Rapid urbanization and growing mobility demand are reshaping transportation systems, calling for more advanced intelligence and management capabilities. Artificial intelligence (AI) has emerged as a key enabler for enhancing perception, prediction, and decision-making in transportation. This paper presents a systematic review of AI applications across four major transportation domains: road, rail, air, and maritime systems. Rather than exhaustively surveying all published studies, this review adopts a thematic synthesis approach, organizing representative, recent research by major transportation modes and core AI application scenarios, with an emphasis on influential studies published in leading journals and conferences. The review examines representative applications and key functionalities within each domain, highlighting how AI techniques—such as deep learning, graph-based models, reinforcement learning, and emerging foundation models—are adapted to diverse transportation contexts. Furthermore, this paper analyzes key challenges, including data quality and sparsity, interpretability, uncertainty, and cross-domain generalization, and discusses emerging research directions such as foundation models, physics-informed learning, and continual adaptation. By integrating insights from both methodologies and real-world applications, this review provides insights for advancing intelligent, scalable, and resilient transportation systems.
The accelerated servitization of manufacturing and the rapid advancements in emerging information technologies have enabled flexible process configuration, efficient resource collaboration, and dynamic value chain reconstruction. However, the complex behavioral interactions among users, firms, and platform systems, alongside massive data flows, pose significant challenges to the operational management of service-oriented manufacturing (SOM). This study conducts a systematic literature review of data-driven, behavior-based SOM using a mixed-methods approach that integrates Bibliometric-Systematic Literature Review (B-SLR) and snowballing techniques. Drawing on an in-depth analysis of 367 core publications from 1988 to April 2026, this study develops a lifecycle-based framework comprising four key operational stages, covering product design, production, smart operations and maintenance, and product lifecycle management, including recycling and remanufacturing. The review synthesizes major findings from both academic and practical perspectives and identifies future research directions across four behavioral dimensions, including value co-creation in design, collaborative sharing in production, human–AI interaction in operations, and multi-agent collaboration in lifecycle management. The findings provide practical insights that advance both research and practice in SOM to support the development of resilient, sustainable, and intelligent manufacturing systems.
The functional robustness and structural integrity of real-world networks disproportionately on a subset of influential nodes. Identifying these influential nodes is a central objective in network science, underpinning applications spanning from epidemic control to predicting high-impact journal publications. In this study, cascade mitigation is defined as a pivotal, yet overlooked, canonical problem in network science that has received comparatively limited attention, and a generalized benchmarking framework is established to unify this interdisciplinary field. A decade of progress is synthesized, tracing the evolution from classical structure-centrality-based to artificial intelligence (AI)-based influential-node identification techniques. Beyond providing a systematic survey, nearly 20 distinct methodologies were empirically evaluated across diverse synthetic and real-world data sets using standard spreading models such as the susceptible-infected-recovered, linear threshold, and independent cascade models. This large-scale cross-domain comparison elucidates the performance differences between structural and AI-based strategies. By standardising the evaluation of node influence, this review offers a critical roadmap for researchers and identifies promising avenues for future interdisciplinary innovations.
Data-driven decision-making plays an increasingly important role in engineering management and complex operational systems under uncertainty and dynamic environments. This article reviews the major paradigms in data-driven optimization, including offline learning and stochastic optimization, robust and distributionally robust optimization under small-data regimes, and adaptive online and reinforcement learning approaches. We examine the methodological foundations of these paradigms and discuss their applications in engineering management contexts. Finally, we highlight emerging research directions at the intersection of artificial intelligence and decision-making.
With the rapid expansion of urban scale, traditional Ground-Based Response Units (GBRUs) face increasing challenges in maintaining efficient response times for traffic incidents, particularly under congestion and spatiotemporal uncertainty. This study develops an application-oriented strategic planning framework for an Unmanned Aerial Vehicle (UAV)-Assisted Traffic Incident Management System (UATIMS). The framework combines data-driven robust facility location with simulation-based operational validation. To represent the heterogeneous attributes of urban traffic incidents, we introduce an Urgency-Importance-Ambiguity (UIA) assessment framework that translates incident severity, network impact, and early-stage information uncertainty into plan-ning-relevant risk categories. Based on this characterization, a weighted structural uncertainty set is constructed and embedded into a min-max robust location model to support capacity allocation under long-tail demand risks. The model incorporates a congestion-aware stability constraint and is reformulated into a tractable Mixed-Integer Linear Programming (MILP) model through strong duality. Using real-world traffic incident data from Hangzhou, China, we evaluate the proposed framework through computational experiments and Discrete Event Simulation (DES). The results show that the Full UIA model provides incremental value over homogeneous and urgency-weighted robust baselines, especially in reducing high-urgency service shortages and tail response risks under resource-constrained conditions. The simulation further indicates that the resulting UAV-assisted layout can achieve high net service coverage and maintain average response times within an operationally acceptable range. These findings suggest that heterogeneous risk representation can complement existing drone-service design models by supporting strategic hangar siting and fleet allocation for traffic incident management.
Zero-carbon industrial parks are core demonstration carriers for global industrial deep decarbonization and a key research hotspot in climate change and sustainable development. However, three structural issues have long restricted the formation of a globally comparable research paradigm: fragmented accounting benchmarks, imbalanced research priorities, and poor generalizability of findings. This comment systematically analyzes the above deviations: divergent accounting rules across mainstream frameworks weaken cross-study comparability; studies overfocus on energy system transition while neglecting core process-level decarbonization; case-specific conclusions cannot adapt to the transition needs of developing economies. To correct these deviations, this comment proposes a universal minimum consensus benchmark, advocates a rebalanced research agenda, and outlines a differentiated globally adaptable framework with seven priority propositions, to advance the systematic development of the field.
Under the “One Country, Two Systems, Three Legal Jurisdictions” framework, cross-regional collaboration in the construction industry of the Guangdong–Hong Kong–Macao Greater Bay Area (GBA) is more complex than other urban agglomerations, and this makes traditional methods relying on manual analysis of influencing factors inefficient and subjective. While existing large language models (LLMs) can meet the needs of intelligent applications, they lack specific domain knowledge. Considering the intelligent advantages of LLMs, this research proposes a lightweight expert agent through providing an optimised low-rank adaptation (LoRA) model SVDSR-LoRA, integrating domain knowledge using the fine-tuning method. Experiments on the 1.5b lightweight base model of qwen2.5 and deepseek-r1 show that the proposed SVDSR-LoRA training method can increase the mid-term convergence speed by 36
Decision-makers in contemporary operational settings must navigate uncertainty arising from rapidly shifting contextual factors such as volatile market dynamics and intricate supply-chain interactions. Although distributionally robust optimization offers a principled means of guarding against distributional ambiguity, conventional formulations typically ignore the contextual information that shapes uncertainty in real applications. This paper outlines the emerging framework of Contextual Distributionally Robust Optimization (CDRO), which embeds covariate information into robust decision-making. We cast CDRO as a conditional worst-case optimization problem and organize existing research into two overarching streams: (1) predict-then-robustify, where ambiguity sets are constructed around estimated conditional distributions or moments; and (2) joint contextual robustification, where robustness is imposed on the joint distribution of covariates and uncertain parameters. Illustrative studies in transportation, healthcare, and inventory systems reveal that incorporating contextual information can substantially enhance adaptability and reliability in decision policies. The paper concludes with a discussion of open challenges, including data fidelity, causal confounding, computational scalability, privacy concerns, and model interpretability, that highlight promising directions for future research.
Urban air mobility (UAM) demands reliable sensing and communication to ensure safe and efficient electric vertical take-off and landing (eVTOL) operations in complex urban environments. This paper proposes a multitier tower base station (TBS) and tethered unmanned aerial vehicle (TUAV)-assisted integrated sensing and communication (ISAC) framework for the eVTOL approach and landing, where the rigid shapes of eVTOLs are represented by the millimeter wave (mmWave) radar point clouds rather than simplifying them as a single point, which is closer to the real world. Then, an integrated radar–communication fair optimization problem is formulated to jointly maximize radar and channel capacities while satisfying tether, safety, and terminal constraints. To address the randomness of radar point clouds, a point-cloud-aware deep reinforcement learning (PCDRL) method is developed to solve the above optimization problem. The proposed framework extracts radar point clouds features and learns adaptive TUAV 3D trajectory control and dynamic power ratio allocation across multiple eVTOLs. Simulation results demonstrate that PCDRL improves both radar and communication performance, achieving challenging performance among all the benchmarks.
Frequent extreme weather events not only threaten the safety of subway engineering construction but also pose severe challenges to project organizations. Facing continuous external shocks, enhancing organizational resilience (OR) has become a promising way for subway projects to effectively address climate risks. However, the existing theoretical understanding of OR in subway projects remains limited, and there is a lack of empirical evidence explaining how OR can be systematically constructed to cope with extreme weather. To this end, this study employed web crawling and expert interviews to collect multi-stakeholder interview data and diverse historical materials that characterize OR in subway projects under extreme weather events. Based on this, topic modeling techniques and focus group discussions were employed to model and optimize multi-source data, analyzing the structural, distributional, and multi-stakeholder differentiation characteristics of OR in subway projects. The findings show that OR in subway projects represents a collective capability composed of perceptual, planning, coping, adaptive, recovery, and bounce-forward learning capabilities, encompassing 43 dimensions. mong these, planning capability serves as the primary foundation for shaping OR, while bounce-forward learning capability provides critical support for transitioning from reactive response to proactive leapfrogging. Additionally, the five key stakeholders exhibit different patterns of OR, yet each plays a unique role in collaboratively constructing OR. This study not only advances the theoretical understanding of the multidimensional characteristics of OR in subway projects but also provides practical insights for project stakeholders in formulating targeted strategies to enhance resilience.
Human–AI collaboration is becoming central to operational prediction, decision-making, and control. As AI systems become embedded in operational workflows, organizations face a recurring risk: decision support may improve while the practical space for human intervention, accountability, and learning becomes too constrained for meaningful contestation. This article develops an operational contestability framework to explain when and why this risk becomes collaboration failure. The key question is not simply whether humans remain formally “in the loop,” but whether AI-enabled operational decision systems preserve practical room for questioning algorithmic representations, reallocating decision authority, and revising collaboration routines before execution makes intervention costly or infeasible. The framework identifies three driving mechanisms: cognitive asymmetry, dynamic delegation, and meta-learning. These mechanisms progressively constrain what can be represented, who can intervene, and how collaboration routines can be revised. It also specifies two amplifying conditions: cognitive coupling and accountability-defensibility. These conditions raise the operational and justificatory costs of deviating from algorithmic baselines. The erosion of operational contestability unfolds through a recurrent process of Encoding, Habituation, and Closure (EHC): representations and authority rules are built into system design, algorithmic outputs become planning baselines, and those baselines become operational commitments under time pressure, interdependence, and accountability exposure. Generative AI intensifies this process by turning recommendations, explanations, and justifications into portable decision artifacts, thereby accelerating habituation and making closure less visible. The framework reframes human–AI collaboration as an operations management problem: governing decision systems so that representation, authority, and learning remain contestable across repeated decision cycles.
Achieving energy-optimal urban mobility requires precise predictive energy management for electric buses (EBs). However, existing trip-level predictive models largely overlook the topological structure and complex driving dynamics of stop-route networks. This study proposes an edge-enhanced Graph Transformer framework that models each trip as a directed graph, with stops as nodes and inter-stop micro-trips as edges, to predict trip-level energy consumption rates (ECR). Using real-world data from 219 EBs across 47 routes in Guangzhou, the model integrates stop, route, kinetic, and environmental features. A novel edge enhancement mechanism is introduced to strengthen micro-trip dynamics within the attention computation. Feature contributions are quantitatively analyzed using SHAP values, and internal decision-making patterns are interpreted through attention weight visualization. The proposed model achieves MAE of 0.0756 kWh/km, RMSE of 0.0972 kWh/km, MAPE of 10.14
Industrial classification tasks often face challenges such as class imbalance, noise and non-stationary data distributions. Most feature engineering methods aided by evolutionary algorithms and large language models (LLMs) of-ten rely on the predictive performance of downstream classification metrics, while neglecting the feature distribution structure and the relationship between features and labels under distribution shifts. To address these issues, we propose the Feature Meta-Model Agent (FMM-Agent), a framework that evolves features within a meta-model space defined by statistical information, rather than operating directly on raw data. FMM-Agent enables LLMs to perform operator restructure and refine the chain-of-thought to obtain better feature shaping. We further introduce a unified scoring mechanism to jointly evaluate label relevance and distribution stability, allowing the feature pool to gradually move toward better feature distribution shapes. Experiments conducted on 11 data sets show that FMM-Agent consistently improves the recognition ability of minority classes in terms of balanced accuracy, g-mean, and recall, and its performance is superior to other comparative methods. Ablation studies confirm the necessity of evolutionary restructure and generation mechanism strategies. In addition, experimental results with different LLMs show that although stronger models can produce more stable evolutionary process, the overall performance improvement of FMM-Agent does not depend on a specific model. It is worth noting that although FMM-Agent incurs additional inference time costs due to evolutionary feature generation, it achieves a good balance be-tween computational overhead and performance improvement.
This study investigates a metro-integrated freight delivery optimization problem under time-of-use electricity pricing, where an entire planning horizon is divided into multiple rolling decision periods. Across these rolling decision periods, shipment arrivals may exhibit different distributions. In each decision period, trains operating along a metro corridor coordinate with external transport modes to deliver shipments. The resulting total energy cost includes both metro train energy cost and external transport energy cost, and is jointly affected by train timetables, speed profiles, shipment access-station choices, and train assignments. For this practical problem, a mixed-integer linear programming model is formulated to minimize the total energy cost by jointly optimizing these interdependent decisions. To improve computational efficiency, we design an online continual reinforcement learning (CRL)-guided algorithm to add cuts to the mixed-integer linear programming model, thereby reducing the solution space. Additionally, we adopt a progress-and-compress-based CRL framework to enable the agent to continually adapt to varying shipment arrival distributions across decision periods. Computational results show that the proposed algorithm obtains high-quality solutions more efficiently than both Gurobi and traditional reinforcement learning. Managerial insights further reveal that (i) minimizing energy cost is not always equivalent to minimizing total energy consumption under time-of-use electricity pricing, especially before the onset of higher electricity prices, and (ii) introducing a small shipment waiting-time penalty can promote just-in-time shipment transfers so as to reduce unnecessary storage pressure at stations.
Geopolitical risks increasingly threaten global supply chain and trade stability. As a strategically vital resource, rare earths hold substantial economic value in bilateral trade and have become a key arena for international competition and negotiation. Amid geopolitical tensions, rare earth trade is especially susceptible to becoming a focal point of conflict. Therefore, this paper examines how geopolitical risk affects the volume of bilateral rare earth trade, with a focus on the underlying mechanisms and its varied impact across different segments of the rare earth industry chain. We collect bilateral trade data for the complete rare earth industry chain, encompassing upstream ores, midstream smelted products, and downstream functional materials, from key trading entities such as China, the United States, Australia, the Netherlands, Germany, Malaysia, Vietnam, Japan, Republic of Korea, and Thailand between 2000 and 2021. Subsequently, a panel regression model is used to evaluate the extent. The findings indicate that geopolitical risk has a negative effect on the scale of bilateral trade in rare earth. Specifically, it inhibits the scale of bilateral trade in rare earth by increasing trade costs and altering bilateral political relations between countries. However, a favorable institutional environment of trading partner countries can mitigate this negative effect. The heterogeneous results show that geopolitical risk has a negative effect on the scale of bilateral trade in upstream and midstream rare earth products. In contrast, it has a promotive effect on the scale of bilateral trade in downstream rare earth products. This is attributed to the greater competitiveness and irreplaceability of downstream products compared to other rare earth products. This research helps countries address the shock of geopolitical risk and establish a more stable rare earth trade system in a complex and constantly changing world.
The development of emerging industries is vital for a city’s sustainable growth and represents new momentum for urban development. Most of present researches concentrate on specific industries without addressing systematic development and targeted promotion. This study uses data mining and fuzzy-set qualitative comparative analysis (fsQCA) to examine multiple industries and identify key resources for a city’s competitive advantage. Comparing 290 cities in China, emerging industries can be divided into natural resource endowment-based industries and non-natural resource endowment-based industries. For natural resource endowment-based industries, economic capacity and natural resources play crucial roles in enabling cities to gain high industrial comparative advantages. For non-natural resource endowment-based industries, the technology innovation entity and government support capability are more valuable in forming high industrial comparative advantage. The findings aim to provide a theoretical basis for fostering emerging industries toward sustainable urban development.
High energy-consuming firms are vital in industrial and energy systems of China. Against the backdrop of energy transition, coal price shocks could be an important external disturbance that may challenge the operational continuity and risk management capacity of these firms. Using listed firm panel data sets covering six energy-intensive industries from 2007 to 2022, this study measures corporate expected default probability (EDP), and employs the SVAR method to decompose the sources of coal price shocks, examining their impact on corporate default risk and the mechanisms. The empirical results demonstrate that: (1) Coal price shocks significantly affect firms’ default risk. Among different shock types, supply-driven shocks show the strongest significance (coefficient = 0.0854, p < 0.01). Moreover, the impact demonstrates asymmetrical patterns between positive and negative shocks. (2) Mechanism analysis reveals that coal price shocks elevate default risk by increasing cost pressure and eroding firms’ profitability and long-term value. (3) The moderating role of financial constraints exhibits strong regional heterogeneity. (4) The effects of environmental regulations are also heterogeneous. Command-and-control regulations tend to exacerbate default risk under price shocks, whereas market-based regulation has not yet exhibited a significant moderating role. Overall, this study demonstrates that coal price shocks propagate through energy-intensive production systems and financial structures, ultimately manifesting as elevated corporate default risk. The findings provide insights into enterprise risk management and system-level coordination between energy markets and financial systems, offering managerial implications for enhancing the resilience of high energy-consuming firms during the energy transition.
The Difference-in-Differences (DID) method relying on observational data has become a well-established tool for causal inference in the evaluation of energy and environmental policies. Despite its popularity and rapid methodological advances, there is still a lack of practical guidelines for the full DID research design, which may in turn lead to limited credibility and misleading policy implications. This study presents a comprehensive, up-to-date, critical, and practical guide to the DID research design, which is intended to help early-career researchers improve the credibility, transparency, and replicability of policy evaluation studies. This guide offers a step-by-step framework covering real-world questions, clean identification strategies, appropriate controls, proper standard errors, transparent data, robust checks, and insightful analyses of heterogeneity and mechanism. By focusing on research logic and design principles rather than complex methodological details, this study helps researchers and policymakers obtain more credible evidence for policy learning and optimization.
Tracking time-sensitive space targets, characterized by high uncertainty and highly dynamic motion, requires multi-satellite coordination while accounting for the risk of target loss. This study proposes a mathematical model and a Time-Sensitive Space Multi-Target Observation Scheduling (TSSMTOS) algorithm that considers both tracking rewards and target loss scenarios. First, we establish a generalized real-time multi-satellite scheduling framework for tracking these unpredictable, rapidly evolving space targets. Subsequently, we introduce a Double Deep Q-network for Variable-Number Targets (DDQN-VNT). This approach enables the real-time allocation of dual satellites to each target, guided by heuristic rules, effectively addressing dynamic observation requirements. Experimental results demonstrate that DDQN-VNT outperforms traditional rule-based algorithms, the Parallel Dual Adaptive Genetic Algorithm (PDA-GA), and Adaptive Large Neighborhood Search (ALNS) in complex scenarios, exhibiting superior performance, enhanced efficiency, and robust generalization capabilities.
A popular strategy for network vulnerability assessment is to identify critical nodes, whose deletion maximally degrades the connectivity of the original network. Decision makers usually have information about the network structure they intend to obtain, but do not know the number of nodes to target. In this context, we study a distance-based critical node problem with unknown budget, known as the β-distance-based vertex disruptor problem. To solve it, we first derive three integer linear programming formulations, i.e., recursive, triangular connectivity-based, and reduced path-based ones. They are then solved using the CPLEX solver. To approximately solve large instances that an exact solver fails to solve, we propose a simple and effective integrated strategic oscillation search that combines hill climbing and strategic oscillation. It examines a large neighborhood by alternating between a tabu-enhanced destruction procedure and a random order-based improving construction procedure. Extensive experiments on both real-world and synthetic benchmark instances demonstrate the advantages of the proposed formulations and heuristic. In particular, the reduced path-based formulation outperforms both recursive and triangular connectivity-based ones. The proposed heuristic is significantly faster than exact and state-of-the-art algorithms. Finally, we perform a case study on air transportation networks to gain insights into the impacts of COVID-19.
