To address the limitations of prediction accuracy caused by high nonlinearity, non-stationarity, and complex driver coupling with information redundancy in transportation carbon emission series, a hybrid prediction model which combines a two-stage feature selection (TLMIC-LASSO), Successive Variational Mode Decomposition (SVMD), Improved Animated Oat Optimization algorithm (IAOO), and Bidirectional Long Short-Term Memory network (BiLSTM) is proposed. The two-stage feature selection strategy based on Time-Lagged Maximal Information Coefficient (TLMIC) and LASSO regression is constructed to identify the time-lag effects of key driver accurately and eliminate redundant variables. The SVMD is introduced to decompose the original series of carbon emission adaptively into multiple stationary modal components, thereby reducing the non-stationarity of data. The animated oat optimization algorithm(AOO) algorithm is improved via hybrid chaotic perturbation, adaptive tdistribution mutation, and dynamic opposition-based learning strategies, and the IAOO algorithm is utilized to optimize the key hyperparameters of BiLSTM network adaptively to prevent the model from falling into the local optima. Finally, an IAOO-BiLSTM prediction model is built for each modal component, and the prediction results are integrated. The model is validated with the data of transportation carbon emission from 1990 to 2023 in China. Results indicate that compared with the optimal contrast model, the proposed model reduces the RMSE, MAE, and MAPE by 21.88% , 23.33% , and 24.32% respectively, which significantly improves the prediction accuracy of transportation carbon emission. 

The existing studies on the choice behavior of the commuter of customized bus ignore the heterogeneity of car dependence, and lack the strategies of customized bus demand response for commuter groups with different car dependence. Based on the survey data, commuters are classified into three groups—low, medium, and high car dependence. A conditional Logit model is used to systematically examine the effects of socio-demographic and travel attributes on the willingness to choose the customized buses. It is also used to explore the customized bus choice behavior mechanism of commuters considering the car dependence heterogeneity. The results indicate that the user profiles are different among the three groups. The positive impact of education level on the choice of customized bus increases progressively, while the income effects exhibit notable heterogeneity across groups. Flexible working hours significantly influence the willingness to choose customized buses for all groups, and the effect direction reverses depending on car dependence levels. Regarding travel attributes, peak hours have the strongest positive impact on the willingness to choose, especially for the medium dependence group. The key concerns of different groups also vary: low, medium, and high dependence groups prioritize congestion time, commuting costs, and walking time, respectively. The low-dependence group is willing to pay an additional 1.3 yuan per minute to save on congestion time, while the high-dependence group is willing to pay an additional 0.8 yuan per minute to reduce walking time. This paper proposes a stratified, differentiated and precise customized bus promotion and pricing strategy tailored to commuters with varying car dependence levels, aiming to enhance the appeal of customized buses, and reduce the excessive reliance on private cars, alleviate commuting pressure, and promote the sustainable development of urban transportation systems.

Against the background of the four-network integration in rail transit, the connection relationships among multi-system rail transit lines at comprehensive passenger hubs become highly complex. Local-line and cross-line passenger flows are intertwined, which makes the coupling among train timetabling, stop planning, and passenger routing more prominent. This study considers the origin-destination (OD) demand of both local-line and cross-line passengers across multiple rail transit systems. It constructs a multilayer train-passenger space-time network and introduces arc-based decision variables to represent train and passenger route choices.On this basis, a model of multi-commodity network flow is established with the objective of minimizing the total passenger travel time, which is used to address the collaborative optimization of multi-modal rail transit timetabling, stop planning, and passenger routing. Given the computational complexity of the model, this study develops a Lagrangian relaxation algorithm and designs a dual-solution-based heuristic to generate feasible solutions to the original problem. Finally, the effectiveness of the proposed model and algorithm is verified through a real-world case study on a network composed of the Beijing West-Shijiazhuang section of the Beijing-Guangzhou High-Speed Railway, the Beijing-Xiong'an Intercity Railway, and Beijing Sub-Center Line. The case includes 49 trains and 345 passenger groups.The results show that 76% of passengers have a transfer time within 30 minutes, and the average transfer waiting time of passengers is 11 minutes. Sensitivity analyses of relevant parameters are also conducted.

To enhance the resilience of multi-modal transportation networks in urban agglomerations, this study investigates resilience recovery strategies for multi-modal transportation networks with the goal of maximizing resilience. This paper develops an undirected weighted multi-modal transportation network for urban agglomerations. A resilience assessment method is proposed for multi-modal transportation networks with consideration of performance resilience and connectivity resilience. The performance resilience consists of structural performance and functional performance. A bi-level programming model for resilience restoration is established with the objective of maximizing resilience, in which the upper-level model is a bi-objective mixed-integer programming model, and the lower-level model is a user equilibrium assignment model. The Reinforced Learning Artificial Hummingbird Algorithm (RLAHA) and the Frank-Wolfe algorithm are designed to solve the upper-level model and the lower-level model, respectively. The results show that the RLAHA algorithm exhibits superior convergence performance and solution quality. Under the optimal recovery strategy obtained by the resilience restoration decision model, resilience is improved by 15% and 20% respectively compared with the degree-priority recovery strategy and the random recovery strategy. The optimal repair scheduling scheme derived from this model enhances resilience from the commencement of repair operations. The allocation of maintenance teams has a considerable impact on the resilience and overall repair duration of multi-modal transportation networks in urban agglomerations in western China. The maintenance duration of a single transportation network decreases with an increase in maintenance teams, yet the overall repair duration does not decline monotonically with the growing number of maintenance teams. The structural performance weight has a minor effect on performance resilience and connectivity resilience, the average rates of change both are 1.28% . The decision maker's preference exerts a significant influence on performance resilience, but a slight impact on connectivity resilience, their average rates of change are 12.37% and 3.39% , respectively. The research findings can provide a decision-making basis for resilience assessment and emergency repair recovery of multi-modal transportation networks in urban agglomerations across China.

The low-altitude economy in China is rapidly developing but challenges still remain, such as fragmented airspace resources, low utilization efficiency, and high safety risks. To address these issues, this study proposes a method for constructing and restoring urban low-altitude logistics air-route networks, enabling rapid and efficient recovery under disruptive events. An incremental optimization strategy is used to design the spatial layout of nodes and develop a hierarchical topology consisting of trunk, secondary, and branch routes. A multi-objective genetic algorithm is introduced to dynamically prioritize node and edge restoration, yielding an efficient recovery strategy. Using logistics station coordinates and airport no-fly zone data from the cities of Tianjin and Jinan, the study develops a representative urban air- route network with road- freight- based demand and operational constraints. The experimental results indicate that the proposed restoration strategy markedly improves network efficiency and flow resilience, while Pareto frontier analysis confirms the advantage of the multi-objective optimization approach. The framework provides a practical foundation for enhancing Unmanned Aerial Vehicle logistics operations, urban air mobility, and emergency response under the emerging low-altitude economy. 

Under the framework of international transportation, cross-border multimodal transport is restricted by customs clearance time and the volatility of tariff levy rates. This paper employs triangular fuzzy numbers to characterize the uncertainties inherent in customs clearance time and tariff rates. A path optimization model for multimodal transport is constructed with the objectives of minimizing transportation cost, transport time, and carbon emissions. Based on the fuzzy chance-constrained programming theory, the uncertain model is transformed into a tractable mixed-integer programming model. An improved multi-objective evolutionary algorithm is designed to solve the model, which mitigates the loss of high-quality solutions and enhances the robustness of the algorithm. Finally, a case study is conducted on cross-border transportation from Chongqing to Singapore, covering 15 nodes across 8 countries. The computational results demonstrate that the proposed method can reasonably generate a set of optimized multimodal transport schemes addressing uncertain cross-border conditions. Compared with other state-of-the-art multi-objective evolutionary algorithms, the hypervolume indicator is improved by 13.21% and the stability indicator by 69.63%, which indicates a significant performance enhancement of the proposed method. Therefore, it can effectively provide high-quality path decision support for relevant carriers engaged in cross-border multimodal transport.

In practical transportation, the transshipment time associated with different transport mode conversions at each node exhibits significant uncertainty, which increases the complexity of multimodal transport path optimization and poses challenges to carriers' decision-making. This study investigates the multi-objective path optimization for multimodal transport considering uncertain transshipment times, and develops a multi-objective optimization model aiming to minimize transportation cost, transportation time, and carbon emissions. To characterize the uncertainty of transshipment time, sample data are first tested for normality, and uncertain transshipment times are simulated using a combination of Latin hypercube sampling and the inverse cumulative distribution function. Furthermore, to address the issues of premature convergence and loss of population diversity in the traditional NSGA-II algorithm when solving high-dimensional, the multi-objective, and multi-constraint problems, adaptive crossover and mutation operators are introduced to dynamically adjust the crossover and mutation probabilities based on the fitness status of the population. The search flexibility is improved at different evolutionary stages and the quality and convergence performance of the Pareto solution set is enhanced. A case study is conducted based on the multimodal transport network for exports from Heilongjiang Province to Russia. The results show that the compromise solution yields a transportation cost of 52 138.7 yuan, a transportation time of 26.3 hour, and carbon emissions of 4 964.8 kg. Compared with the standard NSGA- II algorithm, the improved algorithm reduces transportation cost and transportation time by 1 690.5 yuan and 13.4 hour, respectively.Moreover, compared with the deterministic transshipment time condition, the model under uncertainty tends to select solutions with fewer transshipments and shorter transportation times, albeit at the expense of increased transportation cost and carbon emissions by 8 705.5 yuan and 1 675.7 kg, respectively. In addition, Jiamusi, Fujin, and Tongjiang are identified as high-frequency transshipment nodes, and rail-to-water transshipment is a key operation. Prioritizing the optimization of these nodes and their transshipment operations can effectively improve the overall efficiency of multimodal transport. The findings provide references for path decision-making in multimodal transport.

In the routine patrol operations of urban traffic police, ground patrol vehicles are constrained by the road network topology and traffic congestion, which limits their mobility and makes it difficult to simultaneously achieve full-area patrol visibility and rapid response at congested locations. This paper proposes a bi-objective mixed-integer programming model for air-ground collaborative traffic police patrol routing, with the joint objective to maximize the risk capture of congestion and the scale of covered patrol nodes. The model characterizes the heterogeneous motion constraints of ground patrol vehicles and police UAVs, as well as incorporates a collaborative task logic of "airborne detection- ground response- task reassignment". To address the computational challenges over large-scale road networks, an improved adaptive large neighborhood search algorithm is developed: an improved Shaw removal operator incorporating dual weights of spatial distance and congestion risk difference is proposed to enhance identification of risk-clustered zones; a unified greedy repair operator is designed to enable dynamic air-ground resource allocation; and a local incremental time evaluation formula is derived to reduce computational overhead. Experiments are conducted on a real road network in the cross-river district of Hangzhou. Results show that the proposed algorithm outperforms the genetic algorithm, NSGA- II, and tabu search baselines in terms of captured risk score, number of covered nodes, and composite score. Compared with the tabu search, the proposed algorithm improves the composite score by 19.3% and the number of covered nodes by 24.7%. The ablation study shows that, after incorporating the improved Shaw operator, the proposed algorithm improves the composite score by 4.5%, the captured risk score by 12.5%, and reduces the running time by 21.6% compared with ALNS-G; Pareto frontier analysis shows that the proposed algorithm clearly dominates the tabu search in the bi-objective space of risk capture and node coverage, which provides decision-makers with patrol scheme options under different operational preferences.

To address the challenges of coupled failures and collaborative emergency recovery in urban transportation and public New Energy Vehicles (NEVs) charging networks under extreme rainstorm-induced waterlogging, this study proposes a service-release-oriented optimization strategy for road network repair sequencing. First, the cross-layer mapping mechanism from "road network disruption" to "topological islands" is characterized within disaster scenarios. A temporal optimization model is developed to maximize the System Service Efficacy Release Rate (SSR) and the corresponding Area Under the Curve (AUC). Then, the Gravity-Routed Adaptive Frontier Topology algorithm is introduced to mitigate the limitations of heuristic searches, specifically their susceptibility to disconnected redundant repairs and sparse rewards in fragmented networks. This algorithm strictly circumvents off-network invalid repairs through a dynamic frontier constraint and incorporates an isolated island gravity field fallback mechanism. This mechanism effectively guides repair resources through zero-return windows, prioritizing the reconnection of high-capacity charging hubs. Simulation results from a large-scale coupled traffic-power network in Haidian District, Beijing, demonstrate that the algorithm achieves a full-cycle recovery performance (AUC is 57.58) that consistently outperforms traditional strategies based on demand- priority (54.43), topology- priority (52.83), and reachability- priority (52.77). Ablation experiments confirm that the dynamic frontier constraint eliminates off- network invalid repairs, while the gravity field mechanism advances the reconnection of high-value hub stations to the 36th hour. Sensitivity analysis reveals a non-linear relationship and diminishing marginal returns between repair investment scale and system recovery efficacy under resource constraints. These findings quantitatively demonstrate the critical role of underlying traffic network connectivity in supporting upper- layer charging services, providing a theoretical foundation for urban flood control and collaborative emergency resource allocation.

As the critical nodes in urban transportation networks, the accident rates of intersections remain persistently high. The development of assisted driving technologies puts forward some new challenges to the safety of intersection. To improve the prediction of collision risk at intersections, this study proposes a collision prediction method that integrates the driving behavior entropy with a bidirectional Long Short-Term Memory (Bi-LSTM) network. Based on a simulation platform of high-fidelity driving, three typical conflict scenarios at urban intersections were constructed: the conflicts of right-turn pedestrians and left-turn oncoming vehicles, and straight-driving obstacle avoidance. The data of longitudinal speed and lateral steering wheel angle control behavior were collected from 55 drivers. A sliding time window, speed entropy, steering wheel angle entropy, and their corresponding standard deviations were calculated to construct the feature indicators of multidimensional driving stability. Repeated-measures analysis of variance (ANOVA) was employed to statistically analyze the feature differences across different scenarios and driving phases. A Bi-LSTM model was further developed to predict the collision risk in real time within a 3-second time window. The results indicate that both standard deviation-based indicators and driving behavior entropy measures during conflict phases were significantly higher than those during the stable driving phases, and entropy indicators effectively captured the uncertainty and complexity of driver operations. The incorporation of entropy features reduced the miss rate by 30.07% and the false alarm rate by 42.45%, while it improves the average early warning time by 36.84%. As an effective complement to traditional fluctuation- based indicators, the entropy of driving behavior can enhance the sensitivity and prediction stability of model under hazardous conditions. It provides a feasible reference method for intersection risk warning in the assisted driving environments and contributes to the improvement of road traffic safety.

This study addresses start-stop delay, queue spillback, and reduced traffic efficiency caused by platoon arrivals at multiple intersections along an arterial under mixed traffic flow with connected and automated vehicles (CAVs). A bi-level cooperative control method integrating signal timing and variable speed limit (VSL) control is proposed. At the upper level, a rolling optimization strategy is adopted. Lane-level queue states and short-term arrival predictions are used to generate dynamic phases, green durations, and lane-level green service windows, while real-time corrections are made through early green onset and delayed green termination. At the lower level, the lane-level green service windows are used as constraints. The speed limit for each controlled CAV is calculated according to its distance to the stop line and the target green window, so that vehicles are guided to arrive at the stop line during the green interval as much as possible. Speed boundaries, additional-delay constraints, speed-limit holding and smoothing, and car-following risk checks are further introduced to reduce unnecessary speed-limit interventions and speed fluctuations. Simulation results show that, under moderate traffic demand, the proposed cooperative control reduces average delay by approximately 51.2% to 65.8% compared with the fixed- time signal scheme, and the delay reduction becomes more significant as the CAV penetration rate increases. Under high demand with a high CAV penetration rate, the average number of stops per vehicle is reduced by about 57%, indicating that the proposed method can effectively suppress stop-and-go oscillations and approach queue accumulation. In addition, under moderate demand, average fuel consumption per vehicle is reduced by approximately 40.9% to 48.6% compared with the fixed-time scheme. The results indicate that the proposed cooperative control method can improve traffic efficiency and vehicle running smoothness while reducing energy consumption under different traffic demand levels.

Traditional capacity estimation methods based on Fundamental Diagram (FD) theory are inadequate to capture the dynamic characteristics of adaptive signal control and the inherent instability of traffic flow states. To accurately estimate the capacity of adaptive signal control systems, this paper proposes a dynamic capacity estimation method based on Gaussian Process Regression (GPR). To address the complex traffic data generated by adaptive control, this study develops a heterogeneous multi-dimensional feature space, which encompasses the traffic state x(t) , signal control strategy Φ(t) ,and decision time series H(t) . A bidirectional Long Short-Term Memory (LSTM) neural network is then introduced to encode the temporal data and extract relevant features, and a multi-head attention mechanism is applied to identify critical decision points within the adaptive control strategy. A Gaussian Process (GP) model is then used to learn the functional mapping from multi- dimensional input features to traffic volume. By leveraging the nonparametric Bayesian properties of GPs and their inherent capacity for uncertainty quantification, the proposed model flexibly accommodates the complex nonlinear relationships between signal control strategies and traffic capacity, thereby overcoming the static modeling limitations of conventional FD theory under fixed signal timing plans. Simulation results demonstrate that the proposed model outperforms not only the conventional FD-based static capacity model and a static GP model, but also four additional baselines—Support Vector Machine (SVM), Random Forest (RF), K-Nearest Neighbors (KNN), LSTM, and Transformer—achieving optimal performance across the evaluation metrics. The proposed GPR method with extended input features effectively reflects the impact of adaptive control decision changes on traffic capacity, with improved capacity prediction accuracy ( EMAPE is 3.96% , R2 is 0.951 0). The findings demonstrate the practical value of the proposed method for the performance evaluation and systematic optimization of adaptive signal control systems.

To address the challenge of balancing longitudinal-lateral cooperative control accuracy and real-time performance for connected heterogeneous vehicle platoons in dynamically complex scenarios, this paper proposes a distributed longitudinal-lateral coupled control method enhanced by an artificial potential field. First, based on a directed graph communication topology, a three-degree-of-freedom longitudinal-lateral coupled dynamic platoon model is developed considering tire nonlinear characteristics. Then, a distributed nonlinear model predictive controller enhanced by an artificial potential field is designed, and the alternating direction method of multipliers is used for distributed solution. The simulation experiments are conducted under various operating conditions. The results show that in a vehicle cut-in scenario, compared with the traditional distributed nonlinear model predictive control, the proposed method can reduce the lateral position error and heading angle error by approximately 36.3% and 22.1%, respectively, while maintaining the desired safe inter-vehicle distance. In a high-curvature S-shaped curve scenario, compared with the distributed linear model predictive control, the lateral position accuracy and heading angle accuracy are improved by about 57.1% and 64.2%, respectively. Compared with the centralized nonlinear model predictive control, the computational efficiency is increased by approximately 87%, which verifies the rationality of the distributed nonlinear architecture and achieves an optimal trade-off between platoon tracking accuracy and real-time performance. Furthermore, under the distributed nonlinear model predictive control enhanced by the artificial potential field, driving comfort is improved by about 8.9%, and the fuel consumption index is reduced by approximately 3.6%, indicating that the proposed method also exhibits advantages in terms of comfort and fuel economy.

Autonomous vehicle lane-changing decisions often focus on the ego vehicle and adjacent vehicles, overlooking overall traffic flow. This paper proposes a lane-changing trajectory optimization model involving traffic flow conditions. The risk potential field theory is used in the modeling and an adaptive obstacle potential field is constructed by embedding obstacle vehicle acceleration into a two-dimensional Gaussian distribution. A road potential field is coupled to analyze risk distribution in dynamic traffic. Lane-changing trajectories are generated with a quintic polynomial. A multi-objective cost function, which integrates safety, efficiency, comfort, and traffic flow disturbance, identifies the optimal lane-changing time and trajectory. The model is validated on a CarSim/MATLAB co-simulation platform at 60 km·h^-¹ and 120 km·h^-¹. Simulations show smooth ego vehicle tracking with peak longitudinal accelerations well below the 0.3g ride-comfort threshold. The high-speed scenario reduces the peak velocity deviation of the ego vehicle by 34.9% compared to the low-speed scenario. Transient disturbances on the target lane rear vehicles dissipate quickly, demonstrating good disturbance control of the model.

At signalized urban roads, smooth trajectories planned for Connected and Automated Vehicles (CAVs) often involve relatively low local speeds, which can respond to overtaking by following human-driven vehicles (HVs). To enhance the operational efficiency and energy economy of mixed traffic flow in overtaking scenarios, this paper proposes a CAV trajectory planning method that responds to overtaking by rear HVs is proposed. First, HVs are categorized into general HVs and following HVs based on their willingness to follow CAVs and mixed platoons led by CAVs. A simulation framework is designed for mixed traffic on signalized road segments, which uses finite state machines to model vehicular interactions. A mixed-integer quadratic programming model is developed to plan CAV trajectories, aiming to minimize travel time and fuel consumption while maximizing driving comfort. A trajectory prediction and overtaking detection algorithm is developed and an improved passive yielding strategy (IPYS) is proposed to update CAV trajectories and alleviate the negative effects of HV overtaking. Simulation results demonstrate that the IPYS consistently outperforms benchmark strategies across various CAV penetration rates and traffic demand levels. The differences among strategies are most pronounced when CAV penetration ranges between 20% and 30%. An increase in the proportion of following HVs contributes to improved traffic efficiency. Sensitivity analysis further indicates that road segment length and signal cycle length significantly influence optimization outcomes. When the road segment length is between 500 and 700 meters, the frequency of CAVs being overtaken varies considerably with changes in cycle length.

To improve the computational time in optimal control for a large-scale mixed road network which consists of a multi-bottleneck expressway and joint urban networks, this paper develops a coordinated optimization method that balances control performance and computational cost. The proposed method extends the existing traffic flow model in the following aspects: (a) the Macroscopic Fundamental Diagram (MFD) model is jointly introduced for both the expressway and urban components of the mixed network; (b) the traffic flow model of multi-region urban networks accounts for complex scenarios where on-ramps have insufficient storage capacity and must relyon upstream urban intersections for control, while incorporating the factor of cordon queues into each network reservoir MFD. The extended traffic model serves as the prediction model within the Model Predictive Control (MPC) framework, with the optimization objective of minimizing the Total Time Spent (TTS) in the system. This enables real-time solution of the total perimeter inflows for the expressway stretch and each urban network. Based on local state feedback, each total perimeter inflow is allocated to individual controlled perimeter links. Simulation results indicate that, by incorporating cordon queues, the proposed method effectively reduces the mismatch between the prediction model and the simulation model, thereby achieving control performance superior to existing related methods and approaching that of the theoretically optimal control method. Leveraging MFD-based modeling for both components of the mixed network, the proposed method significantly shortens computational time and meets real-time control requirements. Specifically, when queue storage space is sufficient, the proposed optimization-based control method reduces the TTS by 8.6% compared to the no-control scenario; under limited queue storage space, the TTS reduction is 7.3%.

Job-housing spatial separation, as a typical feature of modern metropolises, has led directly to the widespread prevalence of long-distance commuting. To overcome this ground transportation bottleneck and promote job-housing balance, the electric vertical take-off and landing (eVTOL) aircraft provides an innovative solution and a new transport mode. With advantages such as high speed, strong flexibility, and independence from ground road networks, eVTOLs are expected to become a key vehicle for future urban medium-distance commuting. However, current research and practices in urban air mobility (UAM) mostly focus on unmanned aerial vehicles, with studies on manned operations still in their early stages. Therefore, this study proposes a systematic method for site selection and route planning based on eVTOL operations. First, by extracting valid commuting trip samples and incorporating no- fly zone constraints, this study applies an improved greedy coverage clustering algorithm based on K-Dimensional Tree (KD-Tree) to achieve eVTOL site selection, aiming at "maximum coverage with a minimum number of sites". A route planning model was then constructed with the objective of minimizing operational costs and the constraints of passenger capacity, flight range, fixed costs, and distance-based costs. A two-stage hybrid strategy combining a greedy algorithm for initial solution generation and an improved genetic algorithm for iterative optimization is adopted for solving the model. The proposed method is validated through a concrete case study, resulting in a feasible route plan. A comparative evaluation is conducted from the perspectives of travel distance and commuting time. The results indicate that, compared with private cars, the eVTOL save on average 28.29% in travel distance and 61.58% in travel time, demonstrating its advantages in improving commuting efficiency and spatial utilization.

To address the non-convex optimization challenge of synergizing energy efficiency and safety path planning for heavy-lift Unmanned Aerial Vehicle (UAV) in low-altitude wind fields with strong spatial non-uniformity, this paper proposes a collaborative path planning framework (PI-GNN-A*) using an A* search guided by a Physics-Informed Graph Neural Network integrating flow field dynamic features. First, based on fluid dynamics mechanisms, a sparse physical graph incorporating vorticity features is constructed to empower the model with the capability of perceiving flow field anisotropy. Then, an enhanced asymmetric loss function is designed. By penalizing underestimation errors, this function internalizes safety constraints into a conservative bias within the heuristic potential field, effectively balancing search efficiency with risk avoidance. Idealized prototype validation and statistical tests on two-dimensional steady-state Rankine vortex fields demonstrate that the proposed method achieves a highly robust synergistic trade-off among computational efficiency, flight safety, and aerodynamic energy efficiency. Compared to conventional neural network heuristic algorithms lacking explicit physical constraints, the proposed method, after embedding vorticity features, effectively suppresses blind spatial exploration and achieves a statistically significant improvement in search efficiency. Compared to the Dijkstra algorithm, the proposed method obtains a 50.8% reduction in search space as a reasonable trade-off for an approximate 8.0% degradation in comprehensive cost. Compared to traditional wind-field-aware methods based on artificial potential fields, the proposed method, at the reasonable expense of a moderate search overhead, broadens the wind-resistant safety margin of the system from 5.08% to 6.17% (a relative increase of 21.46%) while ensuring noninferior global comprehensive cost, providing faster path replanning capabilities for airborne platforms with limited computing power.

Focusing on trips within 0~3 kilometers, this study examines residents' mode choice behavior among four green transportation options: electric bicycles, shared bicycles, buses, and the metro. This study incorporates granular environmental and service attributes alongside classical dimensions to develop an Integrated Choice and Latent Variable model (ICLV), thereby deepening the characterization of psychological and micro-environmental factors in travel decisions. Through systematic comparison with the traditional Multinomial Logit model, the ICLV model demonstrates advantages in both goodness- of- fit and behavioral interpretability. The results indicate that psychological latent variables, socio-economic attributes, and mode-specific characteristics all significantly influence choice behavior, with the psychological pathways enhancing the model's explanatory depth. Among the findings, an increase in metro headway exerts a pronounced positive effect on electric bicycle choice (with the probability of choosing an electric bicycle rising by approximately 7.27 percentage points when the headway increases from 3 to 9 minutes), while the effects of shared-bike cost and bus in-vehicle travel time are relatively more gradual. Furthermore, the presence of physically segregated bicycle lanes significantly increases the attractiveness of shared bicycles, whereas air pollution encourages a shift toward bus and metro use. These findings reveal the complex competitive and substitutive relationships among short-distance green travel modes. The conclusions provide a quantitative basis for the design of short-trip environments, the coordinated development of multimodal green transport systems, and the formulation of targeted policies.

To address the difficulty in identifying the causes of taxi transfer congestion at transportation hubs during holidays, this study develops a state-recursion model of queue dynamics based on the principle of flow conservation. A dual-gap congestion diagnosis method is further proposed through consisting of the potential capacity gap and the effective matching gap. The dual-gap indicators enable a stratified identification of the types of supply-demand imbalance. Together with the vehicle utilization rate, empty-departure rate, passenger abandonment rate, and supply-demand response lag, they form a multi-dimensional cause identification framework. This framework systematically characterizes the dynamic evolution of queue accumulation and the temporal mismatch between supply and demand. A case study is conducted by using 5 min observational data from three typical peak days during the National Day holiday in 2024 at Lanzhouxi Railway Station. The results reveal three distinct queue evolution patterns: evening-concentrated imbalance, continuous accumulation, and fluctuating release. These patterns demonstrate pronounced time-of-day dependency and pattern heterogeneity. The overall capacity of vehicles at the hub is sufficient to cover the passenger demand during holidays. Therefore, the primary cause of congestion is not a vehicle shortage but the combined effect of inadequate service conversion and temporal supply-demand imbalance. Passenger abandonment reduces the observed queue length and average waiting time and weakens their ability to reflect the true extent of the imbalance. The proposed method can serve as a reference for congestion cause identification and operational diagnosis at comparable transportation hubs.

With the large-scale deployment of new energy school buses, the fire hazard triggered by lithium-ion battery thermal runaway has become increasingly prominent. Constraints such as narrow aisles within vehicles significantly limit the evacuation efficiency of young children during emergencies. To address this issue, this paper designed and conducted individual and single-file controllable movement experiment of young children, focusing on the constraint of school bus passage width. Based on image processing methods, the movement trajectory data of the children were extracted, and the influence of passage width and walking posture on movement trajectory, speed, acceleration, spatiotemporal characteristics, and fundamental diagram were quantitatively analyzed. The results show that aisle width has a significant impact on the walking speed in individual and single-file experiment of young children. In the individual movement experiment, when the passage width increased from 0.30 m to 0.50 m, the average speed of forward and sideways walking increased by approximately 42.70% and 60.50%; in the single-file movement experiment, the average speed of forward and sideways walking increased by approximately 88.90% and 18.00% , respectively. This study reveals the coupling relationship between forward distance and acceleration and quantitatively defines three functional zones-deceleration and collision avoidance, steady- state following, and acceleration and traction- using 0.70 m (mean zero point) and 2.05 m (gain inflection point) as boundaries. This study provides empirical data support for the interior spatial design of school buses and the emergency evacuation safety of young children.

In existing theoretical research, the high-speed railway train operational plans generally rely on time-varying passenger demand, facing challenges such as difficulty in obtaining accurate and comprehensive data, and disconnection between theory and reality. By contrast, the data based on the passenger arrival and departure time windows is simpler and more available in optimizing train operational plans. This paper introduces the convenience levels of passenger boarding and alighting at both departure and arrival ends during different time periods throughout the day as the indicator to measure arrival-departure time windows. The travel time of trains, including delays due to conflicts, is considered as the operational cost for enterprises. The passenger travel cost is determined based on the impact of train stops and the convenience level of passenger boarding and alighting, and the objective function is built by integrating these two costs. The constraints such as train candidate set and train coverage during convenient time periods are also included in the model. A genetic algorithm is used and the passenger flow distribution strategies are designed in conjunction with railway ticket selling practice. Multi- neighborhood genetic operators for train operation plans are designed based on train operation periods, departure time and stops. An adaptive mechanism for genetic operations is introduced to enhance the search efficiency of neighborhood solutions. A case example is analyzed for the Beijing-Shanghai High-speed Railway to verify the effectiveness of the model and algorithm. The result shows that the average passenger occupancy rate of the line is 69.4% and the train operation structure, time period distribution and capacity utilization are reasonable. Compared to the actual train operation plan, the total convenience level of passengers in the optimized plan increases by approximately 18.2% and the number of Origin to Destination (OD) pairs with high convenience level significantly increases, resulting in more balanced passenger travel experience and significant improvement in train operation efficiency. The result indicates that train operation plans optimization can be solved effectively by the passenger arrival and departure time windows and the proposed optimization method provides practical references for railway passenger transport departments in making decisions on train operation plans.

Through-running services are a key path toward regional rail integration, but shared-corridor operations with multiple operators require coordination of service quality, capacity, and revenue attribution. This paper compares separate and integrated planning for suburban railways and develops a collaborative optimization model that maximizes host-line net benefit under capacity, rolling-stock, and economic-viability constraints. The model uses excess travel time (ETT) to measure passenger time loss, jointly optimizes train frequency, composition, and passenger assignment, and is reformulated as a mixed-integer linear program solved by Gurobi. In a Shanghai Airport Link case study, through-running increases host-line operating revenue and reduces passenger travel time relative to no through-running. Integrated planning achieves a host-line net benefit 15.2% higher than that under separate planning, mainly through higher operating revenue, whereas separate planning gains mainly from reduced ETT. Since ETT accounts for only about 12.4% of total travel time but represents the main component that can actually be improved, it identifies service improvements and planning-mode differences more sensitively than total travel time. The sensitivity analysis shows that separate planning is beneficial when the through-running frequency limit and co-line revenue requirement are low, while integrated planning should be adopted when both are high. The framework supports quantitative evaluation of through-running operations and planning-mode selection.

To alleviate traffic congestion at high-volume toll plazas under land-use constraints, this study proposes a lane width and function optimization method based on movable toll islands. Considering the minimum safe lane width required by different vehicle types and the spatial requirements for tolling equipment deployment, the configuration of toll islands (including number, type, position, and sequence) and toll lanes (including width, functional allocation, travel direction, and tolling mode) is dynamically adjusted. A mixed-integer linear programming model with the objective of maximizing capacity is established and efficiently solved using the CPLEX solver. The results indicate that under complex traffic scenarios characterized by tidal flows, stochastic fluctuations in demand, and frequent changes in traffic composition, the proposed flexible lane design significantly outperforms the traditional lane function switching scheme, increasing capacity by approximately 20% ~80% . As the width of small-vehicle-exclusive lanes increases from 2.6 to 3.0 meters, the capacity improvement gradually decreases. With higher proportions of small vehicles and electronic toll collection (ETC) traffic, more narrow toll islands and dedicated lanes can be configured, providing greater flexibility for optimization. In addition, the optimization of lane width and functions can accommodate oversized special transport vehicles. The findings provide technical support for refined operational management and smart capacity expansion of high-volume toll plazas.

To address the strong coupling among multiple decision variables and the large solution scale in the joint optimization of dry bulk fleet configuration and operational scheduling, this paper develops a mixed-integer linear programming model based on a space-time network, with the objective of maximizing total system profit. A dynamic vessel addition algorithm is proposed to solve the model, which decomposes the multi-vessel joint optimization problem into a sequence of single-vessel subproblems and employs a marginal profit termination criterion to reduce computational complexity effectively. Randomly generated instances and a large-scale realistic case based on iron ore transportation are used to validate the model and algorithm and to conduct sensitivity analyses. Numerical results show that, for most test instances, the objective value deviation between the dynamic vessel addition algorithm and Gurobi is within 5%, while computation time is substantially reduced, with particularly significant advantages in large-scale cases. Fixed costs and fuel prices exhibit a threshold effect on fleet operational decisions: within a certain range, they compress profit margins without altering fleet configuration, whereas beyond the critical threshold they trigger a reduction in the number of deployed vessels and the abandonment of low-profit cargoes. In addition, system net profit is more sensitive to freight rate changes in high-volume backbone cargo categories, while freight rate fluctuations of a small number of high-value long-haul cargoes have a relatively limited impact on overall profitability.

To investigate the problem of heterogeneous fleet deployment on shipping routes under collaborative agreements between port terminal operators and liner shipping companies, this paper analyzes the relationships among multiple time windows, multiple origin-destination times, and multiple loading/unloading rates within the collaborative agreement, as well as their interactions with vessel speed, ship configuration, and arrival/departure times. A mixed-integer nonlinear programming model is developed to minimize the total transportation cost over the planning horizon. A partial linearization solution method is designed, and the model is solved using a nonlinear solver. A numerical example based on the SEA3 route operated by COSCO Shipping Group is conducted to validate the applicability and effectiveness of the proposed model. The results show that, compared with scenarios featuring fixed port time windows, fixed loading/unloading rates, and homogeneous fleets, the collaborative agreement combined with a heterogeneous fleet reduces the total cost of the liner company by 5.98% , 15.31% , and 4.27% , respectively. Sensitivity analysis indicates that when fuel prices are high, the liner company does not continuously increase the deployment of large vessels but instead turns to small and medium-sized vessels as substitutes. When demand fluctuations are significant, the liner company proactively increases the number of deployed vessels to reduce the loading pressure per vessel and avoid excessively high speeds or prolonged port operation times caused by high-load operations. The findings provide references for liner companies decision-makings on heterogeneous fleet deployment and operational cost control under collaborative agreements.

To improve the accuracy and stability of short-term airport departing passenger flow prediction at an hourly scale, this paper proposes a large language model-based airport departing passenger flow forecasting method, ChatPAX, where Chat denotes a dialogue-/prompt-based large language model modeling approach, and PAX is a commonly used aviation abbreviation for passengers. The proposed method formulates departing passenger flow prediction as a conditional generative sequence forecasting problem constrained by flight schedules, and jointly characterizes demand-side inertia from historical passenger volumes and the constraint effect of future flight supply on passenger flow formation within an end-to-end framework. ChatPAX uses a causal language model as the predictor. Through structured prompts, the historical departing passenger volume sequence over the past 48 h and the forward-looking flight information within the prediction horizon are organized as conditional inputs, and the departing passenger volume sequence for the next 24 h is generated in a fixed format. During training, an answer-masking mechanism is introduced to concentrate supervision signals on the target prediction segment, thereby improving generation stability and convergence efficiency. Meanwhile, parameter-efficient fine-tuning and low-precision quantization are combined to enable a trainable and deployable forecasting workflow under single-GPU memory constraints. Experiments based on multi-month real operational data from a domestic airport show that ChatPAX achieves a mean all- day weighted mean absolute percentage error (WMAPE) of 13.003% on the test set. Compared with the autoregressive integrated moving average model (ARIMA), long short-term memory network (LSTM), Random Forest, and extreme gradient boosting algorithm (XGBoost), the prediction errors are reduced by approximately 53.9%, 29.5%, 26.2%, and 28.4%, respectively, and the proposed method exhibits higher stability in cross-date prediction. Ablation experiments further indicate that introducing flight schedules as a supply-side constraint can bring stable performance improvements, whereas exogenous factors such as holidays and weather do not produce consistent marginal gains.

As the airport mixed operations grow increasingly complex and the energy demands across various operational units become more diverse, the importance of optimizing the scheduling of heterogeneous shuttle fleets becomes increasingly prominent. This study constructs a mixed integer linear programming model that integrates vehicle-specific energy constraints and a temporary stopping strategy for unmanned vehicles, with the core objective of minimizing the total travel distance. To address the optimization problem of peak-hour routing for heterogeneous fleets, a Quantum-behaved Particle Swarm Optimization algorithm was designed for efficient solutions. An empirical analysis based on typical peak-hour operational data from Beijing Daxing International Airport demonstrates that the Quantum-behaved Particle Swarm Optimization algorithm significantly outperforms other heuristic algorithms and the traditional First-Come-First-Served rule in reducing total travel distance, balancing vehicle workloads, and enhancing overall airport surface operational efficiency. Specifically, the algorithm reduces the total travel distance from 625.58 kilometers to 229.27 kilometers, with a reduction rate of approximately 63.3%. While it also decreases the required number of vehicles from 55 to 35, with an optimization rate of 36.4%. These results fully demonstrate the notable advantages of the algorithm in convergence speed and solution accuracy.

Extreme weather events and other emergencies may disrupt airport operations. The conventional recovery strategies typically resume flight schedules through time extension after disruptions, but often overlook passenger cancellations caused by prolonged delays and the flexibility of transport capacity structures. To address this, this study develops a dual-layer optimization model for departure recovery that incorporates both passenger cancellations and flight structure adjustments. The upper layer optimizes capacity allocation through structural strategies such as flight consolidation, aircraft swaps, and standby replacement to minimize operational costs. The lower layer optimizes departure times under constraints like capacity limits, balancing cancellations with total delay duration. Using the Weibull distribution to characterize cancellation probabilities under delays, the model introduces flight duration and high-speed rail accessibility parameters to reflect flight heterogeneity. A co-evolutionary algorithm integrating simulated annealing, constraint methods, and non-dominated sorting is designed to solve the model. The validation was performed using Kunming Changshui International Airport as an example, and the result demonstrates that compared to baseline recovery plans based on time extension and departure scheduling optimized solely for takeoff times, the proposed method reduced total operational costs by 37.27% and 11.51%. Cancellations decreased by 42.08% and 13.50%, and total delay duration is reduced by 36.46% and 11.21%. With consideration of passenger behavior responses and departure decisions, this study provides actionable decision-making for coordinated recovery during large-scale operational disruptions at airports.

Flight Quick Access Recorder (QAR) data, derived from civil aviation flight operations quality assurance, provides a valuable conduit for investigating aircraft landing load patterns on airport runway structures. This study focuses on the vertical interaction behaviors during aircraft landing. Considering the intrinsic data characteristics, the study identified and extracted QAR data from 933 flights—comprising four representative aircraft types: Airbus A320, A330, and Boeing B737, B747—for the two primary stages: landing and post-landing rollout. A QAR data-driven method for estimating vertical dynamic loads during these phases is proposed based on the impulse-momentum theorem. A first-order Fourier series fitting technique is then used to characterize the oscillation frequency of the vertical dynamic loads during the rollout phase. The Gaussian Mixture Models (GMM) are utilized for precise probabilistic distribution fitting and modeling of raw QAR parameters (total aircraft mass, inertial vertical speed, and pitch/roll angles) and their derived metrics (vertical load-to-weight ratio and rollout dynamic load frequency). Based on the quantitative analysis of GMM for the QAR data and its derived data during the landing and post-landing rollout phase of aircrafts, statistical parameters such as the mean and standard deviation of the single Gaussian principal component associated with potential hard-landing conditions of the aircraft can be obtained. Taking the data in this paper as an example, the statistical upper limit of the vertical load-to-weight ratio of four types of aircraft relative to potential hard-landing conditions is calculated to be between 1.158 9 and 1.441 4, which provides support of real engineering data for the design and operation of airport runway structures.

To assess the low-altitude airworthiness of rail-transit station areas, this paper proposes a ground- risk assessment framework for small and large aircraft operations. The framework focuses on a multi-event coupled risk chain of "failure-impact-damage". For small aircraft, probability calculation models are developed to account for human exposure risk, casualty risk, and road-traffic risk in station areas under the influence of population distribution and entrance/exit passenger-flow distribution. For large aircraft, key parameters are further integrated considering the above risk factors. The key parameters include lethal volume, floor area ratio, and building volume. A probability calculation model is established for building-damage risk in station areas under large-aircraft failure. Spatial constraints of station-area no-fly zones are incorporated to modify the risk field. A spatial aggregation algorithm is used to construct a risk spatial-differentiation map for urban areas. Shanghai Rail Transit Line 1 and Line 3 were taken as the case studies. The results suggest that small-aircraft ground-risk probability is mainly affected by population density and the road environment, and high-risk zones are concentrated at station entrances/exits and surrounding primary roads. Large-aircraft risk probability is also significantly affected by building density and building volume, and risk probability in building-dense areas increased significantly. In terms of the risk probability of urban station clusters, 11% of stations are within the no-risk range for small-aircraft operations, whereas 91% of stations are at the high-risk level for large-aircraft operations. The lethal area of the aircraft is the core element determining the ground-risk level, followed by road area for small aircraft and building volume for large aircraft. The findings can provide quantitative risk support for facility layout and route planning in rail-transit station areas under air-rail integration

To investigate the relationship between Contrast Sensitivity (CS) at different spatial frequencies and flight fatigue, 20 participants were recruited to conduct a simulated flight experiment under different light environments, in which CS at four spatial frequencies and flight fatigue indicators were collected. First, based on the results of repeated-measures analysis of variance, the flight fatigue indicators which are significantly affected by light environment changes were identified. Subsequently, the relational models between CS at the four spatial frequencies and flight fatigue were constructed by using the linear mixed-effects model, followed by the goodness-of-fit tests for these models. Furthermore, marginal effect plots were generated to analyze the reliability and influence trends of the fixed-effect variables in the constructed relational models. Finally, the significance tests were conducted on the random effects in the relational models using the methods of random intercept distribution and one-way variance analysis. The results showed that at the 0.1 significance level, all four relational models achieved good fitting performance, with relatively low AIC and BIC values and conditional R2 ranging from 0.332 to 0.483. The CS at low-to-moderate spatial frequencies (1.5, 3.0, 6.0 cpd) was significantly positively correlated with the Critical Flicker Frequency (CFF), among which the correlation between CS at 1.5 cpd and CFF exhibited the optimal performance. In contrast, the CS at the high spatial frequency (13.0 cpd) was significantly negatively correlated with Pupil Diameter (PD). The results of random intercept distribution and one-way ANOVA demonstrated that the random effects corresponding to the individual differences were all statistically significant (p＜0.01). This research focused on revealing the specific correlation indicators between CS at different spatial frequencies and flight fatigue, and the findings can provide the basic theoretical support for monitoring flight fatigue based on CS.

Airport ground handling covers the key operational links from flight landing to takeoff, and there are conditional temporal dependencies between tasks in each link. At the same time, the multi flight operation process is accompanied by resource competition, and this complex coupling relationship has become a major challenge restricting the development of ground support. Traditional methods rely on manual rule design and are difficult to apply in large- scale and high dynamic scenarios. This paper proposes a deep reinforcement learning algorithm framework that integrates mathematical graph embedding and attention mechanism for the airport ground handling scheduling. The dynamic scheduling problem with random task sequences is graphically modeled as a mixed integer programming model through conditional task graphs, and transformed into a heterogeneous graph. The graph convolutional network is used to extract global graph embedding to efficiently represent the coupling relationship in the graph, which is fused with real-time state to form a deep reinforcement learning state representation. The attention mechanism based policy network is used to learn serialized scheduling strategies, achieving end-to-end scheduling strategy optimization. The test results in a real airport simulation environment show that compared with traditional heuristic methods and reinforcement learning methods, this algorithm framework has significant improvements in multiple core metrics. Among them, compared with deep Q-networks, the average flight delay is reduced by 40.1%, and the resource utilization rate is improved by 6.6%, which verifies the effectiveness and superiority of the algorithm.