Mémoire numérisé par la Division de la gestion de documents et des archives de l'Université de Montréal.

Cette thèse traite du problème de la redistribution dynamique parallèle efficace de la charge pour les problèmes numériques de très grande taille. Nous présentons tout d'abord un état de l'art des algorithmes permettant de résoudre les problèmes du partitionnement, du repartitionnement, du placement statique et du re-placement. Notre première contribution vise à étudier, dans un cadre séquentiel, les caractéristiques algorithmiques souhaitables pour les méthodes parallèles de repartitionnement. Nous y présentons notre contribution à la conception d'un schéma multi-niveaux k-aire pour le calcul sequentiel de repartitionnements. La partie la plus exigeante de cette adaptation concerne la phase d'expansion. L'une de nos contributions majeures a été de nous inspirer des méthodes d'influence afin d'adapter un algorithme de raffinement par diffusion au problème du repartitionnement.Notre deuxième contribution porte sur la mise en oeuvre de ces méthodes sur machines parallèles. L'adaptation du schéma multi-niveaux parallèle a nécessité une évolution des algorithmes et des structures de données mises en oeuvre pour le partitionnement. Ce travail est accompagné d'une analyse expérimentale, qui est rendue possible grâce à la mise en oeuvre des algorithmes considérés au sein de la bibliothèque Scotch. / This thesis concerns efficient parallel dynamic load balancing for large scale numerical problems. First, we present a state of the art of the algorithms used to solve the partitioning, repartitioning, mapping and remapping problems. Our first contribution, in the context of sequential processing, is to define the desirable features that parallel repartitioning tools need to possess. We present our contribution to the conception of a k-way multilevel framework for sequential repartitioning. The most challenging part of this work regards the uncoarsening phase. One of our main contributions is the adaptation of influence methods to a global diffusion-based heuristic for the repartitioning problem. Our second contribution is the parallelization of these methods. The adaptation of the aforementioned algorithms required some modification of the algorithms and data structure used by existing parallel partitioning routines. This work is backed by a thorough experimental analysis, which is made possible thanks to the implementation of our algorithms into the Scotch library.

The present work describes the model of heuristic judgment of Kahneman & Frederick (2002) and two experiments based upon it. According to the model people answer a question with an answer to an easier question when making a heuristic judgment. This process is called the attribute substitution because a target attribute of a question is substituted by an associated and easier accessible heuristic attribute. The first experiment investigated whether two heuristic attributes can be used simultaneously during making of a judgment. A part of participants gave 1 or 4 reasons for one of the statements in Linda problem (Tversky & Kahneman, 1983). Numbers of reasons were selected so as to produce a feeling of fluency of disfluency. Although the conjunction fallacy occurred, the writing of the reasons didn't have any influence on the assessment of the probability of the related statement. The second experiment investigated whether the priming of the relation between processing fluency and risk can influence the effect of pronounceability of a food additive name on the assessment of its harmfulness. In accord with previous study (Song & Schwarz, 2009) it was shown that food additives with less pronounceable names were considered as more harmful. The priming didn't have any effect. The present studies...

Split delivery vehicle rating problem is one of the most studied combinatorial optimization problems in operations research. According to the mathematical difficultness, there should be many problems to find the optimal solution. Therefore, there are many exact algorithms and heuristics, which tries to find the best solution in the short period of time. The theoretical part of this thesis describes the basic facts of the split delivery vehicle routing problem and its heuristics. The practical part focuses on the practical usage of the split delivery vehicle routing problem. The main goals of this thesis are the practical usage of this vehicle routing problem and assistance in strategic decision establishing of the secondary store.

Cette thèse développe des algorithmes pour des problèmes de transport ferroviaire et est réalisée en partenariat avec l'entreprise Eurotunnel qui exploite le tunnel sous la Manche. Ce partenariat s'est établi sous la forme d'une chaire avec l'École des Ponts où cette thèse a été menée. Nous développons trois sujets dans cette thèse: le premier est un problème opérationnel rencontré par Eurotunnel, les deux autres sont plus prospectifs et théoriques, et sont inspirés des problèmes de transport ferroviaire d'Eurotunnel.Le processus de création de grilles horaires pour le transport ferroviaire se découpe en plusieurs phases (estimation de la demande, détermination du réseau, planification des départs, affectation des trains et du personnel). Nous nous intéressons dans une première partie à la phase de planification des départs des trains sur un intervalle temporel, appliquée au cas spécifique d'Eurotunnel. L'objectif est de calculer les horaires des départs des trains depuis chacune des deux stations (Coquelles en France et Folkestone en Angleterre) en respectant des contraintes d'exploitation (sécurité, chargement, ...) et des accords commerciaux signés avec leurs partenaires (Eurostar, ...). De plus, la prise en compte des retards dès la planification des départs est primordiale pour limiter la propagation des perturbations de train en train sur le réseau. Nous avons développé des algorithmes de planification pour Eurotunnel tenant compte des contraintes du réseau et de la probabilité de retard pour chaque train. Ces algorithmes utilisent des outils standard de la Recherche Opérationnelle pour modéliser et résoudre ces problèmes d'optimisation.La tarification des billets est un enjeu majeur pour les entreprises de transport. Pour les compagnies aériennes, de nombreux algorithmes ont été étudiés pour définir le prix optimal des billets pour différentes classes de passagers. Nous appliquons dans une deuxième partie des méthodes standard de tarification (modèles de choix discrets) afin d'optimiser de manière globale les prix et les horaires des départs pour des entreprises de transport ferroviaire. Des outils classiques de l'optimisation stochastique, des modèles de choix discrets et des heuristiques sont utilisés dans nos algorithmes pour donner les meilleures solutions possibles en un temps de calcul limité.Nous nous intéressons dans une dernière partie à une classe de problèmes de transport, inspirés de ceux rencontrés par Eurotunnel, en donnant des algorithmes efficaces de résolution exacte ou approchée. Ces algorithmes permettent de donner une borne supérieure de la complexité temporelle de ces problèmes. La classe de problèmes étudiés consiste en la planification des départs de navettes sur une ligne fixe, pour transporter d'une station A vers une station B des usagers arrivant de manière continue. Les navettes sont éventuellement autorisées à faire de multiples rotations pour transporter plusieurs vagues d'usagers. L'objectif est de limiter le temps d'attente des passagers avant le départ de leur navette. Des combinaisons originales de l'optimisation convexe et de la théorie des graphes (problèmes de plus court chemin) sont utilisées dans nos algorithmes / This thesis develops algorithms for rail transportation problems, conducted in relationship with the company Eurotunnel which operates the tunnel under the Channel. This partnership is a scientific chair with the École des Ponts et Chaussées, where this thesis was realized. We study three topics throughout the thesis: the first one is an operational problem faced by Eurotunnel, whereas the two other ones are prospective and theoretical problems inspired by their process.The planning process for rail transportation can be divided into several phases (demand estimation, line planning, scheduling of the departure times, rolling stock and crew planning). In a first part, we focus on the scheduling phase on a time interval, applied to the specific case of Eurotunnel. The objective is to compute the departure times of the trains for each of the two stations (Calais in France and Folkestone in England), satisfying operation constraints (security, loading, ...) and commercial agreements with their partners (Eurostar, ...). Moreover, taking into account the delays in the scheduling phase is essential to limit the propagation of the disturbances from train to train in the network. We develop scheduling algorithms for Eurotunnel taking into account the operation and commercial constraints, and the random distributions of the delays for each train. These algorithms use standard tools of Operations Research to model and solve these optimization problems.Pricing is a main issue for transportation companies. Many algorithms have been proposed to help airline companies to define optimized prices of the plane tickets for different classes of passengers. In a second part, we apply some standard pricing frameworks (discrete choice models) in order to optimize in a global way the prices and the departure times of the trains for rail transportation companies. Standard tools of stochastic optimization, discrete choice models, and some heuristics are used in our algorithms to compute the best possible solutions in a limited computation time.We focus in a last part on a class of transportation problems, inspired form Eurotunnel. We give efficient algorithms to solve exactly or to approximate the optimal solutions of these problems. These algorithms give an upper bound of the time complexity of this class of problems. The problems studied consist in scheduling the departure times of shuttles on a fixed trip, to transport passengers, arriving continuously at an initial station, to a given destination. The shuttles are potentially allowed to perform several rotations to transport several groups of passengers. The objective is to minimize the waiting time of the passengers before the depart of their shuttle. Original combinations of convex optimization and graph theory (shortest path problems) are used in our algorithms

[pt] A importância do planejamento em logística ferroviária cresce a cada dia devido ao alto custo dos investimentos para o aumento da sua capacidade. Entretanto, planejar é uma atividade que exige uma representação suficientemente precisa da realidade estudada. Neste contexto, os modelos de programação matemática apresentam-se cada vez mais adequados. Isto decorre dos recentes avanços nos algoritmos e computadores disponíveis para sua resolução. Esta dissertação apresenta modelos e algoritmos para o planejamento ferroviário tático e estratégico, isto é feito estudando o Problema de Planejamento de Atendimento (PPA). Primeiramente este problema é considerado assumindo que toda a estrutura ferroviária está definida: a malha, a tração e os vagões disponíveis, os pátios para carga, descarga e transbordo, suas respectivas taxas de carga e descarga e as demandas previstas. Em seguida, a questão adicional de determinar os pátios onde paradas podem ser efetuadas é considerada. Finalmente, em uma terceira etapa, introduz-se a capacidade de se analisar os efeitos do congestionamento de trechos da malha e seu impacto nos tempos de circulação e na capacidade da estrutura logística. Modelos são apresentados para cada um dos níveis de complexidade do PPA. Algoritmos exatos e heurísticos e técnicas de pré- processamento, foram desenvolvidos para os tratamentos dos casos obtidos. Em todos os casos, foi possível resolver de maneira ótima ou quase ótima em tempo razoável, tanto em termos acadêmicos, como para a utilização prática. Resultados computacionais sobre um amplo conjunto de instâncias reais são apresentados. / [en] Planning in Railway Logistic is an activity with growing importance. This is due to the high costs of investment to increase the railway capacity. Nevertheless, planning in this context is a cumbersome task, since a precise representation is necessary to consider most relevant points in this activity. Mathematical programming is becoming one of the best ways derive precise representations and to solve them. This is due to the recent advances on algorithms and computers used in the resolution of mathematical programming problems. This dissertation presents models and algorithms for tactical and strategical railway planning what is done by studying a demand planning problem (PPA). First, this problem is considered assuming that all the railway structure is defined: the network, the locomotives and wagons available, the yards for loading and unloading with their respective rates, and the forecast of demands. Next, the question of deciding the yards to stop is considered. Finally, in a third step, the effect of congestion in parts of the network is introduced to the models. This allows analyzing the variation in the travel times and its consequence in the logistic structure capacity. Models are presented for all cases of the PPA. Exact and heuristic algorithms, as well as pre-processing techniques, are described for the problem resolution. In all cases, the resulting approach allowed to solve the problems optimally or quasioptimally in a reasonable computing time. Computational results are presented on a wide set of real world instances.

Allt fler beslut tas i samband med skärmar, bland annat på grund av den ständigt växande e-handeln. När människor tar beslut påverkas de av hur valmiljön är utformad samt av heuristikeroch fördomar, vilka är mentala genvägar. Människor influeras av heuristiker och fördomar dådet ofta är svårt för dem att ta beslut som grundas på all information som presenteras. Knuffarär ett sätt att vägleda människor till att göra specifika val utan att sätta begränsningar för derasfrihet och har företrädesvis använts i icke-digitala kontexter, men har nu även introduceratsinom informatik. För att knuffar ska kunna utvecklas och användas digitalt är det dock viktigtmed kunskap om hur människor gör val samt om den påverkan knuffar har. Även tidigareforskning om digitala knuffar efterfrågar studier om hur digitala knuffar kan designas och omderas påverkan på människors val online. Då knuffar är ett relativt nytt begrepp inom informatik, och att fler studier om ämnetefterfrågas, undersöker denna studie hur digitala knuffar kan designas för att vägledamänniskor till hållbara val på nätet. Studien utfördes med en designorienterad forskningsansatsdär digitala knuffar designades utifrån fyra olika heuristiker och fördomar. Knuffarnaapplicerades sedan på en e-handelssida och testades med hjälp av en prototyp. Processen fördesign och utvärdering av de digitala knuffarna itererades fyra eller fem gånger för attexplorativt testa olika sätt att designa knuffarna på relaterat till forskningsfrågan. Studien bidrarmed ett antal förslag för hur digitala knuffar kan designas utifrån heuristiker och fördomar medfokus på vägledning och hållbarhet. / Decisions are increasingly made in connection with screens, amongst other things due to thecontinuous growth of e-commerce. When people make decisions they are influenced by thechoice environment as well as heuristics and biases, which are mental shortcuts. People areinfluenced by heuristics and biases as it is often difficult for them to make decisions based onall of the information presented. Nudges are a way to guide people to make specific choiceswithout limiting their freedom and have primarily been used in non-digital contexts, but havenow also been introduced in informatics. However, in order for nudges to be able to bedeveloped and used digitally it is important to have knowledge about how people make choicesand the effect nudges have. Previous research on digital nudges also asks for studies on howdigital nudges can be designed and their impact on people's choices online.As nudging is a relatively new term in informatics, as well as research regarding the topic isinquired, this study investigate how digital nudges can be designed to guide people tosustainable choices online. The study was conducted with a design research approach wheredigital nudges were designed based on four different heuristics and biases. The nudges wereapplied on an e-commerce website and tested by means of a prototype. The process fordesigning and evaluating the digital nudges was iterated four or five times to in an exploratoryway test different ways to design the nudges in relation to the research question. The studycontribute with a number of suggestions on how digital nudges can be designed based onheuristics and biases, with guidance and sustainability as focal points.

Este trabalho trata do problema de programação de operações no ambiente flow shop com máquinas múltiplas, com seus tempos de preparação (setup) assimétricos e dependentes da seqüência de processamento das tarefas. Este ambiente de produção é comum em indústrias gráficas, químicas, têxteis, de papel e de tinta, caracterizadas por sistemas com amplo mix de produtos. Qualquer processo produtivo requer um gerenciamento eficaz por meio do Planejamento e Controle da Produção (PCP). Esta atividade inclui a programação da produção, ou seja, a alocação de recursos para a execução de tarefas em uma base de tempo. A atividade de programação é uma das tarefas mais complexas no gerenciamento de produção, pois há a necessidade de lidar com diversos tipos diferentes de recursos e atividades simultaneamente. Além disso, o número de soluções possíveis cresce exponencialmente em várias dimensões, de acordo com a quantidade de tarefas, operações ou máquinas, conferindo uma natureza combinatorial ao problema. No ambiente estudado neste trabalho as operações de cada tarefa são executadas em múltiplos estágios de produção, podendo variar a quantidade de máquinas em cada um deles. Cada operação é processada por apenas uma máquina em cada estágio. Os tempos de preparação das máquinas possuem uma variabilidade relevante em função da ordem de execução das tarefas nas máquinas. A função-objetivo considerada é a minimização da duração total da programação (makespan). Foram desenvolvidos quatro métodos heurísticos construtivos com base em algoritmos reportados na literatura para solução de problemas flow shop permutacional e máquinas paralelas no ambiente cujo tempo de setup é dependente da seqüência. Como não foram encontrados na literatura métodos para programação no ambiente tratado neste trabalho, os algoritmos construídos foram comparados entre si. O foco da pesquisa foi o estudo da influência da relação entre as ordens de grandeza dos tempos de processamento e de setup em cada método de solução. Os resultados obtidos na experimentação computacional foram analisados e discutidos com base na porcentagem de sucesso, desvio relativo (%), desvio-padrão do desvio relativo e tempo médio de computação / This work adressess the hybrid flow shop scheduling problem with asymmetric sequence dependent setup times. This environment of production system is common in graphical, chemical, fabric, paper and ink industries. It’s characterized by systems with large mix of products. Any productive process requires an efficient management by means of Production Planning and Control. This activity includes scheduling, i.e., the resources allocation for the execution of jobs in a time base. Scheduling is one of the tasks most complex in production management, since it deals simultaneously with different types of resources and activities. Moreover, the number of possible solutions grows exponentially in some dimensions, in accordance with the number of jobs, operations or machines, conferring a combinatorial nature to the problem. In the environment studied in this work, the operations of each job are processed in multiple production stages. The number of machines in each stage can be different. Each operation is processed by only one machine in each stage. The setup times have a significant variability in function of the sequence of job processing on the machines. The objective is minimizing the total time to complete the schedule (makespan). Four constructive heuristic methods were developed on the basis of algorithms reported in the literature for solving permutation flow shop and parallel machine problems with sequence dependent setup times. The proposed heuristic methods have been compared between themselves, since no constructive heuristics have been found in the literature for the scheduling problem considered in this work. The focus of the research was the study of the influence of the relations among the range of the times processing and setup times in each method. The statistics used in order to evaluate the heuristic performances were the percentage of success (in finding the best solution), relative deviation, standard deviation of relative deviation and average computation time. Results from computational experience are discussed

Este trabalho aborda um problema ainda não explorado na literatura denominado 3L-FSMVRPTW (three-dimensional loading fleet sizing and mix vehicle routing problem with time windows), que compreende resolver simultaneamente o roteamento e carregamento tridimensional de veículos considerando frota heterogênea e janela de tempo. Foi desenvolvido um algoritmo específico para resolver o problema, denominado 3DC. Neste algoritmo foram introduzidas algumas inovações, entre elas, um novo operador de busca local (k-IntensiveSwap) e uma nova heurística de carregamento de contêiner. O algoritmo foi comparado aos melhores resultados disponíveis na literatura para problemas particulares ao apresentado. Houve bom desempenho no caso do CLP (container loading problem), bom resultado na redução do tamanho de frota no caso do 3L-VRP (threedimensional loading vehicle routing problem) e desempenho superior ao problema mais complexo estudado, o 3L-VRPTW (three-dimensional loading vehicle routing problem with time windows). Finalmente, apresentou-se um conjunto de avaliação, instâncias e soluções, para o problema completo com frota heterogênea e janela de tempo. / This work presents a problem not treated yet on the literature referenced as 3L-FSMVRPTW (three-dimensional loading fleet sizing and mix vehicle routing problem with time windows), which deals simultaneously with vehicle routing and its three-dimensional loading considering heterogeneous fleet and time windows. The algorithm developed for the specific problem is called 3DC. This algorithm introduces a new local search operator called k-IntensiveSwap and a new container loading heuristic. The results are compared with the best-known results from literature for particular problems embeeded on the general problem presented. The quality of solution was good in comparison other methods for CLP (container loading problem), it has good results in terms of reduction fleet sizing in the case of 3L-VRP (three-dimensional loading vehicle routing problem) and as for 3L-VRPTW (threedimensional loading vehicle routing problem with time windows) the performance was very superior. Finally, it is presented a solution set as benchmark for future comparison with the general problem, with heterogeneous fleet.

Os processos industriais de produção exigem uma programação da produção efetiva. Essa atividade consiste da alocação dos recursos produtivos, a fim de executar tarefas determinadas por um período de tempo definido. Programar a produção é uma das atividades mais complexas do Planejamento da Produção, pois existem diferentes tipos de recursos a serem administrados simultaneamente. E também a quantidade de possíveis soluções aumenta exponencialmente com o aumento da quantidade de tarefas e máquinas presentes no sistema. A proposta deste trabalho é apresentar um método heurístico construtivo para a solução de problemas flow shop permutacional. A função-objetivo utilizada é a minimização do tempo total da programação (makespan). O algoritmo foi desenvolvido com base no melhor algoritmo construtivo presente na literatura, e os resultados obtidos são discutidos e analisados com base na porcentagem de sucesso, desvio relativo médio e tempo médio de computação. / Industrial productive processes demand an effective production scheduling. These activities consist in allocating the productive resources in order to execute determined jobs for a established period of time. Scheduling the production is one of the most complex activities involved in Planning the Production because there are different kinds of resources to be managed simultaneously. Furthermore, the amounts of feasible solutions increase exponentially as the number of jobs and machines in large systems. This dissertation presents a constructive heuristic method to solve the permutational flow shop problem. The evaluation criterion is the total production elapsed time (makespan). The developed algorithm was based on the best algorithm found in the literature, the results are analysed based on the success rate, mean relative deviation and computing time.