ABSTRACT Strict loading requirements necessitating the consideration of three‐dimensional attributes of items and vehicles exist in specific vehicle routing scenarios. These requirements dictate the simultaneous generation of the loading plan and route of each vehicle, that is, solving the integration of the capacitated vehicle routing problem (CVRP) and the container loading problem (CVRP with three‐dimensional loading constraints, 3L‐CVRP). Over the two decades since the inception of the 3L‐CVRP, a substantial body of literature has emerged. These studies encompass diverse problem variants, modelling characteristics, and algorithmic approaches, while utilising varying benchmark instances. To systematise this expansive landscape, this study provides a comprehensive review structured as follows. First, existing 3L‐CVRP variants are distinguished and classified based on their routing and loading characteristics. Second, existing modelling frameworks are summarised and compared. Third, solution approaches, including routing and loading algorithms and methods for integrating them, are reviewed and classified. Finally, benchmark instances are summarised, detailing their scales, characteristics, applicable scenarios, and previous results. Based on this comprehensive synthesis, current challenges and promising future research directions are highlighted.

Addressing nonlinear fluctuations in inbound logistics demand in the intelligent automotive industry, this paper proposes a SARIMA-LSTM-Attention forecasting model incorporating Real-Time Production Progress Feedback (RPF) features. By introducing production progress deviation into a hybrid framework that combines SARIMA-based linear forecasting with attention-enhanced LSTM residual correction, the proposed model effectively captures both linear trends and nonlinear demand disturbances. The model was evaluated using 995 daily observations collected from the inbound logistics system of a large new energy vehicle manufacturer from January 2023 to July 2025, with data from January 2023 to December 2024 used for training and validation and January to July 2025 reserved for testing. Compared with six representative benchmark models, the proposed model achieved the best overall performance, reducing sMAPE from 43.23% to 33.13% relative to the SARIMA baseline, representing an absolute reduction of 10.10 percentage points and a relative improvement of 23.36%. These results demonstrate the effectiveness of integrating real-time production feedback for demand forecasting and provide practical support for lean inventory management and logistics decision-making in automotive supply chains.

In emergency logistics, road damage can make areas inaccessible to trucks, while rescuepersonnel, casualties in transit, and evacuees may remain mobile, causing their service locations to change over time. Conventional fixed-node delivery models are therefore not well suited to such settings. We study the moving-target truck-multi-drone mothership routing problem (MT-TMMRP), in which one truck serves as a mobile launch, recovery, and charging platform for two homogeneous drones operating in parallel. The problem jointly determines truck movement, drone task assignment, target interception, truck-drone meetings, and battery management to minimize total mission completion time. To represent these coupled operations, we formulate a continuous-space optimization model and an event-driven cooperative Markov game, and propose CAMS-MAPPO. The method uses environment feature encoding to capture relations among the truck, drones, and targets. It then employs the context-aware matching and scoring(CAMS) module to evaluate candidate support and service actions, while multi-agent proximal policy optimization (MAPPO) learns a feedback policy reusable across instances. On small instances for which Gurobi proves optimality, CAMS-MAPPO achieves an average objective value gap of only 1.48% and reduces average online computation time by 97.64%. Across tests with different customer scales, the same policy parameters are reused without retraining and yield lower average completion times than the constructed ALNS baseline at every tested scale. Average online computation time remains below 0.7 s and is at least 99.88% lower than that of ALNS. Further analyses show diminishing returns from increases in battery capacity and the drone-to-truck speed ratio, whereas higher energy consumption extends mission completion time. Target speed, travel distance, and origin-destination distributions affect truck-drone routes and delivery performance by altering interception time windows and the operational range.

In the truck-drone collaborative delivery (TDCD) system, vehicle interaction and drone recharging are typically constrained by static, predefined nodes, inevitably incurring non-productive waiting time (NWT), such as waiting interaction and waiting charging, thereby hindering system delivery efficiency. To overcome the interaction constraints of static, predefined nodes, eliminate NWT, and achieve wait-free collaboration between the truck and drone within the continuous spatiotemporal domain, we propose a novel TDCD mode termed Cruise-Recharging Side-Delivery (CRSD) and define the optimization problem as the Continuous Spatiotemporal Truck-Drone Traveling Salesman Problem with En-route Interaction and Recharging (CTSPD-EIR). In this mode, the truck maintains a continuous cruising delivery state, enabling the drone to execute en-route launch, recovery, and recharging operations without necessitating stationary waiting for interaction or recharging, thereby eliminating NWT. To address the modeling complexities and computational challenges arising from the strong coupling between the truck and drone in the continuous spatiotemporal domain, we formulate the CTSP-D-EIR as an event-driven Markov game and propose the DRA-MADDPG to solve it. This algorithm integrates dynamic feasibility action masking to filter out invalid actions, reward-modulated prioritized experience replay to suppress noise samples for algorithm stability, and adaptive oversampling to enhance the utilization of high-reward samples during the early training. Extensive numerical experiments demonstrate that by eliminating NWT, the proposed CRSD mode achieves average reductions of 12.69% in total delivery cost and 16.17% in total delivery time, significantly outperforming various existing TDCD models. Furthermore, our proposed algorithm demonstrates high computational efficiency in solving this problem.

ABSTRACT Efficient transportation of engineering materials is critical to the timely execution of large‐scale infrastructure projects, yet it is often challenged by heterogeneous material demand, multimodal transitions, dual‐layer capacity limitations (transport arcs and transshipment nodes), and schedule‐dependent time‐window requirements. To address these coupled decision features, this paper formulates an Engineering Material Multimodal Transport Path Planning model with Multi‐Commodity Flow under Time Window and Network Capacity (EM‐CMTP3MCTW), which minimizes a weighted‐sum objective of total transportation cost and total transportation time (including time‐window‐related penalties). To solve this NP‐hard problem at practical scales, we develop an improved parallel adaptive large neighbourhood search algorithm (GA‐PALNS) equipped with problem‐specific destroy‐and‐repair operators and a parallel evaluation mechanism. Computational experiments on randomly generated instances show that GA‐PALNS achieves lower objective values than representative single‐objective baselines (average improvement approximately 0.48%) while reducing runtime by roughly one order of magnitude. A real‐world case study from a major infrastructure project in Western China further demonstrates that the proposed framework can generate feasible and economically competitive multimodal transportation plans under complex capacity and time‐window constraints.

As urbanization accelerates, urban public transportation systems are tasked with meeting the growing demand for travel. According to statistics, the regional rail transit network currently handles 60% of the passenger volume for trips originating from around the stations, and this is projected to increase by over 20% within the next decade. However, the current feeder bus services suffer from insufficient coverage, leading to inconvenience for passengers and impacting the overall efficiency of the transportation system. To enhance the public transit service coverage along a rail transit corridor, this study designs feeder bus services to connect the rail stations. A theoretical trunk-feeder transit design problem is firstly solved to furnish the initial feeder bus line designs. The objective is to minimize total generalized system cost with respect to the density of feeder bus routes and their service frequencies. The idealized design of the optimal bus route density function is then discretized into specific locations and fine-tuned with Geographic Information System (GIS) tool. Considering local conditions like road network and demographic information, a four-step adjustment strategy is proposed to furnish the implementation-ready design of feeder bus services. Numerical results show the final design renders at least 3.5% saving in total system cost. This study illustrates how the state-of-the-art theories of transit design can be applied into practice and highlights the connection between theoretical studies and practical application.

With the rapid development of low-altitude economy and unmanned aerial vehicles (UAVs) deployment technology, aerial-ground collaborative delivery (AGCD) is emerging as a novel mode of last-mile delivery, where the vehicle and its onboard UAVs are utilized efficiently. Vehicles not only provide delivery services to customers but also function as mobile warehouses and launch/recovery platforms for UAVs. This paper addresses the vehicle routing problem with UAVs considering time window and UAV multi-delivery (VRPU-TW&MD). A mixed integer linear programming (MILP) model is developed to minimize delivery costs while incorporating constraints related to UAV energy consumption. Subsequently, a micro-evolution augmented large neighborhood search (MEALNS) algorithm incorporating adaptive large neighborhood search (ALNS) and micro-evolution mechanism is proposed. Numerical experiments demonstrate the effectiveness of both the model and algorithm in solving the VRPU-TW&MD. The impact of key parameters on delivery performance is explored by sensitivity analysis.

The rapid evolution of unmanned aerial vehicle(UAV)technology and autonomous capabilities has positioned UAV as promising last-mile delivery means. Vehicle and onboard UAV collaborative delivery is introduced as a novel delivery mode.Spatiotemporal collaboration, along with energy consumption with payload and wind conditions play important roles in delivery route planning. This paper introduces the traveling salesman problem with time window and onboard UAV(TSPTWOUAV) and emphasizes the consideration of real-world scenarios, focusing on time collaboration and energy consumption with wind and payload. To address this, a mixed integer linear programming(MILP) model is formulated to minimize the energy consumption costs of vehicle and UAV. Furthermore, an adaptive large neighborhood search(ALNS) algorithm is applied to identify high-quality solutions efficiently. The effectiveness of the proposed model and algorithm is validated through numerical tests on real geographic instances and sensitivity analysis of key parameters is conducted.

To support intelligent transportation systems and autonomous freight rail development, in this study, we have developed a novel optimization framework for dynamic train formation planning. It optimizes network-dimension decisions including railcar routing, blocking, train makeup, and train routing, while incorporating railcar-to-track assignments in technical yards. This integrates decisions across previously isolated network and yard dimensions, addressing imprecision and infeasibility caused by conceptual reduction in existing frameworks. Furthermore, the framework subdivides schedule length to incorporate railcar and train scheduling, responding to demand fluctuations. A state-space-time network captures spatiotemporal characteristics of railcars and trains. A practical method that partitions classification tracks into two states is implemented to maintain a compact network scale. An integer-linear programming model is developed to minimize overall transportation, operation, and delay costs. Practical factors and operational methods are considered to ensure feasibility and effectiveness of dynamic train formation planning. A multi-variable-exploration branch-and-price algorithm is designed to solve the model, which estimates and branches on multiple variables in each iteration for promising convergence progress. A general constraint-driven coactive branching technique is embedded to impose multiple branches simultaneously to strengthen lower bounds. Numerical experiments based on real-world railroad networks with up to 21 yards and 964 shipments (28 473 railcars) assess the framework’s performance and practicality. The improved branch-and-price algorithm demonstrates significant improvements over conventional approaches, reducing total cost and computational time by up to 4.13% and 98.40%, respectively.

In capacitated vehicle routing problem with three-dimensional loading constraints (3L-CVRP) that combines the container loading and capacitated vehicle routing problems, the relationship between vehicle visiting sequence and item loading sequence is reflected through relocation-ban constraint. This widely applied constraint prohibits the temporarily unloading and repositioning of loaded items during the entire transportation process, simplifying loading operations but also limiting the volume utilization of each vehicle and increasing transportation costs. To address this issue and obtain a trade-off between transportation cost and operational complexity (reflected in relocation cost), two improved relocation constraints that seek to allow necessary and restricted relocations are developed in this study. Under pickup scenario, a mixed integer-linear programming model is developed to describe the 3L-CVRP with the relocation constraints. An improved branch-and-price algorithm is employed to solve the model. Two loading algorithms, incorporated with a backward dynamic programming method, are proposed to simultaneously generate loading and relocation plans. Enhancement strategies, including an improved label-correcting-based algorithm and memory components that collect loading feasibility and relocation cost information, are developed. Numerical experiments were designed to test the performance of the proposed algorithms and validate the significance of necessary relocations. In pure loading instances, necessary relocations bring an increase in volume utilization by 3.75 % on average and 36.48 % at maximum. In large-scale benchmark instances, allowing necessary relocations decrease the overall costs by 4.86 % on average and 13.08 % at maximum. Abbreviations: 3L-CVRP, capacitated vehicle routing problem with three-dimensional loading constraints; CLP, container loading problem; CVRP, capacitated vehicle routing problem; RC, relocation constraints; MILP, mixed integer linear programming; B&P, branch and price; BDP, backward dynamic programming; D-W, Dantizig-Wolfe; ESPPRLC, elementary shortest path problem with resource and loading constraints; LCA, labelcorrecting-based algorithm; 3L-PDP, pickup and delivery vehicle routing problem with three-dimensional loading constraints; RMP, restricted-master-problem; SP, sub-problem; IGHA, improved greedy heuristic algorithm; ITRSA, improved tree search algorithm; TN, tree node; BR, Bischoff & Ratcliff.

Effectively addressing the surge in passenger flow caused by emergencies represents a critical challenge in the emergency management of urban rail transit operations. This article introduces a comprehensive control strategy for passenger evacuation, employing robust optimization methods to tackle the uncertainties associated with sudden increases in passenger numbers. Initially, a comprehensive mathematical optimization model for train scheduling and passenger flow coordination is established to ensure efficient passenger transport while maximizing the reduction of operational costs for the operating company's emergency response. Subsequently, the impact of uncertainty factors on the evacuation model is considered. The fluctuation in passenger flow is represented using a range of intervals, and a robust corresponding model is formulated by introducing an uncertainty budget coefficient. Furthermore, small-scale numerical examples are utilized to discuss passenger evacuation plans, conduct robustness analysis, and assess demand sensitivity. The practical case of the Beijing subway demonstrates that, in contrast to the most optimistic scenario, the robust passenger plan, accounting for a 1% fluctuation in passenger flow, exhibits a reduced cost increase from 10.96% to 9.13%, as compared to the most conservative situation.These findings contribute to enhancing the emergency response capabilities of operational management departments.

This study investigates a Feasibility-assured Mothership System (FAMS) model for truck-drone hybrid delivery. As a mothership system variant, it is considered to be able to bring benefits by using drones for low-cost short distance transportation. Besides, it depicts a scenario aligning closely with the current features of urban delivery services and adopts methods to increase drone utilization. Most importantly, the model addresses potential infeasibility resulting from assumptions in existing relevant literature. In this study, the FAMS is formulated as a mixed integer linear program (MILP) model and micro-evolutionary algorithm (MEA), a population-based algorithm that captures the structural characteristics of individuals in order to find high-quality solutions more efficiently, is proposed. The experimental results demonstrate the effectiveness of the algorithm. The performance of the two proposed crossover operators is analyzed as well. Furthermore, the cost efficiency of the FAMS under numerous situations including different truck-drone unit travel cost ratios and various combinations of drone technical features is confirmed.

With the continuous development of large-scale engineering projects such as construction projects, relief support, and large-scale relocation in various countries, engineering logistics has attracted much attention. This paper addresses a multimodal material route planning problem (MMRPP), which considers the transportation of engineering material from suppliers to the work zones using multiple transport modes. Due to the overall relevance and technical complexity of engineering logistics, we introduce the key processes at work zones to generate a transport solution, which is more realistic for various real-life applications. We propose a multi-objective multimodal transport route planning model that minimizes the total transport cost and the total transport time. The model by using the ε − constraint method that transforms the objective function of minimizing total transportation cost into a constraint, resulting in obtaining pareto optimal solutions. This method makes up for the lack of existing research on the combination of both engineering logistics and multimodal transportation, after which the feasibility of the model and algorithm is verified by examples. The results show that the model solution with the introduction of the key processes at work zones produces more time-efficient and less time-consuming route planning results, and that the results obtained using the ε − constraint method are more reliable than the traditional methods for solving multi-objective planning problems and are more in line with the decision maker’s needs.

In order to enhance the operational performance of urban rail transit networks under overloaded conditions, this study proposes a novel network-level collaborative passenger flow strategy based on a sliding window mechanism. The main idea is to categorize arriving passengers at stations into different groups using variable-length windows and allocate them to specific trains. With the aim of minimizing the total passenger extra waiting time under cost budget, we establish a mixed integer programming model with sliding window constraints, flow distribution constraints, and node flow constraints to build the optimal matching relation between passenger flow demands and transportation resources. Specifically, the model incorporates precise quantity constraints related to transfers, facilitating the accurate quantification of both internal and external demands of network passenger flow. Then we use logical constraint transformation and linearization techniques to enable the model to be solved directly by commercial optimization solvers. For large-scale problems, we adopt a rolling horizon approach to improve computing efficiency. To validate the effectiveness of the sliding-window based passenger control method, we conduct case studies including a small designed example and real-world research on the Beijing Subway network. The experimental results demonstrate that our proposed model assists managers in developing flexible and accurate network-level collaborative passenger flow control schemes.

技术站是铁路网的重要组成部分,其作业组织水平直接决定了铁路运输生产效率.传统的技术站配流问题通常考虑车站内部作业优化,如解编顺序、分类线运用等,以减少车辆在站停留时间.为加强站间互联互通,促进车流有序流动,提出基于货物列车编组计划的多站协同概念.建立以车辆在站停留时间最小为目标的考虑多台调机的单技术站动态配流模型M1和以协同后减少的停留时间之和最大、新增列车走行费用最小为目标的多站协同优化模型M2.考虑到模型M1属于NP-hard问题,设计拉格朗日松弛算法将模型分解为3个子问题求解.模型M2根据模型M1求得结果并结合多站协同概念,调用GUROBI求解器求解验证.通过案例分析可知:在计划时限内,给定的路网中共有6个技术站进行协同配流,通过改变欠轴停运列车的编组去向,实现相邻车站车流供给,车辆在站停留时间减少645.1 h,较协同前降低6.57％,增加6列正点出发列车.研究结果表明,提出的模型及算法能够压缩多站车辆在站停留时间,提高计划兑现率,增强铁路货物运输生产效率.

The capacity bottleneck of a railway hub is one of the main factors leading to the backlog of traffic and reduced efficiency of network transport. Based on the analysis of the characteristics and mechanism of capacity bottlenecks, this paper summarized the bottleneck influencing factors in three modules of congestion intensity, congestion duration and congestion trend, and then extracted the available data indicators and the importance criteria of each indicator. Based on the above indicators, this paper used Genetic Algorithm（GA） to optimize the hyperparameters of Long Short-Term Memory Network（LSTM） and designed a bottleneck identification model based on GA-LSTM network. Finally, a case study based on the actual data of a railroad hub was conducted to compare various deep learning methods, which proved better effect of the GA-LSTM network in identifying bottlenecks. The application scope of the model framework was given, including real-time identification of bottlenecks and simulation of the changes of different indicator data to carry out bottleneck elimination in advance.

With multiple stations being built in the high-speed railway hub, how to schedule trains among stations is of great significance. This paper studies the integrated train routing and timetabling problem in a multi-station high-speed railway hub. Based on a specific bi-level network, a binary integer programming model using cumulative flow variables is proposed to minimize the total cost of trains and passengers. An extended Alternating Direction Method of Multipliers (ADMM) algorithm is developed to decompose the model into train and passenger subproblems. The performance of the algorithm is investigated on small-scale and large-scale networks. Compared with the commonly used Lagrangian Relaxation (LR) algorithm, the results indicate that the extended ADMM obtain better solutions with less iterations, where the lower bound decreases by 3.79% and iterations are reduced by 83.54%. Another comparative experiment shows the total cost of the optimization scheme is 16.22% lower than that of the current fixed scheme.

Transportation enterprises have started to pay more attention to the vast amount of historical operational data. In this paper, real route information is extracted from the freight car trajectory data and a satisfied route is proposed based on this information. The satisfied route not only can reduce the travel cost but also conform to the actual transportation situation. A solution framework driven by trajectory data is developed to obtain the satisfied route. A case study associated with the Chinese railway is used to verify the model and the solution framework proposed in this paper.

AbstractWith the emergence of new information and communication technology, mobile‐based ridesharing services have received more attention and have improved. They offer timely and convenient service and exert an enormous influence on pollution and congestion by allowing passengers to share overlapping routes. However, ridesharing services have also been discussed and called into question by people with respect to their attractiveness, their comfort, and passenger safety. This paper proposes a ridesharing algorithm with a perceived mechanism by introducing multiple dimensions of perceived value to overall assess passengers’ perceptions. The approach involved can continuously track the psychological changes in passengers during travel, capture the dynamic and comprehensive perceptions of passengers, and make real‐time adjustments on this basis. Furthermore, the proposed partition‐based parallel strategy can ensure high‐quality services and further enhance computational efficiency. Finally, numerical experiments are conducted based on the urban transportation network and the actual trip data from Manhattan to evaluate the performance of the proposed algorithm. The results show that it is important to effectively capture the passengers’ perception of ridesharing services and apply them to optimization mechanisms, which will contribute to improving the perception of ride experiences and attracting potential passengers.

In order to improve the accuracy of short-term prediction of parking space, based on the data characteristics of free parking space in parking lots, a combined prediction model combined whale optimization algorithm（WOA） and extreme gradient boosting algorithm（XGBoost） was proposed. Firstly, the randomness of the data of free parking space was analyzed, and the singular spectrum analysis（SSA） was used to deconstruct and reconstruct the original data, so as to extract the main components of the original data and eliminate the noise. Secondly, the whale optimization algorithm was used to optimize the main parameters of XGBoost prediction model and find the global optimal parameters. Finally, the accuracy of the proposed prediction model was verified by an example. The test results show that the WOA-XGBoost prediction model that optimizes parameters has higher prediction accuracy and stability.