Travel-time information is essential in Advanced Traveler Information Systems (ATISs) and Advanced Traffic Management Systems (ATMSs). A key component of these systems is the prediction of the spatiotemporal evolution of roadway traffic state and travel time. From the perspective of travelers, such information can result in better traveler route choice and departure time decisions. From the transportation agency perspective, such data provide enhanced information with which to better manage and control the transportation system to reduce congestion, enhance safety, and reduce the carbon footprint of the transportation system. The objective of the research presented in this dissertation is to develop a framework that includes three major categories of methodologies to predict the spatiotemporal evolution of the traffic state. The proposed methodologies include macroscopic traffic modeling, computer vision and recursive probabilistic algorithms. Each developed method attempts to predict traffic state, including roadway travel times, for different prediction horizons. In total, the developed multi-tool framework produces traffic state prediction algorithms ranging from short – (0~5 minutes) to medium-term (1~4 hours) considering departure times up to an hour into the future. The dissertation first develops a particle filter approach for use in short-term traffic state prediction. The flow continuity equation is combined with the Van Aerde fundamental diagram to derive a time series model that can accurately describe the spatiotemporal evolution of traffic state. The developed model is applied within a particle filter approach to provide multi-step traffic state prediction. The testing of the algorithm on a simulated section of I-66 demonstrates that the proposed algorithm can accurately predict the propagation of shockwaves up to five minutes into the future. The developed algorithm is further improved by incorporating on- and off-ramp effects and more realistic boundary conditions. Furthermore, the case study demonstrates that the improved algorithm produces a 50 percent reduction in the prediction error compared to the classic LWR density formulation. Considering the fact that the prediction accuracy deteriorates significantly for longer prediction horizons, historical data are integrated and considered in the measurement update in the developed particle filter approach to extend the prediction horizon up to half an hour into the future. The dissertation then develops a travel time prediction framework using pattern recognition techniques to match historical data with real-time traffic conditions. The Euclidean distance is initially used as the measure of similarity between current and historical traffic patterns. This method is further improved using a dynamic template matching technique developed as part of this research effort. Unlike previous approaches, which use fixed template sizes, the proposed method uses a dynamic template size that is updated each time interval based on the spatiotemporal shape of the congestion upstream of a bottleneck. In addition, the computational cost is reduced using a Fast Fourier Transform instead of a Euclidean distance measure. Subsequently, the historical candidates that are similar to the current conditions are used to predict the experienced travel times. Test results demonstrate that the proposed dynamic template matching method produces significantly better and more stable prediction results for prediction horizons up to 30 minutes into the future for a two hour trip (prediction horizon of two and a half hours) compared to other state-of-the-practice and state-of-the-art methods. Finally, the dissertation develops recursive probabilistic approaches including particle filtering and agent-based modeling methods to predict travel times further into the future. Given the challenges in defining the particle filter time update process, the proposed particle filtering algorithm selects particles from a historical dataset and propagates particles using data trends of past experiences as opposed to using a state-transition model. A partial resampling strategy is then developed to address the degeneracy problem in the particle filtering process. INRIX probe data along I-64 and I-264 from Richmond to Virginia Beach are used to test the proposed algorithm. The results demonstrate that the particle filtering approach produces less than a 10 percent prediction error for trip departures up to one hour into the future for a two hour trip. Furthermore, the dissertation develops an agent-based modeling approach to predict travel times using real-time and historical spatiotemporal traffic data. At the microscopic level, each agent represents an expert in the decision making system, which predicts the travel time for each time interval according to past experiences from a historical dataset. A set of agent interactions are developed to preserve agents that correspond to traffic patterns similar to the real-time measurements and replace invalid agents or agents with negligible weights with new agents. Consequently, the aggregation of each agent's recommendation (predicted travel time with associated weight) provides a macroscopic level of output – predicted travel time distribution. The case study demonstrated that the agent-based model produces less than a 9 percent prediction error for prediction horizons up to one hour into the future. / Ph. D.

[pt] Com o aumento das questões energéticas e ambientais, os veículos elétricos (EVs) se tornarão um modo de transporte essencial na distribuição logística. Um cenário vital a ser considerado é a dependência do congestionamento do tráfego nos tempos de viagem dos veículos, como é comum nas áreas urbanas hoje. Esse recurso significa que a velocidade de um EV em cada rota pode ser distinta durante diferentes períodos. Como os EVs possuem autonomia limitada, vários trabalhos na literatura propuseram modelos de consumo de energia em função da velocidade e fatores aerodinâmicos. No entanto, sua aplicação permanece limitada e simplificada devido à sua dependência da velocidade e dos tempos de viagem. No caso da velocidade, os modelos da literatura trabalham sob uma velocidade média durante um determinado arco ou introduzem aproximações com métodos de linearização por partes. Em relação aos tempos de viagem, os atuais algoritmos de roteamento de veículos muitas vezes reformulam a rede viária em um gráfico completo onde cada arco representa o caminho mais rápido entre dois locais. Os resultados obtidos por esses métodos divergem da realidade, principalmente para problemas de roteamento de arco envolvendo serviços nos arcos de uma rede rodoviária. Por essas razões, definimos o Problema de Roteamento de Arco Capacitado Elétrico com tempos de viagem dependentes do tempo e taxa de consumo de energia dependente da velocidade. Ao longo de um horizonte de planejamento, cada arco está associado a uma função de velocidade passo a passo. O objetivo é atender um conjunto de arcos que demandam serviços por meio de uma frota de EVs com carga e capacidade de bateria limitadas, minimizando o tempo total de viagem. Além disso, a taxa de consumo de energia por unidade de tempo percorrido é considerada uma função não linear baseada na velocidade. Propomos um algoritmo de pré-processamento de consumo de energia de forma fechada sem aproximações. Nós o incorporamos em uma metaheurística Iterate Local Search e comparamos o impacto no projeto de rotas com os veículos convencionais. / [en] With energy and environmental issues rising, electric vehicles (EVs) will become an essential mode of transportation in logistics distribution. A vital scenario to consider is the dependence of traffic congestion on vehicle travel times, as it is common in urban areas today. This feature means that the speed of an EV on each route may be distinct during different periods. Because EVs have a limited driving range, various works in the literature have proposed energy consumption models as a function of speed and aerodynamic factors. However, their application remains limited and oversimplified due to their dependence on speed and travel times. In the case of speed, the models in the literature work under an average speed during a given arc or introduce approximations with piece-wise linearization methods. Regarding travel times, current vehicle routing algorithms often reformulate the road network into a complete graph where each arc represents the quickest path between two locations. The results obtained by these methods differ from reality, particularly for Arc Routing Problems involving services on the arcs of a road network. For these reasons, we define the Electric Capacitated Arc Routing Problem with Time-dependent Travel times, and Speed-dependent Energy Consumption Rate (E-TDCARP). Over a planning horizon, each arc is associated with a step-wise speed function. Based on this function, a vehicle s speed can change while traveling on a given arc. The objective is to serve a set of arcs that require services through a fleet of electric vehicles with limited load and battery capacity, minimizing the total travel time. Furthermore, the energy consumption rate per unit of time traveled (ECR) is considered a nonlinear function based on speed. We propose a closed-form energy consumption preprocessing algorithm without approximations. We embed it into an Iterate Local Search metaheuristic (ILS) for E-TDCARP and compare the impact on the design of routes between these alternative vehicles and conventional ones.

Les problèmes de localisation-routage à deux échelons (2E-LRP) sont devenus un domaine de recherche important dans le domaine de la logistique et de la gestion de la chaîne d'approvisionnement. Le 2E-LRP représente un problème d'optimisation dans les systèmes de distribution non dirigés, visant à organiser le transport de marchandises entre les plateformes et les clients par le biais d'installations intermédiaires appelées satellites. Ce problème implique de prendre des décisions simultanées concernant l'emplacement d'un ou deux niveaux d'installations (plateformes et/ou satellites) et de créer un ensemble limité d'itinéraires aux deux échelons afin de répondre efficacement à toutes les demandes des clients. Récemment, la communauté scientifique s'est intéressée de plus en plus à l'étude et à la résolution de problèmes plus réalistes. Cet intérêt provient de la reconnaissance du fait que les systèmes de distribution du monde réel sont caractérisés par une multitude de complexités et d'incertitudes qui ont un impact significatif sur l'efficacité opérationnelle, la rentabilité et la satisfaction des clients. Les chercheurs ont reconnu la nécessité d'aborder ces complexités et incertitudes pour développer des solutions pratiques et efficaces. Cette thèse comprend trois études différentes, chacune correspondant à un article de recherche autonome. Dans les trois articles, nous nous concentrons sur différents 2E-LRP riches qui comprennent plusieurs attributs en interaction. Ces variantes du problème sont appelées problèmes de localisation-routage à deux échelons et à attributs multiples (2E-MALRP). Pour analyser l'influence des incertitudes sur les solutions optimales et les processus de prise de décision, nous considérons à la fois les perspectives déterministes et stochastiques. Cette approche nous permet de mieux comprendre le comportement de ces problèmes complexes. Le premier document de recherche abordé dans cette thèse se concentre sur un problème de localisation-routage déterministe à deux échelons et à attributs multiples avec synchronisation de la flotte dans les installations intermédiaires (2E-MALRPS). Le cadre du problème comprend divers facteurs, notamment la demande de marchandises multiples dépendant du temps, les fenêtres temporelles, le manque de capacité de stockage dans les installations intermédiaires et la nécessité de synchroniser les flottes opérant à différents échelons. Dans le 2E-MALRPS, tous les paramètres, tels que les demandes des clients, les temps de trajet et les coûts, sont connus avec certitude. Dans cet article, nous introduisons le cadre du problème, présentons une formulation de programmation en nombres entiers mixtes et proposons un cadre de découverte de discrétisation dynamique comme méthode de résolution du problème. Le deuxième article de cette thèse traite du problème de localisation-routage à deux échelons en cas de demandes stochastiques et corrélées (2E-MLRPSCD). Contrairement au 2E-MALRPS, le 2E-MLRPSCD prend en compte les incertitudes liées aux demandes des clients, ainsi que la corrélation entre ces demandes. Nous formulons le problème sous la forme d'un modèle de programmation stochastique en deux étapes. Au cours de la première étape, des décisions sont prises concernant la conception des installations satellites, tandis qu'au cours de la deuxième étape, des décisions de recours déterminent la manière dont les demandes observées sont servies. Nous proposons une métaheuristique de couverture progressive comme méthode de résolution. Dans cette approche, nous incorporons deux structures de population dans le cadre de la couverture progressive. Ces structures renforcent la diversité des décisions de conception obtenues pour chaque sous-problème de scénario et fournissent des informations pertinentes pour améliorer la qualité de la solution. En outre, nous introduisons et comparons trois nouvelles stratégies différentes pour accélérer la recherche de l'espace de solution pour le problème stochastique. Finalement, le troisième article présenté dans cette thèse se concentre sur un problème de localisation-routage multi-attributs à deux échelons avec des temps de trajet stochastiques (2E-MALRPSTT). Le 2E-MALRPSTT combine un problème multi-attributs riche avec des éléments stochastiques, en particulier en considérant des temps de trajet stochastiques. Pour traiter le problème stochastique complet, un cadre de couverture progressive (PH) est proposé en s'appuyant sur les lignes directrices méthodologiques définies dans notre deuxième article pour le 2E-MLRPSCD. En outre, une heuristique basée sur la décomposition est introduite pour accélérer le cadre PH, et deux nouvelles stratégies d'agrégation sont présentées pour accélérer le processus de consensus concernant les décisions de la première étape. Les contributions présentées dans cette thèse couvrent divers aspects de la modélisation et des méthodologies de solution pour les 2E-MALRP riches, à la fois d'un point de vue déterministe et d'un point de vue stochastique. Les trois articles inclus dans cette thèse démontrent l'efficacité des approches proposées à travers des campagnes expérimentales étendues, mettant en évidence leur efficacité de calcul et la qualité des solutions, en particulier dans les cas difficiles. En abordant les aspects déterministes et stochastiques de ces 2E-MALRP, cette thèse vise à contribuer à l'ensemble des connaissances en optimisation de la logistique et de la chaîne d'approvisionnement, à répondre aux besoins importants de la littérature actuelle et à fournir des informations importantes pour les systèmes de distribution à deux échelons dans divers contextes. / The Two-Echelon Location-Routing Problems (2E-LRPs) have emerged as a prominent research area within the field of logistics and supply chain management. The 2E-LRP represents an optimization problem in undirected distribution systems, aiming to streamline freight transportation between platforms and customers through intermediate facilities known as satellites. This problem involves making simultaneous decisions concerning the location of one or two levels of facilities (platforms and/or satellites) and creating a limited set of routes at both echelons to effectively serve all customer demands. In recent years, there has been a growing interest among the scientific community in studying and solving more realistic problem settings. This interest arises from the recognition that real-world distribution systems are characterized by a multitude of complexities and uncertainties that significantly impact operational efficiency, cost-effectiveness, and customer satisfaction. Researchers have acknowledged the need to address these complexities and uncertainties to develop practical and effective solutions. This dissertation comprises three distinct studies, each serving as a self-contained research article. In all three articles, we focus on different rich 2E-LRPs that encompass multiple interacting attributes. These problem variants are referred to as two-echelon multi-attribute location-routing problems (2E-MALRPs). To analyze the influence of uncertainties on optimal solutions and decision-making processes, we consider both deterministic and stochastic perspectives. This approach allows us to gain insights into the behavior of these complex problem settings. The first research paper addressed in this thesis focuses on a deterministic two-echelon multi-attribute location-routing problem with fleet synchronization at intermediate facilities (2E-MALRPS). The problem setting encompasses various factors, including time-dependent multicommodity demand, time windows, lack of storage capacity at intermediate facilities, and the need for synchronization of fleets operating at different echelons. In the 2E-MALRPS, all parameters, such as customer demands, travel times, and costs, are known with certainty. In this paper, we introduce the problem setting, present a mixed-integer programming formulation, and propose a dynamic discretization discovery framework as the solution method to address the problem. The second paper in this thesis addresses the two-echelon multicommodity location-routing problem with stochastic and correlated demands (2E-MLRPSCD). In contrast to the 2E-MALRPS, the 2E-MLRPSCD takes into account uncertainties related to customer demands, as well as the correlation among these demands. We formulate the problem as a two-stage stochastic programming model. In the first stage, decisions are made regarding the design of satellite facilities, while in the second stage, recourse decisions determine how the observed demands are allocated and served. We propose a progressive hedging metaheuristic as the solution method. In this approach, we incorporate two population structures within the progressive hedging framework. These structures enhance the diversity of the design decisions obtained for each scenario subproblem and provide valuable insights for improving the solution quality. Additionally, We also introduce and compare three different novel strategies to accelerate the search for the solution space for the stochastic problem. Finally, the third paper presented in this thesis focuses on a multi-attribute two-echelon location-routing problem with stochastic travel times (2E-MALRPSTT). The 2E-MALRPSTT combines a rich multi-attribute problem setting with stochastic elements, specifically considering stochastic travel times. To address the complete stochastic problem, a progressive hedging metaheuristic is proposed building on the methodological guidelines defined in our second paper for the 2E-MLRPSCD. Furthermore, a decomposition-based heuristic is introduced to accelerate the PH framework, and two novel selection strategies are presented to expedite the consensus process regarding the first-stage decisions. The contributions presented in this thesis encompass various aspects of modeling and solution methodologies for rich 2E-MALRPs from both deterministic and stochastic perspectives. The three articles included in this thesis demonstrate the effectiveness of the proposed approaches through extensive experimental campaigns, highlighting their computational efficiency and solution quality, particularly in challenging instances. By addressing the deterministic and stochastic aspects of these 2E-MALRPs, this thesis aims to contribute to the broader body of knowledge in logistics and supply chain optimization, fill important gaps in the present literature and provide valuable insights for two-echelon distribution systems in diverse settings.

Cette thèse s’intéresse au problème classique de tournées de véhicules avec contraintes de capacité (CVRP pour Capacitated Vehicle Routing Problem) ainsi qu’une variante beaucoup plus complexe, soit le problème de tournées de véhicules dépendant du temps avec fenêtres de temps et points de transfert défini sur un réseau routier (TDVRPTWTP-RN pour Time-Dependent Vehicle Routing Problem with Time Windows and Transfer Points on a Road Network). Dans le premier article, le TDVRPTWTP-RN est résolu en adaptant une métaheuristique qui représente l’état de l’art pour le CVRP, appelé Slack Induction for String Removals (SISR). Cette métaheuristique fait appel au principe “détruire et reconstruire” en retirant des séquences de clients consécutifs dans les routes de la solution courante et en réinsérant ensuite ces clients de façon à créer une nouvelle solution. Le problème est défini sur un réseau routier où différents chemins alternatifs peuvent être utilisés pour se déplacer d’un client à l’autre. De plus, le temps de parcours sur chacun des arcs du réseau n’est pas fixe, mais dépend du moment où le véhicule quitte le sommet origine. S’inspirant de problèmes rencontrés en logistique urbaine, nous considérons également deux types de véhicules, de petite et grande capacité, où les grands véhicules sont interdits de passage au centre-ville. Ainsi, les clients du centre-ville ne peuvent être servis que suite au transfert de leur demande d’un grand à un petit véhicule à un point de transfert. Comme un point de transfert n’a pas de capacité, une problématique de synchronisation apparaît quand un grand véhicule doit y rencontrer un ou plusieurs petits véhicules pour leur transférer une partie de son contenu. Contrairement aux problèmes stricts de tournées de véhicules à deux échelons, les grands véhicules peuvent aussi servir des clients localisés à l’extérieur du centre-ville. Comme le problème abordé est beaucoup plus complexe que le CVRP, des modifications importantes ont dû être apportées à la métaheuristique SISR originale. Pour évaluer la performance de notre algorithme, un ensemble d’instances tests a été généré à partir d’instances existantes pour le TDVRPTW-RN. Les réseaux omt été divisés en trois régions : centre-ville, frontière et extérieur. Le centre-ville et l’extérieur sont respectivemnt les royaumes des petits et grands véhicules, tandis que la frontière (où l’on retrouve les points de transfert) peut être visité par les deux types de véhicules. Les résultats numériques montrent que la métaheuristique proposée exploite les opportunités d’optimiser une solution en déplaçant autant que possible les clients neutres, soit ceux qui peuvent être servis indifféremment par un petit ou un grand véhicule, des routes des petits véhicules vers les routes des grands véhicules, réduisant ainsi les coûteuses visites aux points de transfert. Les deuxième et troisième article s’intéressent à des concepts plus fondamentaux et font appel au problème plus simple du CVRP pour les évaluer. Dans le second article, un étude expérimentale est conçue afin d’examiner l’impact de données (distances) imprécises sur la performance de différents types d’heuristiques, ainsi qu’une méthode exacte, pour le CVRP. À cette fin, différents niveaux d’imprécision ont été introduits dans des instances tests classiques pour le CVRP avec 100 à 1 000 clients. Nous avons observé que les meilleures métaheuristiques demeurent les meilleures, même en présence de hauts niveaux d’imprécision, et qu’elles ne sont pas affectées autant par les imprécisions qu’une heuristique simple. Des expériences avec des instances réelles ont mené aux mêmes conclusions. Le troisième article s’intéresse à l’intégration de l’apprentissage automatique dans la métaheuristique SISR qui représente l’état de l’art pour le CVRP. Dans ce travail, le principe “détruire et reconstruire” au coeur de SISR est hybridé avec une méthode d’apprentissage par renforcement qui s’inspire des systèmes de colonies de fourmis. L’ap- prentissage automatique a pour but d’identifier les arêtes les plus intéressantes, soit celles qui se retrouvent le plus fréquemment dans les solutions de grande qualité précédemment rencontrées au cours de la recherche. L’inclusion de telles arêtes est alors favorisé lors de la réinsertion des clients ayant été retirés de la solution par le mécanisme de destruction. Les instances utilisées pour tester notre approche hybride sont les mêmes que celles du second article. Nous avons observé que notre algorithme ne peut produire que des solutions lé- gèrement meilleures que la métaheuristique SISR originale, celle-ci étant déjà quasi-optimale. / This thesis is concerned both with the classical Capacitated Vehicle Routing Problem (CVRP) and a much more complex variant called the Time-Dependent Vehicle Routing Problem with Time Windows and Transfer Points on a Road Network (TDVRPTWTP-RN ). In the first paper, the TDVRPTWTP RN is solved by adapting a state-of-the-art metaheuris- tic for the CVRP, called Slack Induction for String Removals (SISR). This metaheuristic is based on the ruin and recreate principle and removes strings of consecutive customers in the routes of the current solution and then reinserts the removed customers to create a new solution. The problem is formulated in a full road network where different alternative paths can be used to go from one customer to the next. Also, the travel time on each arc of the road network is not fixed, but depends on the departure time from the origin node. Motivated from city logistics applications, we also consider two types of vehicles, large and small, with large vehicles being forbidden from the downtown area. Thus, downtown customers can only be served through a transfer of their goods from large to small vehicles at designated transfer points. Since transfer points have no capacity, synchronization issues arise when a large vehicle must meet one or more small vehicles to transfer goods. As opposed to strict two-echelon VRPs, large vehicles can also directly serve customers that are outside of the downtown area. Given that the TDVRPTWTP-RN is much more complex than the CVRP, important modifications to the original SISR metaheuristic were required. To evaluate the performance of our algorithm, we generated a set of test instances by extending existing instances of the TDVRPTW-RN . The road networks are divided into three regions: downtown, boundary and outside. The downtown and outside areas are the realm of small and large vehicles, respectively, while the boundary area that contains the transfer points can be visited by both small and large vehicles. The results show that the proposed metaheuristic exploits optimization opportunities by moving as much as possible neutral customers (which can be served by either small or large vehicles) from the routes of small vehicles to those of large vehicles, thus avoiding costly visits to transfer points. The second and third papers examine more fundamental issues, using the classical CVRP as a testbed. In the second paper, an experimental study is designed to examine the impact of inaccurate data (distances) on the performance of different types of heuristics, as well as one exact method, for the CVRP. For this purpose, different levels of distance inaccuracies were introduced into well-known benchmark instances for the CVRP with 100 to 1,000 customers. We observed that the best state-of-the-art metaheuristics remain the best, even in the presence of high inaccuracy levels, and that they are not as much affected by inaccuracies when compared to a simple heuristic. Some experiments performed on real-world instances led to the same conclusions. The third paper focuses on the integration of learning into the state-of-the-art SISR for the CVRP. In this work, the ruin and recreate mechanism at the core of SISR is enhanced by a reinforcement learning technique inspired from ant colony systems. The learning component is aimed at identifying promising edges, namely those that are often found in previously encountered high-quality solutions. The inclusion of these promising edges is then favored during the reinsertion of removed customers. The benchmark instances of the second paper were also used here to test the new hybrid algorithm. We observed that the latter can produce only slightly better solutions than the original SISR, due to the quasi-optimality of the original solutions.