Large Language Models (LLMs) offer significant potential by serving as human proxies to advance travel demand modeling, but their behavioral misalignment with human travelers remains a critical obstacle. Furthermore, existing alignment methods are often impractical or inefficient when applied to the sparse data sets typically available for travel choices, limiting the adoption of these powerful new tools. We introduce a novel framework to align LLMs with travel choice behavior. Our method first infers a set of traveler personas from empirical data and then estimates a persona loading function that uses learned embeddings to select the appropriate persona for an individual based on their sociodemographics. Validated on the Swissmetro mode choice data set, our approach significantly outperforms established benchmarks in predicting both aggregate and individual choice outcomes. Our research offers a more adaptable, interpretable, and resource-efficient pathway to robust LLM-based travel behavior simulation, paving the way to integrate LLMs into transportation modeling practice in the future.
We introduce a new variant of the Undirected Team Orienteering Arc Routing Problem (UTOARP) that incorporates three key features: required edges, capacitated vehicles, and multiple services. These features have been investigated individually in the literature but have not been considered simultaneously. In this problem, demand is placed at some edges of a given undirected graph, and served demand edges produce a profit. Feasible routes must start and end at a given depot, and there is a time limit on the maximum duration of each route and a capacity limit on the demand served by each vehicle. The problem asks for a set of up to |K|maximum profit routes while ensuring all required edges are served. We exploit optimality conditions for this problem and propose a new unified, undirected formulation with binary variables. We also introduce a logic-based Benders decomposition derived from this formulation, resulting in a new problem reformulation, and show how to strengthen the logic-based Benders cuts. Crucially, the structure of the Benders subproblems remains unchanged regardless of which of the above features are enabled, highlighting the modularity and flexibility of the approach. Furthermore, we design several new families of valid inequalities, where some of them are derived from conflict graphs. Extensive computational tests are conducted to examine the performance of the proposed formulations and valid inequalities under various settings. We further analyze the solution structure of a real-world instance to illustrate the practical impact of the different features. Our findings highlight the pivotal role of logic-based Benders decomposition and conflict graphs in solving the UTOARP, marking their first application in the context of arc routing problems to the best of our knowledge. Moreover, these techniques hold promise for advancing solution approaches in broader arc routing contexts.
This article may be used only for the purposes of research, teaching, and/or private study. Commercial use or systematic downloading (by robots or other automatic processes) is prohibited without explicit Publisher approval, unless otherwise noted. For more information, contact permissions@informs.org. The Publisher does not warrant or guarantee the article's accuracy, completeness, merchantability, fitness inclusion of an advertisement in this article, neither constitutes nor implies a guarantee, endorsement, or support
We consider emergency preparedness for foreseen disasters such as hurricanes and flooding. Proper preparation and planning for timely relief supply in a cost-effective manner can be crucial determinants of how quickly recoveries occur and with the least suffering of the affected populations. Assuming a host of shelter locations aggregated locally, we are interested in determining relief supply locations (distribution centers [DCs]) and routing of supply from the capacitated DCs to the shelters on an underlying time-phased network for cost-effective timely delivery. We present a mixed integer optimization model to address this problem under the assumption of covering the worst case demand at the shelters. To solve our model, we develop an efficient Benders decomposition-based algorithm that handles the challenges of obtaining optimality cuts via master problem solution modifications and various surrogate constraints among other enhancement techniques. We test the performance of the enhancement techniques on an extensive randomly generated test data set to identify the most effective approach. We finally use our model and the solution algorithm on an actual case study in southern Texas data incorporated and managed by a geographical information system to examine the impact of various input parameters on the design and tactical operation of the relief networks as well as for further model verification and validation.
Railway passenger demand fluctuates because of events, holidays, and individual traveler activities, making it difficult to predict several months in advance. Market-oriented railway plans feature strategies for canceling or adding train trips according to short-term forecasted passenger demand. This results in dynamic adjustments of the rolling stock schedule (RSS) and the train platforming plan (TPP) on a daily basis. The RSS defines the schedule of the train units (TUs), that is, the sequence of trips that each TU will execute and the maintenance appointments it has to undergo. The TPP defines the assignment of platforms to TUs: when two trips are executed in sequence by the same TU, a platform has to be assigned to the TU for the trip connection. Because of demand fluctuation, the RSS has to be adjusted to perform a set of trips different from those in the original plan, requiring new trip sequences and rescheduled maintenance appointments for the TUs, whereas the TPP has to be modified to guarantee the feasible assignment of the platforms to the TUs performing the newly defined trip sequences. To assist railway departments in making optimized decisions, this work studies the integrated adjustment of the RSS and TPP by proposing an integer linear programming model and an exact decomposition algorithm. The integrated model consists of two parts: (i) rolling stock scheduling with maintenance constraints, and (ii) platform assignment. The goal is to minimize the operating costs and the deviations from the original plan. The decomposition algorithm consists of a branch-and-check (B&C) method in which maintenance constraints and platform assignment are handled by dynamic cut generation, and problem-specific acceleration techniques are incorporated to reduce the search space and the computation times. The proposed B&C method is tested on real-world instances from the Chinese high-speed railway system, involving up to 981 trips, showing that optimal solutions are obtained within one hour of computation time. It is shown that the B&C method clearly outperforms solving the integrated model by a general-purpose solver even on medium-size instances. Furthermore, comparisons with different sequential methods, which firstly adjust the RSS and then compute a new TPP, highlight the benefits of the integrated framework.
We focus on a logistics service provider (LSP) that organizes the inbound logistics for an original equipment manufacturer (OEM) using just-in-time production. The LSP conducts periodic milk runs, that is, tours visiting a subset of suppliers on a fixed route to collect shipments bound for the OEM. Although the milk-run routes and rough schedules are fixed well in advance, the exact shipment due dates at the OEM become known only once the production sequence has been finalized. During daily operation, the given milk runs are executed such that shipments are picked up from the suppliers and delivered to the OEM ideally exactly just in time. However, the production sequence on the assembly lines at the OEM often changes on the day of execution, which alters shipment due dates at the OEM and makes it necessary to anticipate potential disturbances when assigning shipments to milk runs. We formulate a two-stage stochastic program, where, in the first stage, shipments are assigned to specific milk runs and, in the second stage, when alterations to the production sequence are revealed, arrival times are adjusted and shipments are express delivered if necessary. The assignment induces earliness-tardiness costs depending on the uncertain due date of each shipment at the OEM, and express deliveries induce fixed costs. Besides an extensive-form mixed-integer linear program, we develop a solution procedure based on the integer L-shaped method, enriched with valid inequalities and cuts derived from partial Benders decomposition and a subproblem relaxation. In a computational study on realistic test data, we show that the proposed L-shaped method solves 62 out of 70 instances to optimality. From a managerial perspective, we show that express deliveries are an effective way of mitigating uncertainty in the production sequence, as are overlapping milk-run routes, which are not yet widely used in practice but are relatively easy to implement.
Crowdshipping has emerged as a novel service paradigm that leverages excess capacity in the transportation system by enlisting the traveling public, namely the "crowd", to deliver parcels during their daily trips. To incentivize heterogeneous travelers as crowd carriers, this study proposes effective auction mechanisms for a crowdshipping system that integrates both crowdshipping and dedicated delivery services. An intermediary platform charges a service fare from customers and solicits bids from potential crowd carriers. Based on the auction outcome, the platform either allocates orders and makes payments to the winning bidders or outsources the orders to dedicated delivery services. We develop a single-sided sealed-bid combinatorial auction to allocate orders and determine compensation for crowd carriers. This auction procedure allows carriers to submit mutually exclusive bids for order bundles that align with their original travel routes. We apply the Vickrey-Clarke-Groves mechanism to achieve allocative efficiency and strategyproofness. To enhance scalability, we also devise an approximation mechanism that combines greedy allocation with a second-best pricing policy. The greedy mechanism provides an upper bound on total system cost and satisfies approximate strategy-proofness with bounded ex post regret. A positive ex post regret can only arise when a crowd carrier overbids on certain bundles, and its magnitude is bounded by the true cost savings of the assigned bundle under misreporting. Under the same mild conditions, both mechanisms yield nonnegative platform profit. Our analysis shows that the platform can effectively tune the performance of both auction mechanisms by regulating the maximum number of bids per crowd carrier and the maximum number of orders per bundle. To achieve more effective cost reduction, the platform is advised to conduct auctions when the number of orders substantially exceeds the number of crowd carriers. The two auction mechanisms developed in this paper can be applied to a more general setting where suppliers submit bids on bundles of items and there exists a fixed-price backup option for each item.
The increasing demand for expedited e-commerce deliveries, with delivery times of one to three days, highlights the importance of optimizing the middle-mile network. Most retailers store a considerable portion of their inventory at the regional distribution centers (RDCs) outside urban areas, from where it is moved to the customer zones equipped with last-mile distribution facilities as required. Thus, RDC locations become critical in middle-mile operations, directly impacting the transit times to customer zones and, ultimately, the delivery times in the last mile. This paper presents a middle-mile network design problem arising in the context of e-commerce companies in the presence of customers with different delivery time preferences. Specifically, it allows RDCs to satisfy demands from customer zones using delivery times longer than requested, albeit with penalties, if that helps reduce cost without violating the service level requirements of fulfilling at least a given threshold of the demands within the requested delivery times. The problem is formulated as a mixed-integer linear program, for which an exact Lagrangian relaxationbased branch-and-bound algorithm is proposed. Several enhancements to the algorithm are provided, including an efficient Lagrangian heuristic for the primal-bound, a Benders decomposition framework to solve one of the Lagrangian subproblems efficiently, an analytical approach for obtaining Benders optimality cuts, and a partial analytical characterization of Pareto-optimal Benders cuts. With these enhancements, our final algorithm substantially outperforms the state-of-the-art commercial solver, as highlighted by our computational experiments on an extensive set of 220 instances with up to 80 potential RDC locations and 1,000 customer zones. Our best algorithm solves 204 of the 220 instances to 0.50% duality gap compared with only 108 that CPLEX could solve to the same gap within an allowed 10-hour CPU time limit. Furthermore, it achieves an average time savings of 63.24% compared with CPLEX across all the instances.
Maritime shipping faces stringent exhaust emission requirements because of sulfur emission regulations and the European Union Emissions Trading System (EU ETS), driving shipping companies to adopt a range of emission reduction technologies, such as scrubbers, liquefied natural gas (LNG) propulsion systems, and methanol propulsion systems. Given that many shipping companies operate fleets equipped with multiple emission reduction technologies, this study investigates an integrated fleet deployment and speed optimization problem for a shipping company operating three or more types of ships (traditional ships, scrubber-equipped ships, and LNG-or methanol-powered ships) under sulfur emission regulations and the EU ETS carbon emission regulation. A mixed-integer nonlinear programming (MINLP) model is proposed to address this optimization problem. Because of their differing regulatory mechanisms, sulfur and carbon emission regulations affect fleet deployment (i.e., the types and number of ships deployed across all routes) and speed optimization in distinct ways. As the number of ship types increases, the number of feasible fleet deployment plans grows sharply, whereas the inclusion of different ship types further complicates speed optimization, increasing the overall problem complexity. To tackle this challenge, the study performs mathematical derivations and analyses to reveal model properties and construct valid inequalities, significantly narrowing the feasible solution space. The MINLP model is first linearized according to its characteristics. Leveraging the model properties, a Benders decomposition algorithm with a tailored cut pool is developed to solve the linearized model, which serves as the foundation for a highly efficient exact algorithm for the original MINLP model. Numerical experiments show that the proposed exact algorithm achieves a nearly 90-fold reduction in computation time compared with the CPLEX-based algorithm.
The rapid growth of online meal delivery has introduced complex logistical challenges, where platforms must dynamically assign orders to couriers while accounting for demand uncertainty, courier autonomy, and service efficiency. Traditional dispatching methods, often focused on short-term cost minimization, fail to capture the long-term implications of assignment decisions on system-wide performance. This paper presents a novel hybrid framework that integrates reinforcement learning with hyper-heuristic optimization to improve sequential order assignment and routing decisions in meal delivery operations. Our approach combines n-step state-action-reward-state-action with value function approximation and a multiarmed bandit-based hyper-heuristic incorporating seven specialized low-level heuristics. Our approach explicitly models the evolving system state, enabling dispatching policies that balance immediate efficiency with future operational performance. By employing scalable linear value function approximation, we enhance policy learning in high-dimensional environments while maintaining generalization across states and actions. Using real operational data from the food delivery platform Meituan, we develop a comprehensive simulation environment that captures order dynamics, courier behavior, and service times. Through extensive computational experiments, we demonstrate that our framework significantly outperforms traditional benchmark policies, achieving 12% cost reduction through strategic order postponement. Our results reveal that the largest improvements occur during high-demand periods with courier shortages and that a 10% increase in courier availability yields greater benefits than algorithmic improvements alone. The proposed methodology effectively balances immediate operational efficiency with long-term performance while providing valuable insights for meal delivery platforms regarding courier fleet management and order assignment strategies.
Maintaining the quality of temperature-controlled perishable products during distribution is essential. Several factors affect product quality and the energy cost of refrigeration during distribution, including the temperature inside delivery trucks, route duration, number of stops, and vehicle load. The existing literature on routing for perishable products considers only a limited set of these factors, and most solution methods rely on heuristics or commercial solvers. This research formulates a vehicle routing problem with time windows and quality considerations. The objective is to determine a set of vehicle routes that minimize both travel costs and refrigeration energy costs while satisfying customer quality criteria and time windows. To capture product quality decay during transportation, we propose a function that accounts for the effect of temperature, route duration, number of stops, and vehicle load, thereby extending existing quality decay models in the literature. To solve the problem, we developed a tailored exact framework with several specialized features to efficiently manage quality decay and energy cost functions. Computational experiments demonstrate that the proposed algorithm is capable of solving instances with up to 100 customers. In addition, we revisit the trade-offs in temperature-controlled routing and provide useful insights into the impact of quality decay and temperature-related coefficients on routing decisions and costs.
One of the more novel recent innovations in the logistics world, both in theory and in practice, is the use of small autonomous vehicles to facilitate last-mile delivery. One particular scheme that has received considerable recent attention is the "sidekick" scheme, in which a large cargo truck acts as a mobile "host" that deploys smaller vehicles such as aerial drones or unmanned ground vehicles (UGVs). In this paper, we develop a continuous approximation model that estimates the improvements to total completion time that such a system provides in the asymptotic limit as many demand points are drawn from a continuous probability distribution in the plane. Our key finding is that sidekick systems can be beneficial even when the sidekicks are slower than the host, provided there are sufficiently many of them.
Disruptions to high-speed train operations result in deviations from the official timetable and train delays, leading to scheduling conflicts. A conflict-free rescheduled timetable and corresponding executable speed profiles are essential to restore disrupted train operations. In this paper, we introduce an integrated framework for the train rescheduling and speed management problem (TRSMP) during disruptions caused by partial segment blockages. Under a flexible rescheduling strategy, affected trains can switch to the opposite track to bypass the blockage. A mixed-integer linear programming model is formulated to simultaneously optimize train times, orders, routes, and speed profiles, aiming to minimize the total deviation time and traction energy consumption. To reduce the complexity of the integrated model, we decompose the model into a reformulated integer programming model and a series of train operation plans. This decomposition approach transforms the TRSMP into a train plan selection problem, effectively decoupling both the interactions among trains and interdependencies between rescheduling and speed management. Moreover, we develop an adaptive algorithm with acceleration techniques to speed up the solving of the reformulated model. Case studies based on the Chinese highspeed railway show that our decomposition approach provides near-optimal solutions within acceptable computational time in large-scale instances. Our integrated method outperforms the sequential optimization method in both solution time and feasibility guarantees of speed profiles. The proposed rescheduling strategy exhibits enhanced effectiveness in mitigating delays while preserving line capacity during partial blockages.
Capturing latent demand has a pivotal role in designing transit services, as omitting these riders can lead to poor quality of service and/or additional costs. This paper explores this topic in the design of transit networks by considering the perspectives of both the transit agencies and riders. The paper presents a generic bilevel optimization model- namely, the Transit Networks Design with Adoptions (TN-DA)-that considers the network design decisions in the leader problem and routing of the riders in the follower problem under the given network design, while allowing a black-box choice function for representing the adoption behavior of latent demand. The paper then identifies structural properties of the optimal solution of the TN-DA problem, which are desirable for transit agencies for capturing adoption behavior of the riders. The paper further provides guideline metrics for the transit agencies based on these desired adoption properties. Because of the computational complexity of this bilevel problem, the paper proposes five efficient heuristic algorithms to solve large-scale instances, which leverage an iterative procedure by solving a simpler version of the TN-DA problem and integrating the evaluation of rider choices. These algorithms either satisfy the desired properties of the optimal solution or provide fast approximations. The paper presents extensive large-scale case studies on two different transit systems by utilizing real data sets: (i) On-demand Multimodal Transit Systems (ODMTS) and (ii) Scooters-Connected Transit Systems (SCTS). The results demonstrate that under time limits, the heuristic algorithms can find high-quality solutions satisfying key adoption properties of the optimal solutions much faster than the exact approaches over various large-scale instances of ODMTS and SCTS.
With the normalization of telecommuting (TLC) and the rapid advancement of autonomous vehicle (AV) technology, commuting behavior has become increasingly flexible across both temporal and spatial dimensions, exhibiting greater heterogeneity than in the past. Consequently, conventional models that focus solely on the morning peak fail to capture commuters' full-day decision-making processes and the systematic evolution of congestion. To address this gap, this study develops a full-day commuting equilibrium model that jointly incorporates AV technology and the TLC option. The model explicitly characterizes commuters' choices of work mode, departure timing, and parking location within a classical bottleneck framework, integrating AV parking capacity and the demand adjustments induced by TLC. Multiple equilibrium traffic patterns are derived along with both the endogenous TLC proportion and the socially optimal exogenous TLC proportion under each pattern. The model further investigates how key parameters-such as the cost of early arrival schedule delay and parking density-affect the equilibrium TLC proportion and congestion dynamics. Analytical derivations and numerical simulations demonstrate that neglecting evening commuting leads to systematic biases and highlight the critical role of a full-day framework in identifying intertemporal decision making and behavioral heterogeneity. The findings provide theoretical insights and quantitative tools for commuting modeling and policy design in the era of autonomous driving and widespread telecommuting.
We consider a planning problem for freight transportation carriers that seek to profitably match supply with demand while recognizing uncertainty in shipment volumes. On the supply side, the problem determines transportation network design decisions regarding hub locations and the number of vehicles to be dispatched within the network in each period of the planning horizon. On the demand side, the problem incorporates the carrier's ability to expand its service coverage by selectively accepting additional customer demands beyond its existing contractual base. Furthermore, although some of these additional customers seek a long-term commitment from the carrier, others are transactional and only require the transportation of a single set of shipments. We refer to this problem as the demand-driven hub network design under uncertainty problem and formulate it as a twostage stochastic program. Further, we develop an enhanced Benders decomposition-based solution method for solving instances of this model. The solution methodology is inspired by partial Benders decomposition, leveraging a problem reformulation that embeds subsets of subproblem variables and constraints into the master problem while also incorporating valid inequalities to strengthen the formulation. We illustrate with an extensive computational study that the proposed method outperforms adaptations of benchmarks proposed for similar problems. We validate the benefits of solving the proposed model, which integrates decisions that have not yet been jointly modeled, with an analysis based on sample average approximation.
Traditional single-vehicle systems often fail to provide efficient solutions in applications, such as rescue operations, postdisaster relief, environmental mapping, and last-mile delivery. Coordinated systems composed of two echelons and heterogeneous vehicles with complementary capabilities offer a promising alternative. This paper focuses on the carriervehicle traveling salesman problem, which involves a slow carrier vehicle (mothership) transporting a fast vehicle (e.g., a drone) that must take off from and return to the mothership to serve a set of customers. We propose a mixed-integer nonlinear programming formulation and an approximation that can be modeled as a mixed-integer linear programming model. An exact algorithm is proposed that corrects the approximation error of the nonlinear term. Combined with a set of valid inequalities, our branch-and-cut algorithm guarantees optimal solutions for the original nonlinear problem. We evaluate our algorithm on a wide range of benchmark instances from the literature, ranging from 10 to 200 customers. Our method outperforms all previous exact and heuristic approaches in the literature in terms of objective value or computational time. Moreover, we are the first to solve all benchmark instances with up to 25 customers optimally, and notably, we find a proven optimal solution for an instance with 35 customers, the largest one solved to optimality in the literature.
This study investigates several classes of integrated supply and inventory (SI) planning problems, addressing supplier selection with limited or unlimited capacities, distribution, and centralized or decentralized inventory planning for various products with uncertain demand. Our focus is on retailers specializing in fashion products characterized by seasonal demand, extended lead times, and the inability to restock during midselling seasons, requiring salvage of unsold items at diminished returns. We formulate the capacitated problems as convex mixed-integer programs and show they are strongly NP-hard. To address these complex problems, we propose employing Lagrangian relaxation methods for analytically deriving lower bounds, establishing optimal multiplier ranges, and deriving worst-case error bounds. Additionally, we introduce heuristics to generate feasible solutions. Computational studies demonstrate that our relaxation-based heuristics produce solutions closely approaching optimality. Furthermore, we demonstrate that uncapacitated SI problems can be solved analytically, providing closed-form optimal solutions. This research contributes significant managerial insights for practitioners involved in supplier selection, distribution strategies, and inventory management within centralized and decentralized supply chains. Particularly noteworthy are our analytical findings on the benefits of centralization, especially under correlated Gaussian and heavy-tailed stable demand distributions. Our computational experiments further explore the influence of critical factors such as economies of scale, demand correlation, demand distribution characteristics, and upstream and downstream costs. Specifically, we assess how these factors impact the efficacy of centralization versus decentralized inventory management in minimizing expected inventory and overall system costs.
Teleoperated vehicles are a promising concept for increasing the attractiveness of car-sharing services. Such vehicles can be remotely steered by an operator to the location of a customer requesting a vehicle on demand. It therefore eliminates both the need for customers to walk to a car-sharing vehicle and the need for providers to relocate vehicles with drivers on-site to meet the temporal and spatial vehicle demand. The key to a successful teleoperated car-sharing service is to avoid longer service delays by effectively utilizing the fleet of vehicles and the limited number of available operators. The corresponding sequential decision process therefore involves a matching problem deciding which vehicle should be steered next by an available operator in order to fulfill which customer request. This decision is challenging, as both future demand and future availability of vehicles are uncertain, because the rental duration and return location of vehicles are unknown. We propose an approximate dynamic programming approach that combines predictions of future vehicle returns and customer requests with an approximation of the opportunity cost of matching decisions. We demonstrate the merits of our approach in comparison with benchmark policies in a comprehensive computational study based on New York demand data. We derive several important insights, among others, that vehicle predictions are especially valuable, and that a ratio of about one operator to six vehicles is sufficient in our setup.
