ABSTRACT. In a logistics transport chain, multiple logistics service providers (LSPs) work together to deliver the cargo container on time to their final customers, the cargo owners. What usually happens is that these LSPs offer services with different quality levels and disclose information about those service quality levels to customers. Customers then choose services that best satisfy their requirements; and communicate their choices back to the LSPs. Many LSPs, therefore, strive to provide customers with feedback on their performance in order to assist them in making better decisions.

Customers' considerations when choosing suitable services lie mainly in their cargo urgency. On the one hand, they want to minimize total service costs incurred; on the other hand, they want to minimize total value lost from cargo depreciation, for instance, when the cargo sits inside containers for a long amount of time.

Given customers' considerations and LSPs' assistance, we want to further understand how these two elements - customers' cargo urgency and LSP's performance feedback influence customers' preferences for logistics services. We will study this problem by means of a laboratory experiment.

ABSTRACT. This paper addresses a highly constrained Facility Location-Routing Problem (FLRP) inspired by the strategic consolidation of multi-echelon supply chain networks under corporate mergers. When two autonomous logistics structures fuse, the optimization architecture must simultaneously determine the topological status of the existing warehousing assets—specifically evaluating options for physical integration, complete suppression, or singular hub centralization—while dynamically re-routing the associated delivery fleets. Given the NP-hard nature of the interconnected location and routing decisions, exact mathematical programming formulations encounter severe computational scaling limits.

ABSTRACT. Efficient aircraft turnaround operations are essential for maintaining airport punctuality and reducing operational costs. Among the critical ground handling activities, aircraft refueling requires careful coordination of limited fuel trucks while respecting strict departure schedules. Delays in refueling may propagate throughout the airport network, affecting subsequent flights and resource utilization. Despite its operational importance, aircraft refueling scheduling has received comparatively limited attention as a standalone optimization problem.

This work addresses the Aircraft Refueling Scheduling Problem (ARSP), in which a fleet of fuel trucks must service multiple aircraft located at different stands within predefined time windows. The objective is to determine the assignment of aircraft to fuel trucks and the sequence of refueling operations while minimizing total operational costs, truck travel times, and delay penalties. A mixed-integer linear programming (MILP) model is first developed to formulate the problem. To efficiently solve larger instances representative of busy airport operations, a matheuristic combining exact optimization with Adaptive Large Neighborhood Search (ALNS) is proposed. The exact model is used to optimize promising neighborhoods generated during the heuristic search, enabling high-quality solutions within practical computational times.

Computational experiments are conducted on problem instances of different sizes to verify the performance of the proposed matheuristic. For small instances, the results are compared with the optimal solutions obtained by the MILP model. For medium and large instances, the proposed approach is benchmarked against alternative metaheuristic algorithms and evaluated using the best bounds obtained by the exact solver. Solution quality, computational time, scalability, and optimality gaps are analysed to assess the effectiveness of the proposed method.

ABSTRACT. Climate-driven changes in the Arctic are opening new navigational opportunities that may reshape global shipping networks. Shorter Arctic passages offer reductions in distance, fuel use, and emissions compared with traditional canal routes. In this work, we address the liner shipping network design problem (LSNDP) when Arctic routes are available, incorporating their characteristics such as seasonal accessibility, reduced sailing speeds, and icebreaker fees. We extend existing benchmark instances to include Arctic connections and evaluate how both conventional and ice-going vessels can be deployed within such networks. Because Arctic routes are only viable during part of the year, we also investigate bi-seasonal network structures that coordinate a winter network relying solely on conventional routes with a summer network that activates Arctic alternatives. This approach allows rotations to shift seasonally with minimal redesign. Computational results demonstrate potential profitability gains and notable shifts in cargo flows when Arctic connections are strategically integrated.

ABSTRACT. The global shipping and transport network transports approximately eighty percent of all commercial products traded internationally. Seaports are a critical component of this global trade infrastructure. Seaport operations create major environmental issues contributing to climate change. By 2030, ports will need to reduce their greenhouse gas (GHG) emissions thirty percent from 2008 levels. Ports must also be GHG free by 2050. This study reviews 120 articles published between 2018 and 2025 regarding artificial intelligence (AI) and computer vision (CV) based applications for improving sustainability within port operations. Three new conceptual frameworks were developed as a result of these studies. The first was the Sustainability Intelligence Loop (SIL). SIL is a cyber-physical loop utilizing CV in real time to improve operational sustainability of seaports. The second was the Green Port Intelligence Maturity Model (GPIMM). GPIMM identifies five stages for seaports to become sustainable using AI and CV. The third was the Port Environmental Digital Twin (PEDT). PEDT is an extension of digital twins with the addition of five architectural components to enhance environmental performance. The study found that application of AI/CV-based technologies resulted in potential reductions in emissions ranging from ten to thirty eight percent , improved energy efficiencies from seven to eighty percent and reduced costs from ten to forty percent.

ABSTRACT. Artificial Intelligence and Supply Chain Resilience: Comparative Case Studies of Amazon and Alibaba The increasing complexity of global supply chains and the growing frequency of disruptions have highlighted the importance of resilience and agility in logistics systems. In this context, Artificial Intelligence (AI) has emerged as a strategic tool capable of enhancing decision-making processes, forecasting demand, optimizing inventory management, and improving operational responsiveness. This communication aims to explore the contribution of AI to supply chain resilience through a comparative analysis of two leading global e-commerce companies: Amazon and Alibaba. These organizations have invested heavily in AI-powered logistics solutions, including predictive analytics, intelligent warehousing, automated inventory management, route optimization, and real-time decision support systems. The study adopts a qualitative case-study approach based on secondary data from academic literature, corporate reports, and industry publications. The objective is to identify how AI technologies contribute to the anticipation of disruptions, operational flexibility, rapid recovery capabilities, and overall supply chain performance. Preliminary findings suggest that AI strengthens resilience by improving visibility across the supply chain, enabling proactive risk management, and supporting faster and more accurate decision-making. Amazon and Alibaba provide valuable examples of how AI-driven logistics ecosystems can enhance both agility and resilience in highly dynamic environments. The communication contributes to the ongoing debate on digital transformation in logistics and offers practical insights for organizations seeking to leverage AI to build more resilient supply chains.

ABSTRACT. With the urgency to address sustainability challenges on a global scale and the rapid digital transformation, the field of artificial intelligence (AI) in green supply chain management (GSCM) has become a rapidly growing research area. A bibliometric analysis of the scientific literature on AI-supported green supply chains is provided in this paper and is based on 768 peer-reviewed articles and review papers from the Web of Science Core Collection database. Using bibliometric tools such as Bibliometrix and VOSviewer, the study analyzes the publication trends, top journals, most important authors, geographical distribution and main research themes. The results show that the number of publications has increased exponentially after 2020, Asian economies dominate, sustainability and machine learning are the two key fields of research. There are six thematic clusters defined relating to AI-driven supply chain optimization, Industry 4.0 integration, applications of the circular economy and resilient logistics systems. The paper indicates research needs and suggests paths for future research such as increased methodological integration , and broadening of geographic and sectoral coverage of underrepresented areas.

ABSTRACT. The Vehicle Routing Problem (VRP) is a fundamental NP-hard combinatorial optimization problem with major implications for logistics, supply chain management, and urban freight distribution. Artificial intelligence (AI) techniques, including machine learning, deep reinforcement learning, and hybrid metaheuristics, have increasingly shaped VRP research in recent years. This paper presents a bibliometric review of AI-driven VRP studies with an emphasis on claimed real-world applicability, i.e., studies that explicitly mention practical, industrial, or case-study relevance. Using Web of Science and Scopus, a final corpus of 195 peer-reviewed journal articles published between 2015 and April 2026 was retained through a PRISMA-guided screening process. The analysis combines publication trend analysis, co-authorship mapping, keyword co-occurrence analysis, burst detection, and BERTopic thematic modeling, computed and visualized using PyBibX and VOSviewer. Results show a strong increase in AI-driven VRP research after 2021, with 2025 recording the highest annual output. Six thematic clusters are identified, led by hybrid AI-metaheuristics and deep reinforcement learning solvers. Four research gaps are discussed relative to real-world validation, explainability, ethical routing, and geographic under-representation. The findings offer a structured overview of the AI-VRP research landscape and directions for more transparent and industrially validated routing methods.

ABSTRACT. Safety-stock estimation from forecast errors depends on a historical sample that is both sufficiently long and representative of current forecasting performance. In practice, firms often use only recent error histories, implicitly assuming that recency guarantees representativeness. Our empirical study in an FMCG setting shows that this assumption is inadequate. Using the confidence-interval logic for service-level estimation (Walpole et al., 2012 chap. 9 [1]), a 98% target service level may require around 133 replenishment cycles, well above the one-year horizon currently used in practice. However, extending the history creates a second problem: forecast errors are not temporally homogeneous. We analyze weekly SKU-level forecast errors from six years of data and work only with calendar years to include all seasonal events (Hyndman & Kostenko, 2007 [2]). We exclude 2020 because it reflects a structural break associated with COVID-19 (Chatfield, 2000 chap. 6 [3]), and reserve 2025 for discrete-event simulation validation, leaving four years for statistical analysis. Nonparametric comparisons across years using Kruskal-Wallis and Kolmogorov-Smirnov tests show that only a small fraction of SKUs remain homogeneous for more than three years. This rules out pooling past errors indiscriminately and motivates a longitudinal analysis of how error distributions evolve over time. We propose a robust, nonparametric framework (Hampel et al., 1986 [4]) to characterize this evolution through two yearly indicators per SKU: the absolute median, as a measure of central bias, and the interquartile range, as a measure of dispersion. Linear trends on these indicators reveal a damped oscillatory behavior in forecast errors: in 61% of SKUs both bias magnitude and dispersion decrease over time; in nearly 90% of SKUs, at least one of these two indicators improves. This pattern is consistent with a learning process in which forecast errors become progressively more centered and less variable. These findings have direct implications for inventory control. If extreme historical errors are transient and become less representative as the forecasting system learns, retaining them mechanically in safety-stock calculations may overstate protection and reduce efficiency. The results therefore support cleansing strategies based on distributional influence, aimed at distinguishing structural variability from non-recurrent extremes when estimating safety stock.

ABSTRACT. Truck Appointment Systems (TAS) are widely used to regulate truck arrivals at container terminals and mitigate congestion. However, most existing approaches focus on appointment allocation, scheduling, and capacity control while treating truck arrivals as exogenous. As a result, limited attention has been given to demand-side congestion management through behavioural responses to pricing signals.

This study presents a behaviour-aware decision-support model for congestion management in container terminal truck operations. Using operational truck and straddle-carrier data from the Port of Southampton, a behavioural nonlinear optimisation model has been developed in which truck arrival choices are represented through a multinomial logit formulation grounded in random utility theory. Congestion is represented through an empirically calibrated bottleneck waiting-time function informed by OLS and Random Forest analyses of operational data. The model determines time-dependent prices that influence truck arrival decisions while balancing three competing objectives: revenue generation, congestion reduction, and demand retention through explicit consideration of diversion to outside alternatives.

Results demonstrate that pricing can redistribute demand away from peak periods and reduce waiting times by encouraging shifts toward less congested arrival windows. The strongest effects are observed under lower-demand operating conditions where greater spare capacity is available. By reducing truck idling and congestion, the proposed approach can contribute to more sustainable port operations through lower fuel consumption, reduced emissions, and improved local air quality.

The study also highlights limitations of static congestion pricing approaches. While the model captures behavioural responses and congestion effects within a representative operating day, it does not fully represent congestion spillovers and queue accumulation across consecutive periods observed in terminal operations. Nevertheless, the results demonstrate the potential of integrating behavioural choice modelling, data-driven congestion characterisation, and optimisation within a practical decision-support framework for port logistics.

ABSTRACT. Emergency procurement operates under severe time pressure, uncertainty and resource constraints, creating tensions between rapid humanitarian response and fiduciary accountability. Existing procurement governance frameworks recognize the need for procedural flexibility during crises but provide limited analytical guidance on how governance controls should be reconfigured when conventional oversight mechanisms become operationally infeasible.

This paper views emergency governance as an adaptive reallocation of controls rather than a relaxation of accountability. It develops a stochastic decision-support framework for emergency procurement governance, modeling governance control allocation as a sequential decision problem under uncertainty. Governance decisions are represented within a finite horizon stochastic control framework that captures the dynamic trade-off between preventive procedural controls and accountability-oriented oversight mechanism.

To examine how different governance philosophies perform under uncertain emergency conditions, the framework is implemented as a simulation-based decision-support model comparing representative governance regimes. The analysis focuses on evaluating alternative governance configurations and their implications for accountability and humanitarian responsiveness.

Simulation results show substantial variation in the performance of the evaluated governance regimes. Rigid compliance approaches achieve lower integrity risk but generate significantly higher humanitarian delays, whereas low-control governance approaches improve responsiveness at the expense of accountability. In the reference simulation, the accountability dominant regime achieves the lowest overall loss. The adaptive governance regime performs better than rigid compliance across the volatility scenarios examined, and its relative advantage increases as emergency volatility rises.

Methodologically, this paper introduces a stochastic governance-control allocation framework for evaluating representative governance policies under uncertainty. Practically, the findings highlight the role of digital governance infrastructures, including traceability, auditability, and workflow visibility, in maintaining accountability while enabling rapid emergency response.