The crew-scheduling module in the GIST system

The public transportation is gaining importance every year basically duethe population growth, environmental policies and, route and streetcongestion. Too able an efficient management of all the resources relatedto public transportation, several techniques from different areas are beingapplied and several projects in Transportation Planning Systems, indifferent countries, are being developed. In this work, we present theGIST Planning Transportation Systems, a Portuguese project involving twouniversities and six public transportation companies. We describe indetail one of the most relevant modules of this project, the crew-scheduling module. The crew-scheduling module is based on the application of meta-heuristics, in particular GRASP, tabu search and geneticalgorithm to solve the bus-driver-scheduling problem. The metaheuristicshave been successfully incorporated in the GIST Planning TransportationSystems and are actually used by several companies.

On optimal location with treshold requirements

The optimal location of services is one of the most important factors that affects service quality in terms of consumer access. On theother hand, services in general need to have a minimum catchment area so as to be efficient. In this paper a model is presented that locates the maximum number of services that can coexist in a given region without having losses, taking into account that they need a minimum catchment area to exist. The objective is to minimize average distance to the population. The formulation presented belongs to the class of discrete P--median--like models. A tabu heuristic method is presented to solve the problem. Finally, the model is applied to the location of pharmacies in a rural region of Spain.

Location of hubs in a competitive environment

We offer a formulation that locates hubs on a network in a competitiveenvironment; that is, customer capture is sought, which happenswhenever the location of a new hub results in a reduction of thecurrent cost (time, distance) needed by the traffic that goes from thespecified origin to the specified destination.The formulation presented here reduces the number of variables andconstraints as compared to existing covering models. This model issuited for both air passenger and cargo transportation.In this model, each origin-destination flow can go through either oneor two hubs, and each demand point can be assigned to more than a hub,depending on the different destinations of its traffic. Links(``spokes'' have no capacity limit. Computational experience is provided.

Adaptive approach heuristics for the generalized assignment problem

The Generalized Assignment Problem consists in assigning a setof tasks to a set of agents with minimum cost. Each agent hasa limited amount of a single resource and each task must beassigned to one and only one agent, requiring a certain amountof the resource of the agent. We present new metaheuristics forthe generalized assignment problem based on hybrid approaches.One metaheuristic is a MAX-MIN Ant System (MMAS), an improvedversion of the Ant System, which was recently proposed byStutzle and Hoos to combinatorial optimization problems, and itcan be seen has an adaptive sampling algorithm that takes inconsideration the experience gathered in earlier iterations ofthe algorithm. Moreover, the latter heuristic is combined withlocal search and tabu search heuristics to improve the search.A greedy randomized adaptive search heuristic (GRASP) is alsoproposed. Several neighborhoods are studied, including one basedon ejection chains that produces good moves withoutincreasing the computational effort. We present computationalresults of the comparative performance, followed by concludingremarks and ideas on future research in generalized assignmentrelated problems.

A new chance-constrained maximum capture location problem

The paper presents a new model based on the basic Maximum Capture model,MAXCAP. The New Chance Constrained Maximum Capture modelintroduces astochastic threshold constraint, which recognises the fact that a facilitycan be open only if a minimum level of demand is captured. A metaheuristicbased on MAX MIN ANT system and TABU search procedure is presented tosolve the model. This is the first time that the MAX MIN ANT system isadapted to solve a location problem. Computational experience and anapplication to 55 node network are also presented.

Consumer choice in competitive location models: Formulations and heuristics

A new direction of research in Competitive Location theory incorporatestheories of Consumer Choice Behavior in its models. Following thisdirection, this paper studies the importance of consumer behavior withrespect to distance or transportation costs in the optimality oflocations obtained by traditional Competitive Location models. To dothis, it considers different ways of defining a key parameter in thebasic Maximum Capture model (MAXCAP). This parameter will reflectvarious ways of taking into account distance based on several ConsumerChoice Behavior theories. The optimal locations and the deviation indemand captured when the optimal locations of the other models are usedinstead of the true ones, are computed for each model. A metaheuristicbased on GRASP and Tabu search procedure is presented to solve all themodels. Computational experience and an application to 55-node networkare also presented.

Location models for airline hubs behaving as M/D/c queues

Models are presented for the optimal location of hubs in airline networks, that take into consideration the congestion effects. Hubs, which are the most congested airports, are modeled as M/D/c queuing systems, that is, Poisson arrivals, deterministic service time, and {\em c} servers. A formula is derived for the probability of a number of customers in the system, which is later used to propose a probabilistic constraint. This constraint limits the probability of {\em b} airplanes in queue, to be lesser than a value $\alpha$. Due to the computational complexity of the formulation. The model is solved using a meta-heuristic based on tabu search. Computational experience is presented.

Differential evolution based classification with pool of distances and aggregation operators

The objective of this thesis is to develop and generalize further the differential evolution based data classification method. For many years, evolutionary algorithms have been successfully applied to many classification tasks. Evolution algorithms are population based, stochastic search algorithms that mimic natural selection and genetics. Differential evolution is an evolutionary algorithm that has gained popularity because of its simplicity and good observed performance. In this thesis a differential evolution classifier with pool of distances is proposed, demonstrated and initially evaluated. The differential evolution classifier is a nearest prototype vector based classifier that applies a global optimization algorithm, differential evolution, to determine the optimal values for all free parameters of the classifier model during the training phase of the classifier. The differential evolution classifier applies the individually optimized distance measure for each new data set to be classified is generalized to cover a pool of distances. Instead of optimizing a single distance measure for the given data set, the selection of the optimal distance measure from a predefined pool of alternative measures is attempted systematically and automatically. Furthermore, instead of only selecting the optimal distance measure from a set of alternatives, an attempt is made to optimize the values of the possible control parameters related with the selected distance measure. Specifically, a pool of alternative distance measures is first created and then the differential evolution algorithm is applied to select the optimal distance measure that yields the highest classification accuracy with the current data. After determining the optimal distance measures for the given data set together with their optimal parameters, all determined distance measures are aggregated to form a single total distance measure. The total distance measure is applied to the final classification decisions. The actual classification process is still based on the nearest prototype vector principle; a sample belongs to the class represented by the nearest prototype vector when measured with the optimized total distance measure. During the training process the differential evolution algorithm determines the optimal class vectors, selects optimal distance metrics, and determines the optimal values for the free parameters of each selected distance measure. The results obtained with the above method confirm that the choice of distance measure is one of the most crucial factors for obtaining higher classification accuracy. The results also demonstrate that it is possible to build a classifier that is able to select the optimal distance measure for the given data set automatically and systematically. After finding optimal distance measures together with optimal parameters from the particular distance measure results are then aggregated to form a total distance, which will be used to form the deviation between the class vectors and samples and thus classify the samples. This thesis also discusses two types of aggregation operators, namely, ordered weighted averaging (OWA) based multi-distances and generalized ordered weighted averaging (GOWA). These aggregation operators were applied in this work to the aggregation of the normalized distance values. The results demonstrate that a proper combination of aggregation operator and weight generation scheme play an important role in obtaining good classification accuracy. The main outcomes of the work are the six new generalized versions of previous method called differential evolution classifier. All these DE classifier demonstrated good results in the classification tasks.

Feature Selection and Classification Using Age Layered Population Structure Genetic Programming

The curse of dimensionality is a major problem in the fields of machine learning, data mining and knowledge discovery. Exhaustive search for the most optimal subset of relevant features from a high dimensional dataset is NP hard. Sub–optimal population based stochastic algorithms such as GP and GA are good choices for searching through large search spaces, and are usually more feasible than exhaustive and deterministic search algorithms. On the other hand, population based stochastic algorithms often suffer from premature convergence on mediocre sub–optimal solutions. The Age Layered Population Structure (ALPS) is a novel metaheuristic for overcoming the problem of premature convergence in evolutionary algorithms, and for improving search in the fitness landscape. The ALPS paradigm uses an age–measure to control breeding and competition between individuals in the population. This thesis uses a modification of the ALPS GP strategy called Feature Selection ALPS (FSALPS) for feature subset selection and classification of varied supervised learning tasks. FSALPS uses a novel frequency count system to rank features in the GP population based on evolved feature frequencies. The ranked features are translated into probabilities, which are used to control evolutionary processes such as terminal–symbol selection for the construction of GP trees/sub-trees. The FSALPS metaheuristic continuously refines the feature subset selection process whiles simultaneously evolving efficient classifiers through a non–converging evolutionary process that favors selection of features with high discrimination of class labels. We investigated and compared the performance of canonical GP, ALPS and FSALPS on high–dimensional benchmark classification datasets, including a hyperspectral image. Using Tukey’s HSD ANOVA test at a 95% confidence interval, ALPS and FSALPS dominated canonical GP in evolving smaller but efficient trees with less bloat expressions. FSALPS significantly outperformed canonical GP and ALPS and some reported feature selection strategies in related literature on dimensionality reduction.

Feature Selection and Classification Using Age Layered Population Structure Genetic Programming

The curse of dimensionality is a major problem in the fields of machine learning, data mining and knowledge discovery. Exhaustive search for the most optimal subset of relevant features from a high dimensional dataset is NP hard. Sub–optimal population based stochastic algorithms such as GP and GA are good choices for searching through large search spaces, and are usually more feasible than exhaustive and determinis- tic search algorithms. On the other hand, population based stochastic algorithms often suffer from premature convergence on mediocre sub–optimal solutions. The Age Layered Population Structure (ALPS) is a novel meta–heuristic for overcoming the problem of premature convergence in evolutionary algorithms, and for improving search in the fitness landscape. The ALPS paradigm uses an age–measure to control breeding and competition between individuals in the population. This thesis uses a modification of the ALPS GP strategy called Feature Selection ALPS (FSALPS) for feature subset selection and classification of varied supervised learning tasks. FSALPS uses a novel frequency count system to rank features in the GP population based on evolved feature frequencies. The ranked features are translated into probabilities, which are used to control evolutionary processes such as terminal–symbol selection for the construction of GP trees/sub-trees. The FSALPS meta–heuristic continuously refines the feature subset selection process whiles simultaneously evolving efficient classifiers through a non–converging evolutionary process that favors selection of features with high discrimination of class labels. We investigated and compared the performance of canonical GP, ALPS and FSALPS on high–dimensional benchmark classification datasets, including a hyperspectral image. Using Tukey’s HSD ANOVA test at a 95% confidence interval, ALPS and FSALPS dominated canonical GP in evolving smaller but efficient trees with less bloat expressions. FSALPS significantly outperformed canonical GP and ALPS and some reported feature selection strategies in related literature on dimensionality reduction.

Routage adaptatif et qualité de service dans les réseaux optiques à commutation de rafales

Les réseaux optiques à commutation de rafales (OBS) sont des candidats pour jouer un rôle important dans le cadre des réseaux optiques de nouvelle génération. Dans cette thèse, nous nous intéressons au routage adaptatif et au provisionnement de la qualité de service dans ce type de réseaux. Dans une première partie de la thèse, nous nous intéressons à la capacité du routage multi-chemins et du routage alternatif (par déflection) à améliorer les performances des réseaux OBS, pro-activement pour le premier et ré-activement pour le second. Dans ce contexte, nous proposons une approche basée sur l’apprentissage par renforcement où des agents placés dans tous les nœuds du réseau coopèrent pour apprendre, continuellement, les chemins du routage et les chemins alternatifs optimaux selon l’état actuel du réseau. Les résultats numériques montrent que cette approche améliore les performances des réseaux OBS comparativement aux solutions proposées dans la littérature. Dans la deuxième partie de cette thèse, nous nous intéressons au provisionnement absolu de la qualité de service où les performances pire-cas des classes de trafic de priorité élevée sont garanties quantitativement. Plus spécifiquement, notre objectif est de garantir la transmission sans pertes des rafales de priorité élevée à l’intérieur du réseau OBS tout en préservant le multiplexage statistique et l’utilisation efficace des ressources qui caractérisent les réseaux OBS. Aussi, nous considérons l’amélioration des performances du trafic best effort. Ainsi, nous proposons deux approches : une approche basée sur les nœuds et une approche basée sur les chemins. Dans l’approche basée sur les nœuds, un ensemble de longueurs d’onde est assigné à chaque nœud du bord du réseau OBS pour qu’il puisse envoyer son trafic garanti. Cette assignation prend en considération les distances physiques entre les nœuds du bord. En outre, nous proposons un algorithme de sélection des longueurs d’onde pour améliorer les performances des rafales best effort. Dans l’approche basée sur les chemins, le provisionnement absolu de la qualité de service est fourni au niveau des chemins entre les nœuds du bord du réseau OBS. À cette fin, nous proposons une approche de routage et d’assignation des longueurs d’onde qui a pour but la réduction du nombre requis de longueurs d’onde pour établir des chemins sans contentions. Néanmoins, si cet objectif ne peut pas être atteint à cause du nombre limité de longueurs d’onde, nous proposons de synchroniser les chemins en conflit sans le besoin pour des équipements additionnels. Là aussi, nous proposons un algorithme de sélection des longueurs d’onde pour les rafales best effort. Les résultats numériques montrent que l’approche basée sur les nœuds et l’approche basée sur les chemins fournissent le provisionnement absolu de la qualité de service pour le trafic garanti et améliorent les performances du trafic best effort. En outre, quand le nombre de longueurs d’ondes est suffisant, l’approche basée sur les chemins peut accommoder plus de trafic garanti et améliorer les performances du trafic best effort par rapport à l’approche basée sur les nœuds.

Parallel metaheuristics for stochastic capacitated multicommodity network design

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

Meta-heuristic Solution Methods for Rich Vehicle Routing Problems

Le problème de tournées de véhicules (VRP), introduit par Dantzig and Ramser en 1959, est devenu l'un des problèmes les plus étudiés en recherche opérationnelle, et ce, en raison de son intérêt méthodologique et de ses retombées pratiques dans de nombreux domaines tels que le transport, la logistique, les télécommunications et la production. L'objectif général du VRP est d'optimiser l'utilisation des ressources de transport afin de répondre aux besoins des clients tout en respectant les contraintes découlant des exigences du contexte d’application. Les applications réelles du VRP doivent tenir compte d’une grande variété de contraintes et plus ces contraintes sont nombreuse, plus le problème est difficile à résoudre. Les VRPs qui tiennent compte de l’ensemble de ces contraintes rencontrées en pratique et qui se rapprochent des applications réelles forment la classe des problèmes ‘riches’ de tournées de véhicules. Résoudre ces problèmes de manière efficiente pose des défis considérables pour la communauté de chercheurs qui se penchent sur les VRPs. Cette thèse, composée de deux parties, explore certaines extensions du VRP vers ces problèmes. La première partie de cette thèse porte sur le VRP périodique avec des contraintes de fenêtres de temps (PVRPTW). Celui-ci est une extension du VRP classique avec fenêtres de temps (VRPTW) puisqu’il considère un horizon de planification de plusieurs jours pendant lesquels les clients n'ont généralement pas besoin d’être desservi à tous les jours, mais plutôt peuvent être visités selon un certain nombre de combinaisons possibles de jours de livraison. Cette généralisation étend l'éventail d'applications de ce problème à diverses activités de distributions commerciales, telle la collecte des déchets, le balayage des rues, la distribution de produits alimentaires, la livraison du courrier, etc. La principale contribution scientifique de la première partie de cette thèse est le développement d'une méta-heuristique hybride dans la quelle un ensemble de procédures de recherche locales et de méta-heuristiques basées sur les principes de voisinages coopèrent avec un algorithme génétique afin d’améliorer la qualité des solutions et de promouvoir la diversité de la population. Les résultats obtenus montrent que la méthode proposée est très performante et donne de nouvelles meilleures solutions pour certains grands exemplaires du problème. La deuxième partie de cette étude a pour but de présenter, modéliser et résoudre deux problèmes riches de tournées de véhicules, qui sont des extensions du VRPTW en ce sens qu'ils incluent des demandes dépendantes du temps de ramassage et de livraison avec des restrictions au niveau de la synchronization temporelle. Ces problèmes sont connus respectivement sous le nom de Time-dependent Multi-zone Multi-Trip Vehicle Routing Problem with Time Windows (TMZT-VRPTW) et de Multi-zone Mult-Trip Pickup and Delivery Problem with Time Windows and Synchronization (MZT-PDTWS). Ces deux problèmes proviennent de la planification des opérations de systèmes logistiques urbains à deux niveaux. La difficulté de ces problèmes réside dans la manipulation de deux ensembles entrelacés de décisions: la composante des tournées de véhicules qui vise à déterminer les séquences de clients visités par chaque véhicule, et la composante de planification qui vise à faciliter l'arrivée des véhicules selon des restrictions au niveau de la synchronisation temporelle. Auparavant, ces questions ont été abordées séparément. La combinaison de ces types de décisions dans une seule formulation mathématique et dans une même méthode de résolution devrait donc donner de meilleurs résultats que de considérer ces décisions séparément. Dans cette étude, nous proposons des solutions heuristiques qui tiennent compte de ces deux types de décisions simultanément, et ce, d'une manière complète et efficace. Les résultats de tests expérimentaux confirment la performance de la méthode proposée lorsqu’on la compare aux autres méthodes présentées dans la littérature. En effet, la méthode développée propose des solutions nécessitant moins de véhicules et engendrant de moindres frais de déplacement pour effectuer efficacement la même quantité de travail. Dans le contexte des systèmes logistiques urbains, nos résultats impliquent une réduction de la présence de véhicules dans les rues de la ville et, par conséquent, de leur impact négatif sur la congestion et sur l’environnement.

Meshing highly regular structures: The case of super carbon nanotubes of arbitrary order

Mesh generation is an important step inmany numerical methods.We present the “HierarchicalGraphMeshing” (HGM)method as a novel approach to mesh generation, based on algebraic graph theory.The HGM method can be used to systematically construct configurations exhibiting multiple hierarchies and complex symmetry characteristics. The hierarchical description of structures provided by the HGM method can be exploited to increase the efficiency of multiscale and multigrid methods. In this paper, the HGMmethod is employed for the systematic construction of super carbon nanotubes of arbitrary order, which present a pertinent example of structurally and geometrically complex, yet highly regular, structures. The HGMalgorithm is computationally efficient and exhibits good scaling characteristics. In particular, it scales linearly for super carbon nanotube structures and is working much faster than geometry-based methods employing neighborhood search algorithms. Its modular character makes it conducive to automatization. For the generation of a mesh, the information about the geometry of the structure in a given configuration is added in a way that relates geometric symmetries to structural symmetries. The intrinsically hierarchic description of the resulting mesh greatly reduces the effort of determining mesh hierarchies for multigrid and multiscale applications and helps to exploit symmetry-related methods in the mechanical analysis of complex structures.

A customizable multi-agent system for distributed data mining

We present a general Multi-Agent System framework for distributed data mining based on a Peer-to-Peer model. Agent protocols are implemented through message-based asynchronous communication. The framework adopts a dynamic load balancing policy that is particularly suitable for irregular search algorithms. A modular design allows a separation of the general-purpose system protocols and software components from the specific data mining algorithm. The experimental evaluation has been carried out on a parallel frequent subgraph mining algorithm, which has shown good scalability performances.