial Intelligence, specifically Bayesian Networks. Bayesian Networks are structures in the form of graphs for which the probability distributions are represented by nodes connected by acyclic arcs, which may or may not represent causal relationships between variables. At the end we intend to contribute to knowledge and improvement in the design of public policies and business in general, giving a more detailed look at what affects the difficulties of families, providing a vision that goes beyond the traditional division of economic classes.

Sistemas de transporte marítimo são essenciais para o Comércio Global, em especial, navios-tanques e seus centros de carga e descarga de produtos líquidos ou gasosos inflamáveis; portanto, é crucial entender como estes sistemas podem falhar, para que seus operadores sejam capazes de manter a sua capacidade de operação. É preciso que cada e toda análise quantitativa de risco compreenda algumas das atividades básicas que devem ser desenvolvidas, para permitir a quantificação dos riscos envolvidos e associados, na operação do sistema ou do processo. Basicamente, devem ser calculadas as probabilidades de ocorrência dos eventos indesejados identificados, bem como a magnitude de suas consequências. O objetivo deste trabalho é aferir se a técnica denominada Rede Bayesiana RB é a mais adequada, comparando-a com as técnicas de árvores de falhas e de eventos, para realizar uma Análise de Risco da operação ou processo de descarga de líquidos inflamáveis, como etanol anidro e/ou produtos petrolíferos, de um naviotanque para um terminal portuário específico Terminal de Granéis Líquidos TGL com foco na interface entre dois sistemas: o navio e o porto, observado o elemento humano, ou seja, o erro humano (Análise da Confiabilidade Humana). Além disso, será realizado um estudo das consequências do vazamento de um líquido inflamável transportado pelo navio, olhando para o pior cenário, a partir da ruptura da tubulação ou do compartimento do navio-tanque. A análise tem por base as recomendações da Organização Internacional Marítima OIM (em inglês, IMO). A OIM tem adotado a Avaliação Formal da Segurança AFS (em inglês, Formal Safety Assessment FSA), como seu modo oficial de receber as sugestões de seus membros para criar ou modificar qualquer regulamentação correlacionada. Este processo é composto de cinco passos que a OIM descreve na guia AFS (IMO, 2002). Este trabalho irá mostrar todas as etapas, mas irá focar, com especial atenção, a segunda etapa Risk Assessment, porque será aplicada ao caso sob análise, envolvendo o comportamento humano. Existem muitas técnicas e muito trabalho envolvido na estimação das probabilidades dos eventos. O mesmo ocorre para a avaliação de suas consequências. Uma vez definida a quantidade total de vazamento, um software poderá ser usado para calcular as consequências. O mesmo será feito para na Análise de Risco, utilizando RB, e, neste ponto, o trabalho apresenta uma nova contribuição. / Maritime transportation systems are essential for World Trade, in special, Tankers ships and yours loading and unloading facilities; therefore, it is crucial to understand how these systems may fail, to be able to maintain their capacity. It need that each and every quantitative risk assessment comprises some basic activities that have to be developed to allow the quantification of the risks involved in the operation of a system or process. Basically, it must be estimated the likelihood of the identified undesired events as well as the magnitude of their consequences. The objective of this study is to assess if the technique called Bayesian Networks BN is the best suited, with respect to the Fault Tree Analysis FTA and the Event Tree Analysis ETA, to perform an Risk Analysis of the operation or process of unloading of flammable bulk liquids, such as anhydrous ethanol and/or oil products, from a Tanker to a port terminal specified the Bulk Liquid Terminal BLT, focusing on the interface between the two systems: ship and port with the inclusion of the human factor, i.e., human error: Human Reliability Analysis HRA. Furthermore, a consequence analysis of a specific liquid bulk leakage will be performed, looking at the worst scenario case, from the rupture of a pipeline or tank from a Tanker. The analysis came from based on the recommendations of the International Maritime Organization IMO. The IMO has adopted the FSA (Formal Safety Assessment) as its official way of receiving suggestions of its members to create or modify any regulation correlated. It is a process composed by five steps that IMO has described in its Guidelines for FSA (IMO, 2002). This thesis will to show all steps, but will look carefully to step two (Risk Assessment) because it will be applied in the example situation, involving human behavior (HRA). There are many techniques and much work involved in the estimation of the likelihood of the events. The same occurs for the evaluation of their consequences. Once defined the total leaked quantity, software will be used to calculate the consequences. The same will be done to Risk Analysis, using BN, and at this point, the work is a new contribution.

Dans ces travaux de thèse, nous proposons un modèle de prévision à court terme des flux de voyageurs basé sur les réseaux bayésiens. Ce modèle est destiné à répondre à des besoins opérationnels divers liés à l'information voyageurs, la régulation des flux ou encore la planification de l'offre de transport. Conçu pour s'adapter à tout type de configuration spatiale, il permet de combiner des sources de données hétérogènes (validations des titres de transport, comptages à bord des trains et offre de transport) et fournit une représentation intuitive des relations de causalité spatio-temporelles entre les flux. Sa capacité à gérer les données manquantes lui permet de réaliser des prédictions en temps réel même en cas de défaillances techniques ou d'absences de systèmes de collecte / In this thesis, we propose a Bayesian network model for short-term passenger flow forecasting. This model is intended to cater for various operational needs related to passenger information, passenger flow regulation or operation planning. As well as adapting to any spatial configuration, it is designed to combine heterogeneous data sources (ticket validation, on-board counts and transport service) and provides an intuitive representation of the causal spatio-temporal relationships between flows. Its ability to deal with missing data allows to make real-time predictions even in case of technical failures or absences of collection systems

Les systèmes de prévision des crues décrivent les transformations pluie-débit en se basant sur des représentations simplifiées. Ces représentations modélisent les processus physiques impliqués avec des descriptions empiriques, ou basées sur des équations de la mécanique classique. Les performances des modèles actuels de prévision des crues sont affectées par différentes incertitudes liées aux approximations et aux paramètres du modèle, aux données d’entrée et aux conditions initiales du bassin versant. La connaissance de ces incertitudes permet aux décideurs de mieux interpréter les prévisions et constitue une aide à la décision lors de la gestion de crue. L’analyse d’incertitudes dans les modèles hydrologiques existants repose le plus souvent sur des simulations de Monte-Carlo (MC). La mise en œuvre de ce type de techniques requiert un grand nombre de simulations et donc un temps de calcul potentiellement important. L'estimation des incertitudes liées à la modélisation hydrologique en temps réel reste donc une gageure. Dans ce projet de thèse, nous développons une méthodologie de prévision des crues basée sur les réseaux Bayésiens (RB). Les RBs sont des graphes acycliques dans lesquels les nœuds correspondent aux variables caractéristiques du système modélisé et les arcs représentent les dépendances probabilistes entre ces variables. La méthodologie présentée propose de construire les RBs à partir des principaux facteurs hydrologiques contrôlant la génération des crues, en utilisant à la fois les observations disponibles de la réponse du système et les équations déterministes décrivant les processus concernés. Elle est conçue pour prendre en compte la variabilité temporelle des différentes variables impliquées. Les dépendances probabilistes entre les variables (paramètres) peuvent être spécifiées en utilisant des données observées, des modèles déterministes existants ou des avis d’experts. Grâce à leurs algorithmes d’inférence, les RBs sont capables de propager rapidement, à travers le graphe, différentes sources d'incertitudes pour estimer leurs effets sur la sortie du modèle (ex. débit d'une rivière). Plusieurs cas d’études sont testés. Le premier cas d’étude concerne le bassin versant du Salat au sud-ouest de la France : un RB est utilisé pour simuler le débit de la rivière à une station donnée à partir des observations de 3 stations hydrométriques localisées en amont. Le modèle présente de bonnes performances pour l'estimation du débit à l’exutoire. Utilisé comme méthode inverse, le modèle affiche également de bons résultats quant à la caractérisation de débits d’une station en amont par propagation d’observations de débit sur des stations en aval. Le deuxième cas d’étude concerne le bassin versant de la Sagelva situé en Norvège, pour lequel un RB est utilisé afin de modéliser l'évolution du contenu en eau de la neige en fonction des données météorologiques disponibles. Les performances du modèle sont conditionnées par les données d’apprentissage utilisées pour spécifier les paramètres du modèle. En l'absence de données d'observation pertinentes pour l’apprentissage, une méthodologie est proposée et testée pour estimer les paramètres du RB à partir d’un modèle déterministe. Le RB résultant peut être utilisé pour effectuer des analyses d’incertitudes sans recours aux simulations de Monte-Carlo. Au regard des résultats enregistrés sur les différents cas d’études, les RBs se révèlent utiles et performants pour une utilisation en support d’un processus d'aide à la décision dans le cadre de la gestion du risque de crue. / Flood forecasting describes the rainfall-runoff transformation using simplified representations. These representations are based on either empirical descriptions, or on equations of classical mechanics of the involved physical processes. The performances of the existing flood predictions are affected by several sources of uncertainties coming not only from the approximations involved but also from imperfect knowledge of input data, initial conditions of the river basin, and model parameters. Quantifying these uncertainties enables the decision maker to better interpret the predictions and constitute a valuable decision-making tool for flood risk management. Uncertainty analysis on existing rainfall-runoff models are often performed using Monte Carlo (MC)- simulations. The implementation of this type of techniques requires a large number of simulations and consequently a potentially important calculation time. Therefore, quantifying uncertainties of real-time hydrological models is challenging. In this project, we develop a methodology for flood prediction based on Bayesian networks (BNs). BNs are directed acyclic graphs where the nodes correspond to the variables characterizing the modelled system and the arcs represent the probabilistic dependencies between these variables. The presented methodology suggests to build the RBs from the main hydrological factors controlling the flood generation, using both the available observations of the system response and the deterministic equations describing the processes involved. It is, thus, designed to take into account the time variability of different involved variables. The conditional probability tables (parameters), can be specified using observed data, existing hydrological models or expert opinion. Thanks to their inference algorithms, BN are able to rapidly propagate, through the graph, different sources of uncertainty in order to estimate their effect on the model output (e.g. riverflow). Several case studies are tested. The first case study is the Salat river basin, located in the south-west of France, where a BN is used to simulate the discharge at a given station from the streamflow observations at 3 hydrometric stations located upstream. The model showed good performances estimating the discharge at the outlet. Used in a reverse way, the model showed also satisfactory results when characterising the discharges at an upstream station by propagating back discharge observations of some downstream stations. The second case study is the Sagelva basin, located in Norway, where a BN is used to simulate the accumulation of snow water equivalent (SWE) given available weather data observations. The performances of the model are affected by the learning dataset used to train the BN parameters. In the absence of relevant observation data for learning, a methodology for learning the BN-parameters from deterministic models is proposed and tested. The resulted BN can be used to perform uncertainty analysis without any MC-simulations to be performed in real-time. From these case studies, it appears that BNs are a relevant decisionsupport tool for flood risk management.