The Northeast Sea-Land Corridor involves a large number of transportation nodes. Various risk events will not only directly affect the normal operation of these nodes, but also trigger cascading failures in the network through the transfer of freights flow between nodes, which undermine the efficient operation of the corridor. Therefore, the invulnerability of the Northeast Sea Land Corridor transportation network was studied based on a complex network theory. First, a multi-layer composite network was constructed based on the diverse transportation modes of the Northeast Sea-Land Corridor transportation network, and its topological structure was analyzed. Second, metrics, such as degree centrality, closeness centrality, betweenness centrality, eigenvector centrality, and clustering coefficient were selected to establish a node importance evaluation system. The improved correlation CRITIC-TOPSIS, incorporating information entropy, was used to quantify the node importance. Finally, a load-capacity cascading failure model was constructed based on node importance. The impact of different parameters and attack modes on network invulnerability was simulated and analyzed. The findings indicate that when the top 10 nodes by node importance fail, the global efficiency of the Northeast Sea-Land Corridor transportation network decreases by 58%, and the network connectivity drops by 30%. When the node load parameter α=2, node capacity coefficient β=1.4, and γ=1.6, the network exhibits strong resilience against various attack modes.

Maritime cargo transportation through strait waterways faces elevated accident risks, with incidents exhibiting distinct spatiotemporal clustering according to the statistics from the International Maritime Organization. This study addresses spatiotemporal risk heterogeneity by developing an integrated assessment and optimization framework. Based on the data from the Automatic Identification System (AIS), this paper quantifies the dynamic risk profiles across strait waterways at various temporal scales with grid-based analysis and matter-element extension theory. A mixed-integer programming model was formulated with the aim of maximizing fleet profitability subject to aggregate risk constraints. The model simultaneously optimizes vessel speeds, route selection, and precise navigation paths through straits, and uses the Gurobi solver to solve efficiently. Empirical validation using the crude oil transportation data from a major shipping company reveals the key insights as follows: dynamic speed adjustment and adaptive routing effectively mitigate exposure to high-risk zones; fleet revenue increases monotonically with the relaxation of risk threshold; delay penalties exhibit non-linear impacts, with initial implementation (0~5%) reducing revenue by 7.04%, while subsequent increases (5%~30%) yield only 3.54% additional reduction; lower risk thresholds demonstrate reduced sensitivity to demand volatility; and price elasticity shows near-proportional relationships, with 5% price increases generating approximately 5.5% revenue growth. The findings provide some optimization recommendations for shipping companies based on their risk preferences and practical demands for navigating through straits.

To enhance the modal share and profitability of railway in the non-bulk cargo market, this paper comprehensively formulates utility functions for door-to-door railway and road transport by considering cargo density and value attributes. Accounting for the bounded rationality of shippers, the perceived utility is modeled by the cumulative prospect theory. On this basis, a bi-level programming model is established, with the upper level maximizing the profit of railway operator and the lower level minimizing the perceived generalized cost of shippers. A genetic algorithm which incorporates the CPT-Logit method is developed to obtain the differentiated pricing and operation planning scheme, and a typical freight corridor is used for case study. The results show that, within a freight rate adjustment range of [-15%, 10%], the differentiated pricing strategy that tailors rates to cargo density and value characteristics increases the total profit of railway operator by about 26% and raises the overall rail market share by about 5.6%, which achieves a coordinated optimization between profitability and market share. Specifically, for high- density, low-value goods, the strategy recommends lowering freight rates and operating large-scale trains to boost shipment volume. For medium-density, high-value goods, it maintains premium freight rates to secure stable revenue. For low-density goods, it suggests marginal freight rate reductions to achieve limited market expansion. These differentiated measures enable fine-grained control across cargo types, thereby it not only enhances the willingness of shippers to choose rail, but also improves the overall performance of enterprise.

To address the bottlenecks faced by UAVs in conducting trunk and last-mile delivery operations across complex terrains, this paper investigates the location-routing problem for a two-echelon logistics network composed of regional hubs and local distribution centers. A critical gap in existing research is the tendency to overlook the functional and cost heterogeneity among UAVs operating at different echelons. To fill this gap, we formulate a mixed-integer programming model with the objective of minimizing the total system cost, which comprises the facility construction, differentiated two-echelon transportation, and time penalty costs. The model is designed to jointly optimize the distribution of two-echelon facilities, UAV delivery routes, and timeliness of customer service. To solve this problem, a hybrid algorithm is designed, in which a Genetic Algorithm is employed for global facility location and customer allocation, while a Variable Neighborhood Tabu Search is utilized for local route optimization. A case study of Yunlong County, Yunnan, demonstrates that, compared to a sequential decision-making approach, the proposed joint optimization method reduces the total cost of system by 88.3%. Furthermore, in contrast to a single-echelon direct delivery network, the on-time delivery rate is enhanced to 91.9%, achieving an effective balance between cost and service quality. This research provides an effective decision-making model and methodology for the planning and operation of distributed UAV logistics networks.

Connected and Automated Vehicles (CAV) offer novel data sources and optimization opportunities for traffic signal control. However, the existing methods are generally limited in two aspects: first, most methods rely on fixed decision intervals, which struggle to adapt to the dynamic variations of traffic flow, leading to insufficient global optimality of control strategies; second, there is a lack of in-depth modeling of the complex interaction characteristics of mixed traffic flow in low-penetration scenarios, which restricts the robustness of practical applications. To address these issues, this paper proposes a dynamic decision interval signal control method based on Proximal Policy Optimization (PPO). The approach first constructs a multi-source traffic state representation that integrates information from both CAV and Regular Vehicle (RV) by employing Convolutional Neural Networks (CNN) and a multi-head attention mechanism. Subsequently, it designs a multi-discrete action space that combines dynamic decision intervals with phase selection to adaptively generate signal control strategies, thereby balancing decision- efficiency and control flexibility. In the design of the reward function, a multi-objective adaptive weighting mechanism for cumulative delay, queue length, and delay standard deviation is introduced to co-optimize traffic efficiency and fairness. The simulation tests based on real-world road networks demonstrate the control effectiveness of the proposed model. The results indicate that under varying traffic demands, the proposed method reduces both the average waiting time and the average queue length by over 8.50% compared to the traditional discrete control methods. Notably, the method maintains stable control performance even when the CAV penetration rate is as low as 20%, validating its effectiveness and strong adaptability in mixed traffic environments.

To address the time-varying multi-path demand on arterial corridors and the inability of conventional (Origin Destination) OD estimation methods to meet the accuracy and real-time requirements of signal coordination, this paper proposes a dynamic arterial OD estimation method based on an equality-constrained Kalman filter. The model first builds a spatio-temporal mapping between turning flows and OD demands, incorporates signal timing based arrival equalities into the measurement equation to improve model accuracy. To ensure estimates satisfy flow-conservation and other physical constraints, an equality constrained Kalman filter is used and a state-correction mechanism is derived under the minimum mean-square-error criterion to solve the model. A case study on the arterial traffic of Shandong Road in Dalian city shows that the mean absolute errors (MAPE) for two key paths are 12.3% and 17.5%. Compared with a traditional turning-flow model, the proposed approach improves estimation accuracy and better captures the temporal variation of OD flows, providing reliable real-time inputs for multi-path signal coordination.

To address safety and efficiency challenges in cooperative control between connected and automated vehicles (CAVs) and human-driven vehicles (HDVs) at expressway merging areas, this paper proposes a Priority SAAM MAPPO algorithm that integrates priority safety supervision and action masking for collaborative control of mixed traffic flow in merging areas based on the Multi Agent Proximal Policy Optimization. This algorithm introduces the filtering rules of static and dynamic double-layer action mask, and establishes a priority index based on task urgency, spatial criticality and time risk, and optimizes the collaborative mechanism of strategy and value network by the PPO pruning and GAE long-range benefit estimation. The simulation results show that Priority SAAM MAPPO has a good learning convergence in basic and complex heterogeneous scenarios, and has a stable collaborative optimization on the networks of strategy and value. In terms of safety performance, the collision risk rate in basic heterogeneous scenarios is less than 4%, which is half that of MAPPO. The collision risk rate in complex heterogeneous scenarios is about 8%, which is better than MAPPO (12%) and QMIX (18%). In terms of efficiency performance, the average reward is higher than that of in the benchmark algorithm. The spatiotemporal density in merging area has changed from disorderly fluctuations to regular distribution, and the orderliness of traffic flow is significantly enhanced, which verifies its effectiveness and robustness in the collaborative control of mixed traffic flow in merging areas. Further analysis indicates that Priority SAAM MAPPOis suitable for merging control of mixed traffic flows with high traffic density and strong heterogeneity of HDV behavior.

To systematically analyze the impact of different levels of ride-sharing participation on the fleet size of heterogeneous shared autonomous vehicles, this paper proposed a scheduling optimization approach for the heterogeneous SAV fleet composed of vehicles with capacities of 2 and 4, based on different levels of ride-sharing participation among passengers. The approach aims to enhance the resource utilization efficiency and passenger satisfaction while ensuring service quality. Then, it achieved a maximum weight matching for up to four passengers by introducing the concept of edge contraction from graph theory, and reconstructing the network structure. Based on this, the Kuhn-Munkres algorithm was applied to solve the minimum fleet size problem, and obtained the optimal scheduling solution. Case studies based on the real data of road networks and taxi order from Chengdu show that the heterogeneous fleet effectively reduces the fleet size, and improves the efficiency of resource utilization and lower the costs of fuel consumption and operational. Sensitivity analysis further indicates that as the proportion of ride-sharing participation increases from 10% to 90%, the total fleet size decreases from 1 174 vehicles to 331 vehicles, and the fleet composition shifts from 0.94∶0.06 to 0.05∶0.95. Moreover, the total travel time, total travel distance, and total fuel consumption of fleet all exhibit a decreasing trend, whereas the average travel time, average travel distance, and average fuel consumption per vehicle show an increasing trend. In addition, ride-sharing effectively reduces travel fares through cost-sharing among passengers, and the average waiting time and delay time remain within acceptable ranges.

To improve the traffic efficiency, unstable operations, and increased energy consumption caused by left-turn demand at intersections with reversible lanes, this study proposes a trajectory control method for left-turn vehicles under mixed traffic with autonomous vehicles (AVs) and connected autonomous vehicles (CAVs). A multi-objective trajectory optimization model is developed to maximize traffic efficiency and minimize energy consumption. By incorporating the differences between AVs and CAVs in perception accuracy and cooperative capability, the model enables dynamic regulation of vehicle acceleration, deceleration, and lane-changing behaviors. A branch-and-bound algorithm is applied to jointly determine lane-changing strategies and vehicle trajectories. Furthermore, three typical traffic scenarios—continuous, concentrated, and balanced—are designed, and the simulation experiments are conducted under various CAV penetration rates. The results show that the proposed trajectory control model effectively improves the spatiotemporal distribution of vehicle platoons, reducing stop frequency and queue oscillations. The optimal effects occur at CAV penetration rates of 20%, 50%, and 100% under the continuous, concentrated, and balanced scenarios, respectively. In the continuous scenario, the proposed method reduces the comprehensive index by about 51.5% and 38.9% compared with the energy-first and efficiency-first strategies. The corresponding improvements are observed in the concentrated scenario, with reductions of 16.2% and 40.8%, respectively. In the balanced scenario, with reductions of 10.7% and 33.2% of the comprehensive index, respectively. Moreover, the trajectory control model demonstrates strong robustness to variations in headway and variable-lane length.

Automatic Vehicle Identification (AVI) trajectory data serves as a crucial data foundation for building intelligent transportation systems. However, in practical applications, it is often hindered by the problem of trajectory omission caused by false detections and missed detections. The existing trajectory reconstruction methods are difficult to capture the global time background and local boundary information, this paper proposes a vehicle trajectory reconstruction method combining timing perception and boundary loss. This method utilizes a Transformer encoder-decoder architecture as its core framework. A time embedding module is introduced to fuse the vehicle trajectory sequence structure and global time background information to form a unified vector representation. Then, the encoder-decoder framework is used to describe the deep dependencies of the trajectory sequence, and the reconstruction results are generated in an autoregressive manner. The bidirectional boundary loss was then jointly optimized with the standard decoding loss to strengthen the model's attention to the strong constraint information of the missing trajectory boundary. Experimental results based on nearly 420 000 real vehicle trajectory data show that the method achieves trajectory reconstruction accuracies of 95.57% , 93.62%, and 86.36% under missing rates of 10%, 30%, and 50%, respectively, with all reconstruction performance indicators outperforming various baseline methods. The results shows that the proposed methods can improve the performance of deep learning models in vehicle trajectory reconstruction tasks and improve the data integrity of intelligent transportation systems.

To improve the safety of intelligent vehicles traveling through highway work zones, this study proposes a dynamic path planning algorithm considering the perception characteristics of intelligent driving. First, based on potential field theory, the potential fields for road boundaries, lane lines, vehicles, and the highway work zone are constructed for the road environment of the highway work zone. Then, based on the established multi-source potential fields, a fused potential field for the highway work zone is constructed through normalization and weight assignment. A dynamic low-potential channel path planning algorithm (D LPC) is proposed based on time sliding. The simulation experiments evaluate key metrics such as path trajectory, speed profile, minimum work zone distance, acceleration profile, Jerk index, path curvature, and traffic efficiency. The results were compared with the typical path planning models, and the performance of the D-LPC algorithm was verified in terms of safety, comfort, and traffic efficiency. The results show that the maximum and minimum vehicle speeds in the work zone are 118.80 km·h^-1 and 60.12 km·h^-1, respectively, with a minimum distance from the work zone of 2.85 meters. The total travel time is 272 seconds, and the average speed is 61.81 km·h^-1, ensuring safety while maintaining high traffic efficiency. The study results provide theoretical and technical support for the real-time safe dynamic path planning strategy of intelligent driving vehicles in work area scenarios, and is helpful to improve the adaptability and autonomous planning capabilities of autonomous driving vehicles in unstructured environments.

To enhance the emergency collision avoidance performance of intelligent vehicles under low-visibility conditions, this study proposes a trajectory planning and adaptive optimization model based on traffic flow data collected through unmanned aerial vehicle (UAV). First, a steering decision mechanism combining continuous scoring and supervised learning is developed. The results indicate that under low-visibility conditions, the model achieves a directional decision consistency of 93%, representing a 39.6% improvement over the traditional threshold-based method. The average response time is 0.12 seconds, and the accuracy of collision-avoidance trigger detection reaches 96%. Furthermore, 1 000 trajectory generation simulations were conducted using the quintic polynomial and segmented fourth-order Bézier methods. The comparative results indicate that the average computation times of the two methods are 0.046 1 seconds and 0.038 5 seconds, respectively, demonstrating a 16.45% increase in computational efficiency. The maximum curvature variation of the Bézier method is 0.003 m-², which is significantly lower than that of the quintic polynomial method (0.094 m-²), confirming its superior trajectory smoothness and real-time performance. In addition, a dual-factor fine-tuning mechanism based on visibility and adhesion coefficient is proposed to achieve dynamic correction. When the visibility ranges from 440 to 490 meters and the adhesion coefficient from 0.54 to 0.60, the mid-section control points show the most significant adjustments, with a 3.8% increase in lateral offset. When h_vis ≥470 m and μ≥0.57, the curvature variation stabilizes within 0.027m-¹⋅s-¹,and the system response time remains below 0.12 seconds. The results showed the effectiveness and robustness of the dual-factor adaptive optimization model under low-visibility and low-adhesion conditions, providing an engineering oriented framework for trajectory control and safe collision avoidance of intelligent vehicles in complex weather conditions.

The Connected and Automated Vehicles (CAVs) can acquire real-time information on the driving states of surrounding vehicles, enabling them to make more efficient use of roadway infrastructure while ensuring priority passage for buses. This study analyzes the car-following patterns in mixed traffic flow and calibrates model parameters using the Waymo dataset. A CAV lane- changing model incorporating risk perception is then proposed. Building upon this model, a mixed traffic flow simulation framework for urban arterial roads is developed to examine the operational efficiency and risk characteristics of mixed traffic under dynamic bus lane scenarios. The adaptability of lane management strategies is evaluated in consideration of CAV penetration rates, general traffic demand, and bus stop locations. The results show that under the risk-aware intelligent lane scenario, bus travel time increases by 4.8%, while the average travel time per person decreases by 23.1%, and the number of CAV lane changes is reduced by 50.8%, compared to the non-risk-aware scenario. The intelligent lane management strategy achieves optimal performance when the CAV penetration rate is within [0.5, 0.8) and general traffic demand is within [600, 900] passenger car unit per hour with bus stops located between [50, 250] meters, or when the CAV penetration rate is within [0.8, 1.0] and traffic demand is within (900, 1 400] passenger car unit per hour with bus stops between [50, 450] meters.

Formulating crew routing plans for electric multiple units (EMU) constitutes a critical operational procedure in high speed railway management. Its execution quality directly impacts both crew operation costs and crew member productivity. To address the planning challenge of single-depot fixed-route EMU crew routing, this study proposes an optimization strategy featuring "shift consolidation with rapid transfer, deadhead priority, and balanced overnight stays". An optimization model and algorithm with multi-objective two-stage is developed to simultaneously reduce operational costs, enhance crew productivity, and accommodate crew work preferences as much as possible. Stage 1 implements the "shift consolidation with rapid transfer" strategy. Abi-level optimization model is constructed with the upper level maximizing crew duty section connections and the lower level minimizing total connection time. An enhanced Ant Colony Algorithm is designed, which incorporates a pheromone increment allocation strategy based on the Pareto front and a hybrid elite strategy. This stage yields an initial set of crew duty section connections with the minimum required crew size. Stage 2 executes the "deadhead priority and balanced overnight stays" strategy based on Stage 1 results. A mathematical model is established to minimize deadhead travel and overnight accommodation subsidies at remote locations. A heuristic algorithm is designed to obtain the comprehensively optimized crew routing plan. Using operational data from the Xuzhou-Lanzhou High-speed Railway under the Lanzhou Railway Bureau as a case study, the model and algorithm were tested. Results demonstrate that the proposed method efficiently generates crew routing plans for both paired and unpaired train services. The optimization strategies and planning methodology provide decision support for EMU crew scheduling that balances economic efficiency and personnel satisfaction, and also offers reference value for similar resource scheduling optimization problems.

In the daily operations of onboard mechanics, there are numerous predictable task changes, such as additional tasks, training, and leave requests, which lead to frequent adjustments to the crew rostering plan. However, the adjustment capability of traditional method is limited, which means it difficult to meet the requirements of plan stability and workload balance. Therefore, this paper focuses on the work characteristics of onboard mechanics, and then constructs an optimization model. The model prioritizes the minimum changes to the original task arrangement during plan adjustments, with workload balance as a secondary objective. The constraints, including the upper limit of working hours, scheduling of big breaks, and the continuity of multi-day tasks, are incorporated to ensure the feasibility and rationality of plan. In addition, the model is extended to address the scheduling of out-of-town mechanics, taking into account the impact of travel time on task assignments. Through the linearization technique, the model is transformed into a linear 0-1 planning problem. A case study is conducted based on the real data from the Qingdao North Depot, and solved by using the commercial solver GUROBI. The results show that when 36 predictable task changes occur, the model triggers only 32 task adjustments, whereas the traditional method requires 352 adjustments, meaning that the number of adjustments under the model accounts for only 9.09% of that required by the traditional method. At the same time, the model limits the maximum deviation of the workload of onboard mechanics to a single duty task, at 1 645, which is only 38.7% of that under the traditional method.

To expand the coverage and improve the efficiency of high-speed railway express, this paper proposes the express service optimization plan for the high-speed rail network under pickup-delivery mode. An extended space-time network with explicit physical meanings is constructed to precisely capture the service operation requirements and processes of express shipments within the high-speed rail network. Some practical constraints including express delivery deadlines, loading time windows, and train service capacities are converted into restrictions on arcs and nodes in the extended space-time network. A multi- commodity flow model for express services is subsequently developed, aiming to minimize the total transportation time of express shipments. To overcome the symmetry issue of dual solutions in traditional decomposition algorithms, a dual decomposition framework based on the Alternating Direction Method of Multipliers (ADMM) is designed. This approach decomposes the express service optimization problem into individual shortest path problems for each shipment within the extended space-time network. The proposed model and algorithm are validated within an operational network comprising seven high-speed rail lines along with their train services. The results validate the correctness of the proposed model and the effectiveness of the algorithm. Case study demonstrates that the network-based mode reduces the accumulation of express parcels by approximately 45.7% compared to the single-line mode. The transportation efficiency of cross-line express shipments improve notably, with a 92.5% reduction in stranded shipments, significantly enhancing both delivery timeliness and service coverage. The findings provide crucial theoretical support for the networked operational practices of high-speed rail express services.

To overcome the simplistic division of presale periods in current high-speed rail dynamic pricing research, this study proposes a joint optimization of differentiated dynamic pricing and seat allocation for high-speed trains based on the distinct ticket purchase period distributions of different trains. Considering the differences in the distribution of ticket purchase period for each train during the booking horizon, the trains with similar distribution characteristics are grouped into the same category. Differentiated pre-sale period division schemes of dynamic pricing are designed for different types of trains, while the same presale period division scheme is adopted for trains within the same category. A passenger demand elastic demand function is established for each day during the booking horizon. Considering the constraints, such as the train seat capacity, demand, and ticket price bounds, a nonlinear mixed-integer optimization model is developed to maximize the total railway ticket revenue for the joint optimization problem of differentiated dynamic pricing and seat allocation for high-speed trains. The model is then linearized through linear relaxation and outer approximation techniques, and solved with Gurobi solver. Numerical cases based on the Guangzhou-Shenzhen high-speed rail is used to validate the proposed optimization method. The results show that, compared with the traditional fixed pricing strategy without pre-sale period division or dynamic pricing, the optimized total railway ticket revenue and passenger turnover increased by 13.49% and 4.79%, respectively. This study provides a theoretical support and practical guidance for railway operators in formulating dynamic pricing strategies, contributing to the achievement of revenue maximization goals.

To fully utilize the capacity of high-speed railway during off-peak periods, high-speed railway express delivery (HSReD) emerged as a new trend in the development of railway express delivery. However, pricing and freight flow allocation problems are intertwined to determine the operational efficiency and profitability of the HSReD system. To address the challenges posed by the elastic relationship between freight price and demand, as well as stochastic demand, this paper studies the integrated optimization of pricing and freight flow allocation for HSReD by using passenger trains. A two-stage stochastic nonconvex nonlinear programming model is constructed to maximize the expected net profit of the HSReD system, considering the elastic and stochastic demand, pricing constraints, train capacity and station loading and unloading capacities. This model is reformulated into a convex quadratic constrained programming model using outer piecewise approximate linearization and bilinear linearization techniques. A benders decomposition algorithm combined with primal search strategy is developed for efficient solution. Numerical results show that: 1) Compared with the deterministic model, the proposed model achieves higher profits and revenues while reducing transportation costs, and has lower standard deviations for all indicators, demonstrating stronger robustness. 2) The comparative experiments with the solver show that the proposed algorithm has a superior performance in terms of solution efficiency and quality: in the five groups of medium and small-scale cases, the relative gap values between the objective value obtained by the proposed algorithm and the solver are all within 1×10-4 ; in the large-scale case, the proposed algorithm obtained the result within 683.6 seconds, while the solver failed to complete the solution within the specified time. 3) In the application verification on the Zhengzhou-Xi'an high-speed railway line, the proposed method achieves an expected net profit of 14.653 5 million yuan. This verified that the proposed method can attract more express delivery demands and reasonably allocate the expresses to the trains with limited transportation capacity. It achieves efficient matching between demand and capacity resources, thereby significantly increasing the operational profit of the system.

Fuel-dominated tourist vehicles operate at high frequency and for long durations with substantial energy use, fundamentally differing in energy consumption mechanisms from urban buses and freight vehicles. To accurately estimate the energy consumption and optimize eco-driving strategies for tourist passenger vehicles, this paper proposes a multi-objective eco- driving optimization framework integrating grey relational analysis-extreme gradient boosting(GRA-XGBoost) modeling with reinforcement learning. First, an energy consumption feature database is constructed based on real-world operation data, and key influencing factors are identified using Grey Relational Analysis (GRA) to develop a high-accuracy energy consumption estimation model. Then, a Proximal Policy Optimization (PPO)-based eco-driving strategy is designed with explicitly defined state, action, and reward spaces to balance safety, economy, efficiency, and comfort. A collision-avoidance module is incorporated to enhance safety constraints during decision-making. Finally, simulation experiments on the SUMO platform demonstrate that the proposed model achieves a root mean square error of 0.006 1 and a mean absolute percentage error of 3.1%. Compared with the baseline Krauss car-following model and the LC2013 lane-changing model, the proposed strategy reduces energy consumption by 16.88% in low-traffic scenarios and 8.86% in high-traffic scenarios, while maintaining superior driving stability and safety. The study provides useful insights and technical support for eco-driving control and energy optimization of fuel-powered tourist vehicles.

To enhance the appeal of demand-responsive customized bus systems, this paper addresses a two-stage demand- responsive customized bus route optimization problem that incorporates passenger preferences. In the static stage, passenger preferences for vehicle service types are identified with a spatio-temporal clustering algorithm. These preferences are then characterized as exogenous parameters for each stop and integrated into a demand-responsive customized bus route optimization model. The objective of this model is to minimize the sum of the vehicle fixed cost, vehicle variable cost, operating time cost, and penalty costs for unserved passengers. It simultaneously optimizes vehicle routes, arrival times at stops, and passenger loads after serving each stop. An Adaptive Large Neighborhood Search (ALNS) algorithm is designed to generate initial solutions for this static stage. In the dynamic stage, the problem is approached from the perspective of en-route passengers who makes autonomous decisions about accommodating new ride requests. A group decision-making function is introduced to determine whether a vehicle should accept a dynamic request. A dynamic request assignment algorithm is then designed to update the initial routes generated in the static stage, subject to passenger preferences, time window constraints, and vehicle capacity constraints. Finally, a case study using public transit data from Beijing is presented to validate the proposed method. The results indicate that, compared to the single-stage static model, the two-stage demand-responsive customized bus network design model increases the number of boarding passengers by 2.78%; compared to a standard spatio-tamporal clustering algorithm, the strategy that incorporates passenger preference clustering increases total vehicle operating costs by 17.23% while successfully satisfying passenger preferences in the static stage. Furthermore, compared to a scenario without group decision-making, the group decision-making framework leads to a 33.33% reduction in detour distance for the demand-responsive customized bus. This method can provide a decision support for the route planning of demand-responsive customized bus systems.

This paper addresses the efficiency loss in customized passenger transport caused by the separate optimization of passenger stops and vehicle routes and proposes a joint optimization method for dynamic ride and drop-off site and vehicle routing. A multi-objective optimization model is developed to minimize passenger travel costs while maximizing operational benefits, under the constraint that passenger demand points must be connected either via feeder services or direct vehicle routes. The model incorporates multi-vehicle type capacity limits and simultaneously optimizes passenger ride and drop-off schedules and vehicle service plans. A hybrid metaheuristic algorithm is designed using the NSGA-III (Non-dominated Sorting Genetic Algorithm III) framework, integrating DE and ALNS to achieve coordinated global exploration and local refinement. The ST-DBSCAN (Spatial Temporal Density-Based Spatial Clustering of Applications with Noise) is used to identify dynamic stops from demand points and guide population evolution. The case study of the Chongqing-Tongliang corridor shows that the model adaptively selects optimal vehicle fleets across demand levels while boosting both service efficiency and operational profit, with the algorithm exhibiting superior convergence and solution diversity. Practically, it could reduce the average passenger transfer distance to roughly one tenth that of non-joint optimization, and increase the single-trip profit by 31.62 yuan and the vehicle load rate by 2.77 percentage points.

The rapid growth of instant consumption has made food delivery an essential component of urban daily life. However, delivery operations that rely primarily on electric two-wheelers are highly vulnerable to rainfall, leading to substantial declines in delivery speed, increased delays, and reduced fulfillment reliability. Existing research offers limited insights into the resilience of food delivery systems under weather disruptions. This study develops a resilience assessment framework for rainfall-affected delivery operations based on link reliability and system-level recovery probability. A delivery-link resilience metric is created to measure the performance degradation and subsequent recovery at the task level, while a system resilience model incorporating link failure probability and recovery duration is established to quantify resilience under varying rainfall intensities. Using 6.243 7 million valid delivery trajectories and order records collected in Guangzhou, the empirical analysis reveals that average delivery speed under rainfall decreases to 10.8 km·h^-1, representing a 22.3 percent reduction compared with clear weather. Average delivery time increases to 13.2 minutes, an increase of 16.3 percent. Delivery-link resilience decreases by 16 percent in light rain, 18 percent in moderate rain, and 22 percent in heavy rain, with the reduction during the evening peak approximately 12.4 percent lower than in other periods. System-level resilience declines by about 1.4 percent for every 10 mm increase in rainfall intensity. Residential areas exhibit the highest sensitivity to rainfall, with resilience reductions exceeding those of other urban functional areas by more than 50 percent.

To address the challenges of trajectory jumps, matching ambiguities, and inaccurate travel time estimation caused by sparse and low-quality trajectory data, this paper proposes a generative path inference framework based on structure-behavior joint modeling. The framework achieves end-to-end inference through three core modules. Firstly, a variational autoencoder is employed to learn the latent style variables from raw trajectories, capturing individual preferences in route choice and travel speed. Second, a dual-head Transformer decoder is designed to simultaneously generate the complete path structures and segment-level travel times under the collaborative guidance of style variables and contextual attention mechanisms. Finally, a fixed-point theory is introduced to construct a path-time closed-loop mapping, ensuring physical consistency and structural stability of the outputs through residual constraints. The experimental results on the Porto and Chengdu datasets demonstrate that under sparse sampling intervals of 30 to 180 seconds, the proposed method significantly outperforms baseline models, such as Hidden Markov Model (HMM), Deep Map Match (DeepMM), and Learning based Map Match(L2MM) in path matching accuracy, achieving an average absolute improvement of over 10 percentage under extreme sparsity while maintaining inference latency below 0.11 seconds. Ablation studies reveal the indispensability of each component: style modeling is the central to behavioral consistency, fixed-point optimization ensures topological rationality, and the multi-attention mechanism critically determines the accuracy of travel time prediction.

To address the prevalent issue of sample imbalance in road traffic accident data, this study proposes an analytical framework that integrates hybrid sampling with interpretable machine learning to accurately identify the key causal factors of accident severity. To tackle the poor predictive performance of traditional models on the minority class (severe accidents) under imbalanced data conditions, this study first employed a hybrid method combining ADASYN (Adaptive Synthetic Sampling) over- sampling and Tomek Links under-sampling to balance the US Kaggle traffic accident dataset. Subsequently, four machine learning models—Logistic Regression, K-Nearest Neighbors, Decision Tree, and Random Forest—were constructed. The predictive capabilities of these models for different severity levels were evaluated using a confusion matrix. Finally, the SHAP (SHapley Additive exPlanations) algorithm was applied to analyze the key influencing factors of the best-performing model. The results show that: the hybrid sampling strategy significantly improved model performance. The Random Forest model performed optimally, with its F1-score reaching 0.798, an increase of 25.7% compared to the model trained on the imbalanced data. The feature importance analysis of the Random Forest model revealed that the primary influencing factors are day/night condition, temperature, humidity, visibility, and wind speed. Furthermore, it was found that conditions of low visibility and high humidity are prone to lead to more severe accidents. The conclusions indicate that the proposed hybrid sampling method effectively enhances the identification accuracy of model for severe accidents. The SHAP analysis further reveals that the combinations of environmental factors, such as nighttime, low visibility, and high humidity, constitute the key risk scenarios for inducing severe accidents, providing a scientific support for precise traffic safety warnings and interventions.

This paper proposes an improved grey weighted clustering evaluation method based on traffic conflicts to assess traffic safety in weaving zones under human-vehicle mixed driving environments. The Unmanned Aerial Vehicle (UAV) aerial footage was used to collect vehicle data under congested and free-flowing conditions. Key features like speed, acceleration, lane changes, and rear-end collisions were extracted from the field datasets. A simulation scenario was built on the Simulation of Urban MObility (SUMO) platform, using real-world data to calibrate the behavior of both human-driven and autonomous vehicles. Seven traffic conflict factors, including traffic volume, autonomous vehicle penetration, heavy vehicle ratio, weaving ratio, flow ratio, weaving length, and road grade, were selected for an orthogonal simulation experiment. A negative binomial regression model was used to calculate traffic conflicts. Using the traffic conflict occurrence rate as the evaluation index, an improved safety evaluation method was developed with combination weighting and a sine curve probability function. A simulation comparison was conducted using a typical weaving zone in Chongqing. The results showed that the improved method had a 98% accuracy in determining safety levels, outperforming the linear grey variable-weight clustering (78%) and linear grey weighted clustering (94%). The improved method demonstrated higher accuracy and better applicability. Overall, the proposed method supports policy-making, such as autonomous vehicle admission.

The stacking problem of import containers is highly complex due to conflicts between unloading and retrieval sequences and yard resource constraints. This study focuses on automated vertical yards and proposes a two-stage stacking decision method based on deep reinforcement learning. The process is modeled as a Markov decision process with a phased structure of "block- selection-slot selection" to reduce the dimensionality of state and action spaces. A differentiated reward function is designed: block level decisions promote balanced yard utilization, while slot-level decisions minimize relocations and retrieval distances. In algorithm design, the Deep Q-Networks (DQN) is used for block selection and Dueling DQN for slot selection. Simulation results show that the proposed method produces balanced strategies across the yard and adapts well to different yard densities and container batch scenarios. The average relocation rate is controlled at 15%~27%, and the maximum retrieval distance is 3.84 bays per container, representing reductions of about 61.5% and 38.7% compared with historical yard data. Compared to the single- stage DQN, two-stage Proximal Policy Optimization (PPO), and heuristic optimization, the proposed method achieves faster convergence, fewer relocations, and shorter retrieval paths. These results confirm the effectiveness of phased modeling and differentiated rewards in complex stacking problems and provide a practical solution for intelligent scheduling and resource optimization in large-scale automated yards.

The coordinated low-carbon scheduling of energy and logistics in container ports is a crucial measure for promoting sustainable port development. To address the volatility of wind and solar power generation, random scenarios for wind and solar output are generated using Latin hypercube sampling and k-means clustering methods. A coordinated energy-logistics scheduling model is developed considering vessel arrival schedules, the operational status of port equipment, the balance of energy supply and demand, and the introduction of hydrogen-powered container trucks. The objective of the model is minimizing the port's energy and carbon emission costs, incorporating Conditional Value at Risk (CVaR). The simulation results indicate that the proposed scheduling strategy not only avoids impacting vessel berthing and departure times but also significantly reduces system operating costs by 29.65% and carbon emission costs by 12%. The introduction of hydrogen-powered trucks enhances scheduling flexibility and lowers operational costs. However, when the ratio of electric to hydrogen-powered trucks reaches 1∶2, further increases in hydrogen trucks no longer yield cost benefits, at which point overall operational efficiency is maximized. The CVaR-based coordinated scheduling method can effectively control the risk of high operational costs in extreme scenarios, providing a scientific basis for ports to achieve reliable, economical, and low-carbon operations.

Supercharging infrastructure, capable of meeting the rapid energy replenishment demands of electric trucks equipped with large-capacity battery packs while offering high technological maturity and facility universality, represents a critical pathway for advancing the electrification transformation in port collection and distribution operations. However, large-scale deployment of supercharging infrastructure involves not only internal efficiency optimization within the transportation system but also poses significant impacts on the power distribution network through high-power charging loads. This paper proposes a collaborative planning methodology for electric truck supercharging infrastructure that considers transportation- power network interactions. First, considering the heterogeneous energy consumption characteristics of multi-type container transportation tasks, this study develops a spatial charging demand distribution model to accurately characterize fleet charging scheduling decisions under different transportation modes, including import/export and laden/empty container movements. Then, this study establishes a bi-level optimization framework for charging station capacity configuration, where the upper level coordinates the interests of multiple stakeholders including transportation enterprises, charging operators, and power utilities, while the lower level captures the transportation enterprises charging choices and the power system operations, achieving two- network interaction through locational marginal pricing (LMP). The case study results demonstrate that the proposed method reduces total social costs by 7.66% while significantly decreasing average LMP by 54.53%, effectively mitigating the impact of concentrated charging loads on the distribution network. The optimized charging infrastructure exhibits a multi-center networked layout, with railway hubs and inland distribution areas emerging as charging centers, achieving a balance between charging convenience and grid economics. The collaborative planning demonstrates significant effectiveness at medium demand levels and accurately identifies the diminishing marginal returns point of investment.

Traffic flow prediction is a core capability of intelligent transportation systems, which supports the real-time traffic dispatching and road network optimization for traffic management authorities. However, the traffic flow of highway exhibits rapid variations and dynamic spatial correlations, which makes static topologies inadequate for accurately characterizing inter-segment dependencies. Meanwhile, the errors of prediction tend to accumulate when the forecasting horizon increases, which further increase the difficulty of modeling. To address these challenges, this paper proposes a Two-Level Adaptive Spatio-Temporal Network (TLASTN). The model first extracts temporal features and captures multi-scale sequential patterns through multi-scale convolutions and a bidirectional GRU. Subsequently, a two-level spatial modeling is conducted on frame-wise dynamic graphs. At the first level of this model, the similarity of segment feature is taken as the primary criterion. This model is constrained by the distance of road network topological to generate a sparse dynamic adjacency matrix and mask, which is used to screen physically reasonable candidate neighbors. At the second level, a graph attention mechanism is applied under the mask constraint to assign dynamic weights to the candidate neighbors, which enables a fine-grained modeling of the spatial dependencies under different traffic states. The prediction framework adopts a hierarchical structure with a shared encoder and independent convolutional prediction heads, which allows simultaneous multi-horizon forecasting at 5, 15, 30, and 60 minutes. Experiments on the ETC gantry data from the G15 Shenhai Expressway and the G25 Changshen Expressway in Fujian Province demonstrate that TLASTN achieves the best performance across all forecasting horizons, among which, on the G15 dataset, TLASTN reduces MAPE by 3.0%~5.9% compared with baseline models. The results indicate that adopting a two-level spatial modeling on frame-wise dynamic graphs can effectively improve the accuracy of traffic flow prediction in dynamic scenarios, which provides a feasible technical solution for expressway operation management and decision-making.

To clarify the eye movement characteristics of drivers when passing through the accesses of urban underground roads, the Jiefangbei Underground Ring Road in Yuzhong District, Chongqing was selected as the research site. Tobii Glasses 2 wearable eye tracker was used to collect the eye movement data from 40 drivers when passing through 8 types of accesses. The indicators, such as the pupil area drivers', blink frequency, and saccade amplitude, were statistically analyzed, and further classified according to the function and orientation of accesses and DEA (Data Envelopment Analysis) evaluation. It identifies the key indicators which influence the layout of underground road access on driving behavior. The research results show that: the average pupil diameter of drivers at curved accesses reaches 4.401 mm, which is significantly higher than that at straight segments (4.183 mm) and adjacent accesses (4.123 mm); the visual efficiency of the connecting road accesses is the best while their comprehensive efficiency is strongly effective by DEA, and the average horizontal saccade speed is 261.698 pixel · s-¹, higher than that of other types of accesses; The average pupil diameter at left accesses of straight segments (4.213 mm) is larger than that at right accesses (4.078 mm), and the comprehensive efficiency is less than 1; the comprehensive efficiency of combined accesses is strongly effective by DEA, with a blink frequency of 0.607 blinks·s-¹, which is higher than that of Access S (0.587 blinks·s-¹) and Access C (0.591 blinks · s-¹), while the pupil diameter is 4.123 mm, which is smaller than that of Access C (4.401 mm) and Access S (4.183 mm). These results reveal the mechanism by which the layout of underground road accesses affects driving behavior, and can provide more comprehensive theoretical support for optimizing the layout of urban underground road accesses and improving driving safety.

To scientifically assess the vulnerability of urban rail transit networks and ensure the safe, stable and efficient operation of urban transportation systems, this paper proposes a network vulnerability assessment framework that integrates the structure of traffic communities and the actual distribution of passenger flow. It makes up for the deficiency of existing methods that fail to consider the characteristics of community structure. Firstly, an improved Louvain algorithm considering enhanced transfer station influence (Transit-Enhanced Louvain, TEL) is constructed to partition the rail transit network. By introducing a transfer edge weight adjustment function, the compactness of community partitioning is dynamically optimized. Secondly, based on the community partitioning results, node community indicators centered on inter-community importance, intra-community importance, and station community importance are designed. These indicators are integrated with node passenger flow intensity to construct the Cumulative Comprehensive Importance (CCI), thereby achieving accurate identification of key nodes. Finally, empirical verification is conducted using the real dataset from the urban rail transit network in Beijing. The performance variation trend of urban rail transit network under deliberate attacks is evaluated from three aspects: network efficiency, relative connected subgraph, and passenger flow detour ratio. The results show that when the influence enhancement coefficient of transfer station is 1.4, the maximum modularity obtained by the TEL algorithm reaches 0.822 3, which is better than other baseline models. For the deliberate attack on station sequences based on the CCI indicator, the failure of the top 10% stations leads to a 78.1% decrease in the overall network efficiency, an 83.3% decrease in the relative connected subgraph, and an 86% increase in the passenger flow detour ratio. The network failure efficiency is significantly higher than that of traditional methods, verifying the effectiveness of proposed framework in this study. This study thus provides a scientifically grounded decision-making basis for identifying pivotal nodes and enhancing the resilience of urban rail transit systems.

To evaluate the actual dispatch of bicycle-sharing and attain refined operations, this paper introduces the Bicycle-sharing Inner Circulation Area (BICA) as the basic spatial unit, and proposes a dispatch evaluation and classified management framework with an empirical study in Shanghai. First, the riding and dispatch behaviors are identified using bicycle-sharing lock data, and the BICAs are delineated by combining community detection algorithms and spatial proximity. Then, a multidimensional evaluation index system is constructed, including the Inter-zone Dispatch Ratio (IDR), Batch Dispatch Ratio (BDR), Long-distance Dispatch Ratio (LDR), Average Dispatch Distance (ADD), and Dispatch Intensity (DI), to quantify the operational efficiency and spatial structure of dispatch behaviors. On this basis, the k-means clustering algorithm is used to classify the BICAs. The results show that the study area can be divided into over 200 BICAs, whose dispatch patterns exhibit significant spatial heterogeneity. For instance, the IDR is higher in the city center and areas surrounding the outer ring road, while the BDR and DI are higher in the southwestern suburbs. Furthermore, the clustering identifies four types of areas: Mature Urban, Scattered Long-distance Dispatch, Suburban Expansion, and Suburban Low-efficiency. Differentiated management strategies based on their respective dispatch characteristics are proposed, such as stable maintenance and efficiency improvement, optimized delivery strategies, intensive mode adjustment, and precise reduction of operations.

To investigate the heterogeneity characteristics of human-caused traffic accidents on national and provincial arterial roads, this paper proposes a chain causal inference method with the synchronous analysis to analyze the causal mechanism of traffic accidents and quantify the interaction effects. First, the recent traffic accident information was collected from 11 arterial roads in Xinjiang province, and the severity of the accidents was classified into three categories using the K-prototype clustering algorithm. Based on the structural causal model, causal forest, and the SHAP (SHapley Additive exPlanation) algorithm, the study proposes a chain causal inference model to infer the key causal chains of the accidents, the interaction effects of multi-dimensional causes, and the predicted values of the accident categories. The chain transmission characteristics and heterogeneity of "scene combination-human factors-accident category" were also analyzed. Then, the contribution of multi-dimensional factors was evaluated based on the SHAP value. The behavioral causes were classified according to the strength of the causal effect to identify the key human causes of the accidents and propose targeted accident prevention and control strategies. The results show that:(1) the weighted average F1 score and macro average AUC (Area Under Curve) value of the proposed method are 0.86 and 0.82, which are relatively higher than those of commonly used machine learning algorithms. The proposed method overcomes the limitations of traditional association models in characterizing and quantifying the interaction mechanism of multiple factors, making it suitable for the analysis of the heterogeneous mechanism of human-caused accidents. (2) From the analysis of the causal chain, it can be seen that human factors are the main causes of accidents, and environmental factors such as weather and time periods have significant impacts on the consequences of accidents. The combined effect of adverse environments and dangerous behaviors has a nonlinear impact on the escalation of accident severity. (3) Speeding, fatigued driving, following too closely, and poor observation are the critical behavioral causes. For commercial vehicles, these driving behaviors normally cause higher severity of the accidents(i.e., the proportion of severe accidents exceeding 50%), which indicates an important area for accident monitoring and prevention.