Joseph Geunes

Railcar switching, or shunting operations decisions play a significant role in the efficientoperation of railyard systems, which are in turn critical to the fast and effective movement ofgoods. In flat yards, switching operations are primarily performed using locomotives to pushand pull railcars in order to assemble and disassemble trains. In such settings, railcars withpredefined destinations are located across multiple parallel rail tracks, and must be moved, or switched, in order to form desired outbound trains. This study addresses the Railcar Assignment Problem (RAP) in flat yards with an objective of minimizing the total number of switching movements. We present a novel mixed-integer programming (MIP) model for this problem that incorporates practical operational constraints in rail yards, and demonstrate its NP-hardness. To solve large-scale instances, we propose a comprehensive Zone-based Double Deep Q-Network (Zone-DDQN) heuristic method that integrates railway structure, yard-zone decomposition, and a Double Deep Q-Network (DDQN). The yard-zone decomposition strategy partitions the yard into multiple parallel yard zones, after which the DDQN is applied to solve the problem within each zone individually and sequentially. Computational experiments across small-, medium-, and large-scale yards were conducted on a series of RAP instances. Average results show that the Zone-DDQN heuristic achieves an average optimality gap of 5.71% across small-scale yard instances. For large-scale yard instances containing more than 150 railcars and 30 tracks, the MIP model was not able to obtain solutions within 24 hours. In contrast, the Zone-DDQN heuristic was able to solve these instances with an average running time of 214.42 seconds.

Railcar shunting is a core planning task in freight railyards, where yard planners need to disassemble and reassemble groups of railcars to form outbound trains. Classification trackswith access from one side only can be considered as stack structures, where railcars are added and removed from only one end, leading to a last-in-first-out (LIFO) retrieval order. In contrast, two-sided tracks function like queue structures, allowing railcars to be added from one end and removed from the opposite end, following a first-in-first-out (FIFO) order. We consider a problem requiring assembly of multiple outbound trains using two locomotives in a railyard with two-sided classification track access. To address this combinatorially challenging problem class, we decompose the problem into two subproblems, each with one-sided classification track access and a locomotive on each side. We present a novel Hybrid Heuristic–Reinforcement Learning (HHRL) framework that integrates railway-specific heuristic solution approaches with a reinforcement learning method, specifically Q-learning. The proposed framework leverages methods to decrease the state-action space and guide exploration during reinforcement learning. The results of a series of numerical experiments demonstrate the efficiency and quality of the HHRL algorithm in both one-sided access, single-locomotive problems and two-sided access, two-locomotive problems

We consider a distribution network for delivering a natural resource or physical good to a set of nodes, each of which serves a set of customers, in which disruptions may occur at one or more nodes. Each node receives flow through a path from a source node, implying that the service at a node is interrupted if one or more nodes on the path from a source node experience a disruption. All network nodes are vulnerable to a future disturbance due to a potential natural or man-made disaster, the severity of which follows some measurable probability distribution. For each node in the network, we wish to determine a fortification level that enables the node to withstand a disturbance up to a given severity level, while minimizing the expected number of customers who experience a service interruption under a limited fortification budget. We formulate this problem as a continuous, nonlinear knapsack problem with precedence constraints, demonstrate that this optimization problem is $\mathcal{NP}$-Hard for general tree networks and general disturbance severity distributions, and provide a polynomial-time solution algorithm for serial systems, which forms the basis for an effective heuristic approach to problems on tree networks. Our computational test results demonstrate the ability of the proposed heuristic methods to quickly find near-optimal solutions.

A typical freight railyard at a manufacturing facility contains multiple tracks used for storage, classification, and outbound train assembly. Individual railcar storage locations on classification tracks are often determined before knowledge of their destination locations is known, giving rise to railcar shunting or switching problems, which require retrieving subsets of cars distributed throughout the yard to assemble outbound trains. To address this combinatorially challenging problem class, we propose a large-scale mixed-integer programming model that tracks railcar movements and corresponding costs over a finite planning horizon. The model permits simultaneous movement of multiple car groups via a locomotive and seeks to minimize repositioning costs. We also provide a dynamic programming formulation of the problem, demonstrate the NP-hardness of the corresponding optimization problem, and present an adaptive railcar grouping dynamic programming (ARG-DP) heuristic, which groups railcars with common destinations for efficient moves. Average results from a series of numerical experiments demonstrate the efficiency and quality of the ARG-DP algorithm in both simulated yards and a real yard. On average, across 60 test cases of simulated yards, the ARG-DP algorithm obtains solutions 355 times faster than solving the mixed-integer programming model using a commercial solver, while finding an optimal solution in 60% of the instances and maintaining an average optimality gap of 6.65%. In 10 cases based on the Gaia railyard in Portugal, the ARG-DP algorithm achieves solutions 229 times faster on average, finding an optimal solution in 50% of the instances with an average optimality gap of 6.90%.

This paper considers a disrupted infrastructure network where the repair crew knows the locations of service outages but not the locations of actual faults. Our goal is to determine a route for a single crew to visit and repair the disruptions to restore service with minimum negative impact. We call this problem the Traveling Repairman Network Restoration Problem (TRNRP). This problem presents strong computational challenges due to the combinatorial nature of the decisions, inter-dependencies within the underlying infrastructure network, and incomplete information. Considering the dynamic nature of the decisions as a result of dynamic information revelation on the status of the nodes, we model this problem as a finite-horizon Markov decision process. Our solution approach uses value approximation based on reinforcement learning, which is strengthened by structural results that identify a set of suboptimal moves. In addition, we propose state aggregation methods to reduce the size of the state space. We perform extensive computational studies to characterize the performance of our solution methods under different parameter settings and to compare them with benchmark solution approaches.

This paper considers orders that arrive one-by-one over time to a fulfillment center. Each order requests a product with some degree of customization that needs to be delivered expeditiously to a nearby location using a delivery vehicle. However, each vehicle can batch multiple orders together for delivery within a single trip. The benefits of batching include more efficient capacity utilization, lower total vehicle ownership requirements, and reduced environmental impact. The main drawback of batching is the consequent reduced average quality of service due to associated delivery delays when waiting for additional orders to arrive and executing a delivery route. To address this trade-off, we consider a set of threshold policies for batching and dispatching groups of orders, and characterize the associated long-run average cost per unit time for each policy that explicitly accounts for the customer’s total order lead time, including the time between order dispatch and delivery to the customer which, in turn, depends on route sequencing policies. For the threshold policies, our state variable may not only include the number of outstanding orders, but may also incorporate information on order arrival times and delivery locations. We model the stochastic dynamics of the system and obtain the long-run average cost per unit time, which we compute using a renewal-reward approach. We also consider different delivery sequencing approaches, including first-come, first-served and shortest traveling salesperson. In addition, we evaluate the effectiveness of accounting for all order information in the decision-making process, as opposed to just the number of outstanding orders or the time in the system for each order. Our analysis shows that a generalized class of cost- and quantity-based threshold policies often outperforms existing policies in the literature with the additional benefits of being robust to overestimates of the optimal cost threshold value and achieving strong delay cost performance. History: This paper has been accepted for the Service Science/Stochastic Systems Joint Special Issue. Supplemental Material: The online appendix is available at https://doi.org/10.1287/serv.2022.0042 .