A less-than-truckload (LTL) carrier typically delivers shipments less than 10,000 pounds (classified as LTL shipment). The size of the shipment in LTL networks provides ample opportunities for consolidation. LTL carriers have focused on hub-and-spoke based consolidation to realize economies of scale. Generally, hub-and-spoke systems work as follows: the shipment is picked up from the shipper and brought to an origin terminal, which is the entry point into the hub-and-spoke system. From the terminal, the freight is sent to the first hub, where it is sorted and consolidated with other shipments, and then sent on to a second hub. It is finally sent from the second hub to the destination terminal, which is the exit point of the hub-and-spoke system. However, the flow of shipments is often more complicated in practice. In an attempt to reduce sorting costs, load planners sometimes take this hub-and-spoke infrastructure and modify it considerably to maximize their truck utilization while satisfying service constraints. Decisions made by a load planner may have a cascading effect on load building throughout the network. As a result, decentralized load planning may result in expensive global solutions. Academic as well as industrial researchers have adapted a hierarchical approach to design the hub-and-spoke networks: generate the hub-and-spoke network, route shipments within this hub-and-spoke network (generate a load plan) and finally, balance the empty trailers. We present mathematical models and heuristics for each of the steps involved in the design of the hub-and-spoke network. The heuristics are implemented in a user-friendly graphical tool that can help understand patterns of freight flow and provide insights into the design of the hub-and-spoke network. We also solved the load planning sub-problem in a parallel computation environment to achieve significant speed-ups. Because of the quick solution times, the tool lays the foundation to address pressing further research questions such as deciding location and number of hubs. We have used data provided by Roadway Parcel Services, Inc. (RPS), now FedEx Ground, as a case-study for the heuristics. Our solutions rival the existing industry solutions which have been a product of expensive commercial software and knowledge acquired by the network designers in the industry.

Rail-road terminals are the chain links in intermodal rail-road transportation where standardized load units (containers, swap bodies and trailers) are transfered from trucks to trains and vice versa. We consider three subproblems of the operational planning process at rail-road terminals that terminal operators are facing in their daily operations. These are the optimization problems storage planning, load planning and crane planning. The aim of storage planning is to determine load unit storage positions for a set of load units in a partially filled storage area. Here, different restrictions like non-overlapping of stored load units have to be respected. The objective of storage planning is to minimize the total transportation costs and the number of load units that are not stored at the ground level. For the load planning we assume a scenario of overbooked trains. So, the aim of load planning is to assign a subset of the load units that are booked on a train to feasible positions on the wagons such that the utilization of the train is maximized and the costs for the handling in the terminal are minimized. For the feasible positioning of load units length and weight restrictions for the wagons and the train have to be respected. For the load planning of trains we consider a deterministic version and a robust approach motivated by uncertainty in the input data. The last considered optimization problem is the crane planning. The crane planning determines the transfer of the load units by crane between the different transportation modes. For each crane a working plan is computed which contains a subset of the load units that have to be handled together with individual start times for the transfer operations. For the load units which have to be transfered in the terminal, storage and load planning compute destination positions (inside the terminal). These destination positions are part of the input for the crane planning. The main objective of crane planning is to minimize the total length of the empty crane moves that have to be performed between successive transports of load units by the cranes. We provide MIP-models for all three subproblems of the operational planning process at rail-road terminals. For the storage and crane planning we also propose fast heuristics. Furthermore, we present and compare computational results based on real world data for all subproblems. The main contributions of this thesis concern load and storage planning. For the deterministic load planning we provide the first model that represents all practical constraints including physical weight restrictions. For the load planning we furthermore present robustness approaches for different practical uncertainties. For the storage planning we provide complexity results for different variants. For the practical setting we developed a heuristic which is able to compute solutions of high quality in a small amount of runtime.

No description available.

This thesis focuses on decision support for regional LTL carriers. The basic operating characteristics of regional LTL carriers are similar to those of national LTL carriers, i.e., they operate linehaul networks with satellites, breakbulks, and relays to consolidate freight so as to be able to cost-effectively serve their customers. However, there are also key differences. Most importantly, because the area covered by a regional carrier is smaller, a regional carrier handles less freight (sometimes significantly less) and therefore typically has fewer consolidation opportunities, which results in higher handling and transportation costs per unit of freight. Consequently, competing with national carriers on price is difficult. Therefore, to gain or maintain market share, regional carriers have to provide better service. To be able to provide better service, regional carriers have to be more dynamic, e.g., they have to be able to deviate from their load plan when appropriate, which creates challenges for decision makers. Regional carriers deliver about 60% of their shipments within a day and almost all of their shipments within two days. Furthermore, most drivers get back to their domicile at the end of each day. Therefore, the focus of the thesis is the development of effective and efficient decision models supporting daily operations of regional LTL carriers which provide excellent service at low cost. This thesis presents an effective solution approach based on two optimization models: a dynamic load planning model and a driver assignment model. The dynamic load planning model consists of two parts: an integer program to generate the best paths for daily origin-destination freight volumes and an integer program to pack freight into trailers and trailers into loads, and to determine dispatch times for these loads. Techniques to efficiently solve these integer program solution are discussed in detail. The driver assignment model is solved in multiple stages, each stage requiring the solution of a set packing models in which columns represent driver duties. Each stages determines admissible driver duties. The quality and efficiency of the solution approach are demonstrated through a computational study with real-life data from one of the largest regional LTL carriers in the country. An important "technique" for reducing driver requirements is the use of meet-and-turn operations. A basic meet-and-turn operation involves two drivers meeting at a location in between terminals and exchange trucks. A parking lot or a rest area suffices as a meet-and-turn location. This ensures that drivers return to the terminal where they started. More sophisticated meet-and-turn operations also exist, often called drop and hook operations. In this case, drivers do not exchange trucks, but one of their trailers. The motivation in this case is not to get drivers back to their domicile, but to reduce load- miles. The thesis presents analytical results quantifying the maximum benefits of using meet and turn operations and optimization techniques for identifying profitable meet-and-turn opportunities.

Les trains qui transportent des conteneurs empilés (en deux niveaux) sont un élément important du reseau de transport nord-americain. Le probleme de chargement des wagons correspond un probleme operationnel d'utilisation rencontre dans les terminaux ferroviaires. Elle consiste optimiser l’affectation des conteneurs des emplacements spécifiques sur les wagons. Ce mémoire est centré sur un article scientifique traitant le chargement optimal publié dans le Journal Européen de Recherche Opérationnelle (Volume 267, Numéro 1, Pages 107-119, 2018). Nous avons formule un modele lineaire en nombres entiers (ILP) et apporte un certain nombre de contributions. Premierement, nous avons proposé une méthodologie générale qui peut traiter des wagons double ou simple empilement avec des «patrons» de chargement arbitraires. Les les patrons tiennent un compte des dépendances de chargement entre les plateformes sur un wagon donne. Deuxiemement, nous avons modéliser les restrictions du centre de gravité (COG), les regles d’empilement et un nombre de restrictions techniques de chargement associees certains types de conteneurs et / ou de marchandises. Les resultats montrent que nous pouvons resoudre des instances de taille realiste dans un d´elai raisonnable en utilisant un solveur ILP commercial et nous illustrons que le fait de ne pas tenir compte de la correspondance conteneurs-wagons ainsi que des restrictions COG peut conduire une surestimation de la capacité disponible. / Double-stack trains are an important component of the railroad transport network for containerized cargo in specific markets such as North America. The load planning problem embodies an operational problem commonly faced in rail terminals by operators. It consists in optimizing the assignment of containers to specific locations on the train. The work in this thesis is centered around a scientific paper on the optimization on load planning problem for double stack-trains, published in the European Journal of Operation Research (Volume 267, Issue 1, Pages 1-398) on 16 May 2018. In the paper, we formulated an ILP model and made a number of contributions. First, we proposed a general methodology that can deal with double- or single-stack railcars with arbitrary loading patterns. The patterns account for loading dependencies between the platforms on a given railcar. Second, we modeled Center of gravity (COG) restrictions, stacking rules and a number of technical loading restrictions associated with certain types of containers and/or goods. Results show that we can solve realistic size instances in reasonable time using a commercial ILP solver and we illustrate that failing to account for containers-to-cars matching as well as COG restrictions may lead to an overestimation of the available train capacity.

The work in this thesis is separated into two parts. The first part deals with the load planning and sequencing problem for double-stack intermodal railcars, an operational problem found at many rail container terminals. In this problem, containers must be assigned to a platform on which the container will be loaded, and the loading order must be determined. These decisions are made with the objective of minimizing the costs associated with handling the containers, as well as minimizing the cost of containers left behind. The deterministic version of the problem can be cast as a shortest path problem on an ordered graph. This problem is challenging to solve because of the large size of the graph. We propose a two-stage heuristic based on the Iterative Deepening A* algorithm to compute solutions to the load planning and sequencing problem within a five-minute time budget. Next, we also illustrate how a Deep Q-learning algorithm can be used to heuristically solve the same problem.The second part of this thesis considers sequential models in deep learning. A recent strategy to circumvent the exploding and vanishing gradient problem in recurrent neural networks (RNNs) is to enforce recurrent weight matrices to be orthogonal or unitary. While this ensures stable dynamics during training, it comes at the cost of reduced expressivity due to the limited variety of orthogonal transformations. We propose a parameterization of RNNs, based on the Schur decomposition, that mitigates the exploding and vanishing gradient problem, while allowing for non-orthogonal recurrent weight matrices in the model. / Le travail de cette thèse est divisé en deux parties. La première partie traite du problème de planification et de séquencement des chargements de conteneurs sur des wagons, un problème opérationnel rencontré dans de nombreux terminaux ferroviaires intermodaux. Dans ce problème, les conteneurs doivent être affectés à une plate-forme sur laquelle un ou deux conteneurs seront chargés et l'ordre de chargement doit être déterminé. Ces décisions sont prises dans le but de minimiser les coûts associés à la manutention des conteneurs, ainsi que de minimiser le coût des conteneurs non chargés. La version déterministe du problème peut être formulé comme un problème de plus court chemin sur un graphe ordonné. Ce problème est difficile à résoudre en raison de la grande taille du graphe. Nous proposons une heuristique en deux étapes basée sur l'algorithme Iterative Deepening A* pour calculer des solutions au problème de planification et de séquencement de la charge dans un budget de cinq minutes. Ensuite, nous illustrons également comment un algorithme d'apprentissage Deep Q peut être utilisé pour résoudre heuristiquement le même problème. La deuxième partie de cette thèse examine les modèles séquentiels en apprentissage profond. Une stratégie récente pour contourner le problème de gradient qui explose et disparaît dans les réseaux de neurones récurrents (RNN) consiste à imposer des matrices de poids récurrentes orthogonales ou unitaires. Bien que cela assure une dynamique stable pendant l'entraînement, cela se fait au prix d'une expressivité réduite en raison de la variété limitée des transformations orthogonales. Nous proposons une paramétrisation des RNN, basée sur la décomposition de Schur, qui atténue les problèmes de gradient, tout en permettant des matrices de poids récurrentes non orthogonales dans le modèle.