Transportation infrastructure has a strong driving effect on urban economic development. In particular, the construction of large-scale transportation infrastructure can have a significant impact on land use and spatial form. In response to the need for a comprehensive assessment of the economic benefits of urban transportation infrastructure, this study proposes a framework and method for assessing the Wider Economic Benefits (WEBs) of transportation infrastructure based on the Urban-Integrated Economy, Land Use, and Transport (U-IELUT) modeling. This study extends the traditional "Four-Step" transportation planning model to a "PECAS+(Production, Exchange and Consumption Allocation System)" WEBs assessment model by linking it with an economic and population forecasting, a socio-economic activities allocation, a space development, and a wider economic benefits assessment module. The WEBs assessment model is designed to evaluate both the direct economic benefits and the wider economic benefits, mainly the agglomeration benefits, of urban transportation infrastructure. Taking Wuhan Metro Line 2 as an example, the direct and wider economic benefits were assessed using the "PECAS+" wider economic benefits assessment model. The findings indicate that the direct economic benefits of the Metro Line 2 in 2027 are approximately 1.043 billion yuan. The dynamic agglomeration benefits are about 264 million yuan, which is approximately 25.3% of the direct economic benefits. This demonstrates that the wider economic benefits, especially the agglomeration benefits, should not be overlooked in the economic benefits assessment of transportation infrastructure. Meanwhile, it is also possible to ascertain the impact differences of transportation infrastructure construction on various zones of the study area, that is, the spatial distribution of wider economic benefits, which can provide multidimensional decision-making support for the investment and construction of transportation infrastructure.

As vital components of urban spatial governance systems, rail transit and land use play critical roles in advancing urban green low-carbon development. This study proposes a rail transit and land use supply-demand model using population as the intermediary variable, with two scenarios established: natural development and low-carbon development. Integrating multi-source data including station Point of Interest (POI) data, Land Use and Land Cover (LULC) data, and population thermal dynamics, this study uses a passenger flow potential model and BP neural network to predict supply-demand relationships of passenger flow. Through Voronoi polygon spatial analysis and coupling degree modeling, this study comparatively examines the rail transit and land use coupling characteristics in Shenzhen under both scenarios and identifies mismatch causes. The results demonstrate that: the mean coupling degree between rail transit and land use remains stable within [0.7, 0.8] under both scenarios, exhibiting spatial patterns of "higher values in western regions versus lower values in eastern regions" and "central agglomeration with peripheral dispersion". Coupling mismatch primarily manifests as insufficient rail transit supply capacity relative to population travel demand. Compared with the natural development scenario, 58.63% of Voronoi units achieve superior coupling degrees under the low carbon development scenario, demonstrating its effectiveness in enhancing rail transit passenger flow and reducing transportation carbon emissions. However, decreased population density in central urban areas under the low-carbon scenario generates more mismatched stations (Z<1) characterized by weaker population travel demand relative to rail transit supply capacity. These findings provide strategic references for megacities to coordinate rail transit construction with intensive land use while achieving carbon emission reduction in transportation systems.

Researches on the spatiotemporal impact mechanism of rail transit passenger flow on workdays, weekends, and statutory holidays are much crucial to formulate targeted development strategies and optimize spatiotemporal resource allocation. Previous studies have mainly focused on weekday passenger flow, but have not fully considered the key influencing factors and differences about their effects on passenger flow during different periods. This article implements machine learning regression model through tuning, training, and evaluation screening based on the data from three different periods of passenger flow and the variable of "5Ds+C" (Density, Diversity, Design, Destination Accessibility, Distance and Centrality) influencing factors. The XGBoost-SHAP model is used to analyze the differences from the spatiotemporal impact on passenger flow at three levels: overall feature importance, interaction effects, and local spatiotemporal heterogeneity. The case study of Tianjin Metro shows that the XGBoost has better explanatory power compared to the Random Forest (RF) and Gradient Boosting Decision Tree (GBDT), with a fitting coefficient of over 0.7. There are significant differences in key influencing factors, importance, and mode of influence between workdays, weekends, and holidays in terms of overall feature importance analysis. The importance of diversified travel factors on weekends and holidays reaching 59.8% and 61.3% . According to interaction effect analysis, residential land has significant interaction effects on passenger flow in office land, shopping and leisure land, and tourist attraction land over different periods. Attention should be paid to the residential land at low land use level, the mature residential areas with good facilities construction, and the commercial tourism land with passenger flow agglomeration respectively during weekdays, weekends, and holidays by the analysis of the local spatiotemporal heterogeneity. The stations with complete leisure tourism facilities have passenger flow increased significantly during holidays, with a SHAP impact value increasing by about 5000.

This study addresses the issue of effectiveness evaluation and precise implementation about short-term promotional incentive strategies for Mobility-as-a-Service (MaaS) platforms. This study proposes a "quasi-social experiment-causal inference" integrated framework to quantify the average and heterogeneous treatment effects of short-term promotional incentives on MaaS users' lifecycle, and to achieve individual-level identification of strategy-sensitive user segments. First, a quasi-social experiment combining with propensity score matching is used to eliminate the selection bias across four types of travel characteristics: travel frequency, spending amount, travel regularity, and the proportion of peak-hour travel, and to estimate the average treatment effect at the overall strategy level. Then, a binary causal forest model is introduced to analyze the heterogeneity of user group-level behavioral characteristics, and to calculate the distributions of heterogeneous treatment effects and individual treatment effects. Finally, based on the individual uplift model (Uplift model, UM), a four-quadrant model of natural survival time and uplift values is constructed to achieve dynamic classification and prioritization of sensitive user segments at the individual level. The empirical results show that: (1) The incentive amount is positively correlated with average treatment effects, but with diminishing marginal effects. Although high-value incentives (>10 yuan) yield an average treatment effects of 12.18 days, low-value incentives (1~10 yuan) are more cost-effective (average treatment effects 11.10 days, with a cost reduction of 82.5%). (2) User travel frequency and spending amount are the core e of heterogeneous treatment effects , with higher-frequency and higher-spending users showing more significant responses. (3) The individual-level uplift model effectively identifies marketing-sensitive user segments, with 60.19% of them being low-frequency users. However, among high-spending users, 12.84% are still sensitive, which verifies the bias of the traditional "high-value first" strategy and identifies that 23.61% and 17.37% of low-frequency and low-spending users, respectively, belong to the counterproductive group, requiring avoidance of intervention.

Accurate capture of trip ends information is crucial to guarantee the efficacy of transportation planning strategies. The 4G/5G communication technology enables continuous and dynamic tracking of individuals' entire travel process, offering novel opportunities for enhanced trip ends extraction. However, the inherent uneven spatiotemporal positioning characteristics of mobile signaling will pose considerable challenges to the efficacy of trip ends identification. Therefore, this study introduces an adaptive trip ends identification approach tailored for uneven spatiotemporal location trajectories derived from mobile phone signaling. Firstly, the U-DBSCAN algorithm is designed to accurately identify individual's trip ends across varying data densities. This method incorporates the dual non-uniform constraints of spatiotemporal signaling data, enabling precise discrimination of trip ends, which effectively mitigates the challenges associated with missed or erroneous identifications that frequently arise due to sparse signaling data. Secondly, leveraging the K-average nearest neighbor algorithm, a parameter adaptive framework for U-DBSCAN model is established. This framework is designed to achieve optimal adaptive matching of model parameters in a variable data density environment, thereby enhancing the effectiveness and technical universality of trip ends identification. A significant number of synchronous comparative empirical experiments have been conducted in Guiyang City. The results demonstrate that, within an environment characterized by uneven spatiotemporal positioning, the effective accuracy of individual's trip ends identification reaches 90.98%, with an average coordinate distance error of 344.13 meters. Both the starting and ending time errors of trip ends are maintained below 3 minutes. In comparison to KANN-DBSCAN, ST-DBSCAN, and DBSCAN algorithms, the accuracy has undergone a significant enhancement, ranging from 9.62% to 23.45%, resulting in better accuracy and stability. This study can offer robust support in analyzing the characteristics of residents' travel activities and demand, while simultaneously enhancing the effectiveness of transportation planning schemes.

To address the conflict between bus priority and the traffic efficiency of general vehicles, while also considering the green wave effect, this paper proposes a bus and general vehicle collaborative green band optimization model based on intersection repeated release. By introducing a repeated release mechanism at the intersection, dynamic collaborative optimization between buses and general vehicles at the same intersection is achieved. The model first defines the concept of a "green wave pair" and proposes an optimization method aimed at maximizing the weighted sum of the green bandwidth, with consideration of the different vehicle types' needs and the stop-and-go problem in the front and rear segments of the green wave. During the optimization process, factors such as signal phase differences, whether repeated passage occurs, and signal cycle time are considered, and pedestrian crossing and green wave constraints are established. To validate the effectiveness of the proposed model, this paper uses Zhongshan Road in Zhenjiang as a case study and conducts a simulation analysis using VISSIM software. The results show that compared to the Multiband and overall trunk bus green band optimization control methods, the proposed model effectively reduces the average delay time and the number of stops for general through and left-turn vehicles and buses by 30%~34%, and optimizes the green bandwidth by 32%~40%.

Reinforcement learning has strong adaptability and learning ability, which can continuously adjust strategies and behaviors based on changes in the environment and feedback signals, thereby achieving continuous optimization and providing new technological means for urban traffic signal control. In response to the low efficiency of intelligent agent collaboration and the lack of control area partitioning mechanism in existing reinforcement learning methods for traffic signal collaborative control, this paper proposes a traffic signal hierarchical collaborative control architecture. By constructing intersection intelligent agents, this paper performs the correlation and collaborative design of state space and reward function. A traffic control sub zone partitioning model based on congestion diffusion is established to dynamically partition traffic control sub zones. The regional coordinated traffic signal control framework is established by deep reinforcement learning, and the information interaction mechanism is proposed with multi-level agents. The dynamic division method of traffic control sub-area is developed based on the congestion diffusion relationship of urban intersections. The results show that, compared with the existing timing control and reinforcement learning methods, the average travel time of the proposed method is reduced by 56.78% and 29.23%, respectively. In addition, the proposed method has certain portability in the road network of homogeneous intersections.

Traditional traffic signal control based on either discrete or continuous actions often fails to adapt to the spatiotemporal variability of traffic flows in urban intersections. Existing reinforcement learning (RL) approaches are unable to manage hybrid action spaces effectively, particularly with respect to scalability and interdependence among actions. To address these challenges, this paper proposes a novel hybrid action representation reinforcement learning method for intelligent traffic signal control at urban intersections. Firstly, the action space of each intersection agent is formulated as a combination of the selection of discrete signal phases and the corresponding continuous duration of green lights, through a consistent design for state and reward space. Secondly, a conditional variational autoencoder (CVAE) is employed alongside a discrete action embedding table to encode the original hybrid action space into a continuous latent representation, thus transforming the hybrid policy learning problem into a tractable continuous policy optimization task. Thirdly, the proximal policy optimization (PPO) method is then used to train policies within the latent space, and then the learned actions are decoded back into the original hybrid action domain for real-time interaction with the environment. Finally, experimental evaluations, using real-world data from the Beijing High-Level Autonomous Driving Demonstration Zone, show that the proposed approach reduces the average delay time, average queue length and average number of stops by 2.57% to 14.84%, 4.00% to 9.15%, 7.25% to 20.69%, respectively, which demonstrates the effectiveness of proposed approach in optimizing urban intersection control compared to the state-of-the-art benchmark models.

Ramp vehicle platoons can merge into the mainline utilizing gaps created by mainline vehicles' lane changing. Based on this, a Gap-Optimized Cooperative Merging Control (GOCMC) method is proposed for the mixed traffic flow with both Connected and Automated Vehicles (CAVs) and Connected Human-driven Vehicles (CHVs) in connected environment of freeway merging areas. GOCMC constructs a comprehensive benefit model of lane changing on mainline and ramp flow, with the goal of maximizing comprehensive benefit, achieves the cooperative control of mainline vehicle lane changing and ramp vehicle formation. Then, differentiated control of downstream vehicle trajectories is implemented based on vehicle types and functional characteristics. The simulation results show that under different traffic flows, GOCMC can increase the average speed and reduce the average delay for vehicles passing through the control area. When the traffic flow demand is relatively high (i.e., 1800 veh·h-1 ·ln-1 ), the average speed can still be increased by 24.21%, and the average delay can be reduced by 49.50%. Compared with the Cooperative On-Ramp Merging Control (CORMC) method, GOCMC exhibits better traffic efficiency at low penetration rates and high ramp flow ratios. The sensitivity analysis shows that increasing the penetration rate of CAV can improve the traffic efficiency, and this effect is more significant in low penetration scenarios. The compliance of CHV has limited effect on improving traffic efficiency, but GOCMC effectively reduces the impact of the randomness of CHV lane changing behavior through periodic optimization.

Under heterogeneous traffic conditions, this study proposes an integrated control model which simultaneously optimizes signals and trajectories to address the coordination problem between traffic signal control and vehicle trajectory planning. The model employs a Dueling Double Deep Q-Network (D3QN) through deep reinforcement learning approach to achieve the dual objectives of improving traffic efficiency and promoting eco-driving. The comprehensive validation of proposed model was conducted using the SUMO simulation platform. The simulation results show that, compared to the baseline model, although single-objective optimization strategies can partially enhance the performance of intersection, there are some limitations in overall efficiency improvement. In contrast, the proposed integrated control model effectively combines the optimization of macroscopic traffic flow with microscopic vehicle behavior adjustment, achieving a 66.99% reduction in average vehicle delay, an 11.26% decrease in fuel consumption, and significant reductions in CO2 and other pollutant emissions. Further sensitivity analyses reveal the performance of system trends under varying CAV penetration rate, indicating that performance gains gradually plateau beyond certain penetration thresholds. Moreover, the model demonstrates stable optimization effects under different traffic demand conditions, confirming its adaptability and robustness for urban intersection environments.

Aerial images of traffic scenarios are characterized by complex backgrounds, uneven distribution of vehicle aspect ratios, and dynamic variations in vehicle heading angles, which often lead to missed or false vehicle detection. This paper proposes an improved YOLOv8-OBB (You Only Look Once version 8-Oriented Bounding Box) network tailored for detecting vehicles with different heading angles in aerial images. First, a Large Selective Kernel Attention Mechanism (LSKAM) was integrated into the network's neck to enhance feature extraction capabilities for vehicles with varying aspect ratios. To improve the distinction between backgrounds and targets, a deep feature extraction module with a dimension of 10×10 was added to the Path Aggregation Network (PANet) in the head. Then, a VoV-GSCSP (VoVNet GSConv Cross Stage Partial) based lightweight module was embedded into the neck of the network to balance detection accuracy and speed. Experimental results on the large-scale Drone Vehicle dataset show that the proposed method outperforms typical detection methods such as Oriented-R-CNN(Oriented-Regions with Convolutional Neural Networks), R-YOLOv3-tiny, YOLOv6-OBB, YOLOv8-OBB and YOLOv12-OBB in terms of detection accuracy and computational complexity. Specifically, the detection accuracy for "Car" and "Bus" categories exceeds 95%, with a mean average precision (mAP) of 73.7% and a computational complexity of 26.9 GFLOPs (Giga Floating-Point Operations per Second) for all types of vehicles selected in the experiment. Additionally, verification using data collected in the field by drones indicates that the proposed method can effectively reduce missed and false detection, thereby fulfilling the requirements for vehicle detection tasks from an aerial perspective.

The trajectory data of vehicles has numerous applications in intelligent transportation systems (ITS). However, their practical effectiveness is often hampered by data missing. Although the rapid development of radar-and-video-fused perception technology has enabled all-day collection of vehicle trajectory data, some challenges still persist in intersection scenarios, such as the insensitivity of radar to stationary targets and the occlusion by large vehicles. To address the missing data of vehicle trajectory in intersections, this paper proposed a novel completion algorithm (Transformer-Full-Velocity-Difference, TF-FVD) based on physics-informed deep learning, which incorporates the supervision signal of the FVD car-following model into the training process of the Transformer deep learning model, and adds a traffic light state encoding module to account for the traffic rule constraints on vehicle movement. The experimental results based on the radar-video-fused trajectory dataset show that the introduction of the FVD model and the traffic light state encoding module led to the improvements of accuracy by 11.6% and 15.6% respectively. In public SinD (Signalized INtersection Dataset) dataset, the proposed TF-FVD model achieved a 25.3% accuracy improvement compared to the SOTA (State of the Art) data-driven algorithm. The distribution error of travel delay calculated from the imputed trajectories decreased by 9.14%, which implies its value in following applications.

The accuracy of vehicle trajectory prediction impacts driving safety significantly. However, conventional methods focus on the features of vehicle kinematic predominantly, while neglecting the comprehensive utilization of road environmental information. Furthermore, the existing approaches often encounter gradient vanishing issues in long-term trajectory prediction tasks, leading to a substantial performance degradation. To address these challenges, this study proposes a novel framework, Heterogeneous Multi-Graph Spatiotemporal Fusion, for Long-Term Vehicle Trajectory Forecasting. Initially, historical traffic data is decoupled into road environmental features and vehicle interaction patterns, which are subsequently modeled as distinct graph topologies-an environmental graph and an interaction graph. Subsequently, graph attention networks are employed to perform convolutional and pooling operations on each graph topology, capturing their spatiotemporal dependencies effectively. Then a gated fusion mechanism is introduced to adjust the contribution weights of environmental constraints and interactive behaviors dynamically by generating optimized fusion features. Finally, the integrated feature sequences are decoded through a Mamba network to produce long-term trajectory predictions. The simulation results demonstrate that, over a 5-second prediction horizon, the proposed model achieves a 22.8% reduction in average error, and a 32.6% decrease in terminal error, and an 18.9% reduction in root mean square error compared to the optimal baseline algorithm, which improve the long-term prediction accuracy significantly.

Aiming at the problems that the high-level decision-making of the current autonomous driving system lacks specific execution details and continuous learning ability, this paper focuses on applying the Large Language Model (LLM) in refining the decision-making process of autonomous driving. Based on the powerful reasoning ability of the LLM and the exploration ability of Reinforcement Learning (RL), this paper proposes a method of combining the LLM and RL to refine the vehicle decision-making process. First, based on the high-level actions output of the RL, the reasoning ability of the LLM is used to predict the future trajectory points of the host vehicle. Then, the output of the RL model is combined with the current state information to make a safe, collision-free and interpretable prediction of the next state. At last, the above driving decision-making process is vectorized and stored in the memory module as driving experience, and the driving experience is updated regularly to achieve sustainable learning. The trajectory points predicted by the LLM provide a detailed motion path for the Proportional-Integral-Derivative (PID) controller, providing a basis for adjusting the vehicle's acceleration and speed to ensure that the vehicle travels along the predetermined path. In addition, the trajectory prediction can also evaluate and avoid potential collision risks, and create a safe path by analyzing the traffic state and historical data. The results of the closed-loop experiment show that the proposed decision-making method outperforms other models in all evaluation indicators. Compared to the RL, the decision-making method based solely on the LLM, and the LLM-based car-following model, the driving scores are increased by 35.12, 14.33 and 12.28 respectively. The method with the memory module increases the driving score by 25.59 compared to the method without the memory module.

Aiming at the problems of misdetection, omission and low confidence caused by the rich background and the aggregation of multiple small targets in UAV aerial images, this paper proposes a target detection algorithm to improve YOLOX. First, an attention mechanism (LE-MSA) is designed to avoid small target features disappearing into redundant information by extracting high-frequency feature information. Second, in order to prevent the problem of poor detection effect caused by the imbalance of sample categories, a VarifocalLoss loss function is introduced to participate in the improvement of classification accuracy and target frame localization accuracy, together with the BCEWithLogitsLoss loss function. Finally, a multi-strategy data enhancement method, including adaptive small target enhancement and region enhancement methods, is proposed to improve the generalization ability of the model. The experimental results show that the LE-YOLOX algorithm exhibits good detection ability with a detection accuracy of 90.78%, which is better than Faster R-CNN (71.30%), YOLOv5 (88.15%), YOLOv8 (87.63%), YOLOv10 (86.1%) and YOLOX (87.82%). Meanwhile, the improved YOLOX is able to solve the problem of misdetection and missed detection of dense small targets under UAV aerial images in actual detection effectively, and it has strong small target recognition and dense target processing capabilities.

To address abnormal passenger behavior incidents in metro stations timely and effectively, this paper proposes a two stage fusion model BiFuseNet based on computer vision. The model integrates the lightweight detection network LMD-YOLO and the efficient classification network based on EfficientformerV2 to achieve efficient and accurate abnormal behavior detection. In the first stage, the model incorporates lightweight convolution aggregation blocks (LCAB), the mixed convolution aggregation blocks (MCAB), and the dynamic detection heads (DyHead), which not only reduce the size of model but also enhance the detection capability for small and occluded objects. In the second stage, a multi-level weighted fusion strategy is employed to optimize detection and classification results, further enhancing the robustness of the model. The experimental results show that BiFuseNet achieves an accuracy of 89.3% on the self-built MetroAB dataset, which is 6.1% higher than that in traditional models, and realizes a detection speed of 43.7 frames per second (FPS). On the PASCAL VOC and VisDrone public datasets, the model improves accuracy by 10.1% and 2.7%, respectively, further verifying the advantages of the model in detecting small and occluded targets, as well as its excellent generalization ability. Through these innovative designs, BiFuseNet enhances the efficiency and accuracy of abnormal passenger behavior detection in metro stations significantly.

Modular Transit Vehicles (MTV) represent an innovative form of public transportation, with the core feature of being able to flexibly combine and split modular units according to passenger demand. This characteristic helps optimize resource utilization, operational efficiency, and passenger comfort within public transportation systems. In recent years, with the development of intelligent transportation technologies, the MTV scheduling strategies have become a research hotspot. Therefore, this study summarizes and analyzes the current research on MTV scheduling and evaluation methods. First, the paper introduces how existing MTV scheduling studies are categorized and the content included in each category. Then, the related research on MTVscheduling is classified and organized, and the performance evaluation indicators of MTV are summarized. The classification is primarily based on the service scope of MTV scheduling in existing studies (single line service scope and line network service scope). It is then further refined according to the locations of stops where modular units can be combined or split and the characteristics of MTV operational lines (fixed/flexible single line scheduling and fixed/flexible line network scheduling). Additionally, specific MTV service modes are covered (fixed-route transit, feeder bus transit, direct bus transit, shuttle bus transit, demand-responsive transit, customized bus). At last, the paper summarizes the limitations in the research content and provides suggestions for potential future research directions: exploring passenger transfer schemes within the vehicle, MTV charging and battery swapping strategies, enriching research scenarios, focusing on MTV infrastructure development, and raising public awareness of MTV.

To address the insufficient ride-pooling order correlations in existing ride-pooling demand prediction, this paper proposes a spatiotemporal prediction model for ride-pooling origin-destination demand based on a spatiotemporal multi-graph convolutional neural network. First, ride-pooling order data is processed into time-series data representing demand between origin destination points, and multiple characteristic features of ride-pooling demand are analyzed and extracted. Based on this, multiple semantic graphs reflecting ride-pooling order correlations are innovatively constructed using ride-pooling success rate, route similarity, and land-use similarity, with a novel semantic modeling perspective for ride-pooling demand prediction. Meanwhile, geographic graphs are built to capture adjacency and distance relationships, enabling a multidimensional modeling of geographical correlations and travel semantic associations. For model development, this paper proposes a hierarchical multi-graph information fusion mechanism to fully capture spatiotemporal correlations in the data. Additionally, exogenous factors affecting ride-pooling demand are incorporated to develop a spatiotemporal multi-graph convolutional model that integrates multiple features. Experimental results show that ride-pooling success rate, route similarity, and weather are key factors influencing ride-pooling demand. Compared to the Multi-Graph Attention Network (GMAN), Spatiotemporal Long Short-Term Memory Network (SP LSTM), and Residual Multi-Graph Convolutional Network (RMGCN), the proposed method reduces root mean square error by 11.44%, 7.06%, and 3.89%, respectively, and decreases mean absolute error by 9.45%, 10.85%, and 7.26%, respectively. These results demonstrate that the proposed method achieves higher prediction accuracy and scientific validity.

This study integrates microscopic simulation and machine learning to overcome the shortcomings of existing methods in the characterization of weaving flow and the quantification of factors. It constructs a research framework from the simulation calibration, analysis on factor influence, and estimation of traffic capacity. The proposed DIEGA (DBSCAN Information Entropy Genetic Algorithm) method combines DBSCAN (Density-Based Spatial Clustering of Applications with Noise) clustering, information entropy, and genetic algorithms. Simulation experiments explore the relationships among the weaving length ( LW ), on ramp flow ( QRF ), off-ramp flow ( QFR ), and traffic capacity in a weaving area. A stacking-based model of capacity estimation is developed with the SHAP (SHapley Additive Explanation) analysis to reveal how each factor exerts its influence. The results show that DIEGA holds the delay error for each weaving direction below 3%, and converges faster by 22.2% than a traditional genetic algorithm. Under a constant total weaving flow, different proportions of QRF and QFR lead to about 15% fluctuations in capacity. A nonlinear coupling is observed among QRF , QFR , and LW . Among the stacking models, ML_RF ( R2 =0.969) outperforms other stacking approaches and single models. SHAP indicates that when QRF /QFR is near 1 and LW ranges from 250 to 350 meters, a peak capacity of 4635~4860 pcu per hour can be achieved.

To address the abrupt changes in conflict indicators for vehicles on highways, this paper proposes a real-time prediction model for longitudinal conflict risk based on trajectory data to improve the accuracy of vehicle conflict prediction. The model adopts Time-to-Collision (TTC) as a surrogate safety measure for longitudinal conflict, through transforming a discontinuous indicator prediction into a continuous time-series prediction of speed parameters. A TTC real-time deduction module is used to output the predicted conflict risk values. A time-series Transformer is employed to achieve high-precision predictions, and an adaptive correction module is integrated to address error biases caused by the subjective behaviors of drivers during conflict situations. When the current conflict indicator reaches the threshold, a short-term acceleration fitting is activated to correct the predicted values of transformer using the fitted acceleration. The effectiveness of model is validated on real-world vehicle trajectory data. The results show that the proposed model outperforms benchmark models in performance metrics. Compared with the baseline Transformer model, the Adaptive Risk Adjustment Transformer model (ARA-Transformer), which incorporates an adaptive bias correction module, reduces MSE by 48.33%, RMSE by 21.33%, and MAE by 24.10% under conflict conditions. Additionally, the proposed model demonstrates generalizability across different driver trajectories by providing an effective method for conflict warning and improving system risk intervention responsiveness in assisted driving scenarios.

To investigate the characteristics of vehicle conflicts within maintenance work zones and reduce the risk of conflicts during lane-changing maneuvers, this paper collected the video data of urban road maintenance work zones by drones and extracted the trajectory data for conflict prediction. The study analyzed the impact of static protective facilities and the characteristics of surrounding vehicles on conflict risk. By introducing three key indicators: Extended Time to Collision (ETTC), Stopping Distance Index (SDI), and Work Zone Time to Collision (WZ-TTC), and applying the CatBoost algorithm, the study performed high-accuracy conflict risk prediction with 93.3% accuracy on the test set. It was found that within maintenance work zones, the relative speed and distance between the primary vehicle and surrounding vehicles, as well as the distance between vehicles and static facilities, are the main factors influencing conflict risk. The deployment of static facilities, such as guardrails, significantly impacts traffic conflict risk, and this influence gradually weakens as the distance increases. When the distance between a vehicle and a guardrail is less than 20 meters, the relative speed of the front-left vehicle becomes the most critical risk factor. As the distance increases to the range of 20 to 30 meters, the influence of the rear-left vehicle's speed becomes more prominent. When the distance exceeds 30 meters, the constraining effect of the guardrail essentially disappears. Additionally, vehicle parameters in the left lane have a greater impact on conflict risk compared to those in the right lane.

Under the cross-line operation mode of intercity and suburban railways, the collaborative optimization of train timetabling and train stop planning can reduce passenger transfers and enhance travel quality. First, this study considered the hour dependent origin-destination passenger flow as the input of demand, and used candidate stopping patterns to reflect the possible train stops under the cross-line mode. With the help of space-time network representations, the arc-based variables to address train running and passenger travel were introduced respectively. Then, multi-commodity flow constraints were established to describe the train operation trajectory and the passenger travel process. Coupling constraints were used to achieve the space-time matching of train movements and passenger travel. A mixed-integer linear programming model was formulated to minimize train operating costs and passenger travel costs. Under the Lagrangian relaxation framework, the model can be decomposed into three tractable subproblems: train time-space path, stopping pattern selection and passenger demand assignment. Further, A heuristic method were designed based on dual solutions to obtain feasible solutions of the primal problem. Finally, two real-life numerical experiments were conducted to assess the efficiency and effectiveness of the proposed approach. The results show that the proposed approach can generate train service plans that effectively responds the demand of passengers. Specifically, more than 80% of cross-line passengers choose direct train services to improve the smoothness of travel. Moreover, the generated train stop patterns generally align with the passenger flow distribution at each OD pair. In a case with 17 stations and 80 trains, the proposed method can yield solutions superior to those obtained by the solver within a reasonable computation time.

To address the impact of uncertain passenger load variations on train operation in urban rail transit, this paper proposed a real-time robust optimization method for target speed profiles considering load uncertainty. The implementation included three parts: model design, model training, and model validation. First, a reinforcement learning model for train operation was developed based on the Markov decision process, with reward design balancing the robust optimization of performance metrics and control strategies. Second, the model training convergence performance was enhanced by employing the Potential-Based Reward Shaping (PBRS) technology. Real-time response to passenger load changes was achieved through the Deep Q-Network (DQN) value function estimation. At last, the effectiveness of the model was validated via simulation cases based on train operation scenarios of a Beijing subway line. The simulation results show that the DQN-PBRS algorithm achieves an average computation time of 26 millisecond, enabling real-time generation of target speeds. The generated speed profiles exhibit better robustness under extreme load and load variation conditions compared to the DQN algorithm, while also reducing energy consumption by more than 5%. By conducting a sensitivity analysis of key hyperparameters in the algorithm, the optimal hyperparameter combination for the best training performance was determined.

Virtual coupling (VC) has demonstrated significant potential for enhancing railway transport capacity. However, its operational efficiency and safety are severely compromised by safety risks arising from unreliable communication environments. To address the latency sensitivity challenges in VC systems under unreliable communication environments, this paper proposes a distributed online optimal control framework to enhance both stability and safety guarantees for train platoons under time-varying delays. Specifically, a bidirectional communication topology-based platoon dynamics model is established to characterize the inter train interaction dynamics under asymmetric time-varying delays. A model predictive control (MPC) framework is designed to achieve an effective balance between formation coordination and fuel economy, while accounting for safety constraints and the inherent characteristics of individual trains. Furthermore, string stability criteria are rigorously derived and integrated as real-time constraints to ensure inter-train stability. The experimental results demonstrate that the proposed control method can address the impacts effectively caused by system-varying delays, shorten the time required for VC formation, reduce the maximum fluctuation amplitude, and outperforms traditional methods in efficiency across scenarios by 35%~40% , which demonstrates significant advantages in enhancing platoon formation efficiency and stability.

The track sections play a critical role in train operations within a railway station. The failure of track sections can significantly reduce transportation efficiency and service quality. This study proposes a methodology to identify critical track sections based on complex network theory. Using signaling layouts and interlocking tables as primary data sources, the study first establishes a multi-layer network model of the station and its weighted projection network. Then, 12 representative node importance metrics including degree centrality, closeness centrality, and PageRank are integrated through an improved Rank-Sum Ratio method to derive comprehensive node significance rankings. The validity of the proposed methodology is verified through mapping the resulting ranking in station yards and analyzing network efficiency changes caused by node removal. A case study of a typical railway station demonstrates that the proposed method successfully establishes a multi-layer network consisting of 7 layers and 72 nodes, and derives a weighted projection network based on the actual operational volume proportions. The identified critical track sections are located at the throat areas of arrival and departure routes, which are consistent with the actual operational bottlenecks. Sequential removal of the top 10, 20, and 33 ranked nodes leads to network efficiency reductions of 37.57%, 51.26%, and 97.28%, respectively, confirming the method's marked effectiveness. Furthermore, the correlation analyses confirm that the proposed method effectively integrates topological, communicative efficiency, and influence features, thereby offering a comprehensive and robust evaluation of node importance in multi-layer complex networks.

In order to conduct differential treatment of trains based on their differences during the process of train timetables compilation, this paper first classifies the trains into different priorities according to their speed grades, the number of mandatory stops, and operational benefits. Then, based on the train operation space-time network, a train timetables compilation model considering train priorities is established with the objectives of minimizing the total travel time of trains, minimizing the deviation between the departure time and expected time of trains, and minimizing the cost of train cancellations. Considering the differences in the urgency of occupying the directed arcs in the space-time network by trains of different priorities, an arc combination weight optimization strategy considering train priorities is designed, and a compilation algorithm for train timetables is designed with this strategy as the core. Finally, the effectiveness of the algorithm is verified by taking 143 trains from Beijing South Railway station to Nanjing South Railway station as an example. The result shows that compared with the method not considering train differences and only considering the benchmark train, the travel speed of high-priority train is increased by 13.7 km·h-1 and 2.1 km·h-1, the operational benefits are increased by 0.6% and 3.8%, and the total penalty cost from the deviation of departure time is reduced by 46% and 27%. The results prove that considering the differences among train attributes can better improve the quality of train operation diagram.

To address the significant differences in fuel consumption among various types of vessels during port entry and the challenge of balancing scheduling efficiency with energy optimization, this study investigates a coordinated optimization approach for port entry efficiency and fuel consumption considering the heterogeneity in fuel efficiency. A bi-objective mixed-integer nonlinear programming model is formulated to minimize the total port entry time and the overall fuel consumption of vessels. To solve the model effectively, a cooperative meta-heuristic algorithm driven by a Deep Q-Network (DQN) is proposed. The algorithm incorporates an improved Nawaz-Enscore-Ham (NEH) heuristic to generate initial scheduling sequences and designs a DQN-based dual-population cooperative search framework to adjust vessel entry sequences and speed profiles dynamically. Using a representative scheduling scenario from Tianjin Port, numerical experiments are conducted to validate the performance of the proposed method. The results demonstrate that the DQN-driven cooperative meta-heuristic algorithm outperforms traditional heuristic methods in terms of solution quality and distribution within the objective space. Compared to the commercial solver CPLEX, the proposed algorithm achieves exponential improvements in computational efficiency while maintaining the overall deviation of the two objectives within a range of 2.04% to 12.82%, yielding consistently high-quality approximate solutions. Further comparative analysis reveals that incorporating heterogeneous fuel efficiency significantly alters the scheduling priorities of vessels, underscoring the critical impact of energy consumption structure on port entry organization strategies.

The waybills for the transportation of hazardous goods are usually filled in and reported by transport agencies before shipment in China. Therefore, the related time stamps of origin and destination are often inaccurate, which limit the effectiveness in transport management. A deep learning approach using the trajectory data of hazardous goods transportation vehicles is proposed to improve the quality of waybills. A trajectory imputation model incorporating a multi-head self-attention mechanism is firstly constructed to complete missing trajectories, which can solve the positioning accuracy problems caused by the missing real time vehicle positioning data. Then, based on the characteristics of hazardous goods transportation, a dual-layer clustering algorithm integrating an adaptive mechanism and multiple threshold rules is designed to identify the times and geographic coordinates of origin and destination for a specific waybill. The waybill can be enhanced by extracting precise origin and destination addresses from points of interest around the cluster centers using a text recognition model. The proposed methods were tested using the data of LNG transport vehicles in Guangdong, China. The results indicate that the Mean Absolute Error (MAE) and Mean Absolute Percentage Error (MAPE) of the trajectory imputation model ranged from 2.34~3.33 and 6.05~7.74 under various data missing rates respectively, which outperform other baseline models. The accuracy of identifying the times and geographic coordinates of origin and destination reaches 98.35%. The text recognition model achieves an accuracy of 92.83% in identifying the address information of selected point of interests(POI). The proposed method enables high-precision inversion and correction of key fields in pre-filled waybills, thereby providing important theoretical support and practical guidance for the safety management of hazardous materials transportation.

To investigate the relationship between the human respiratory index and controlled workload, a simulated air traffic control experiment was conducted by the selected 27 subjects to collect and analyze their respiratory parameters under different types of workloads. First, according to the calculation results of Spearman's rank correlation coefficient, the respiratory parameters which are significantly correlated with mental and physical control workloads were obtained. Then, based on the ordered logistic model method with significantly correlated respiratory parameters as independent variables and five graded levels of mental and physical regulatory workload as dependent variables, the prediction models for the severity of mental and physical workload and the likelihood ratio test were constructed and goodness-of-fit tests were carried out. Furthermore, the ROC (Receiver Operating Characteristic) was plotted to evaluate model performance, and finally Confusion Matrix was used to verify the prediction accuracy of model. The results showed that among the respiratory parameters, the respiratory cycle was significantly correlated with mental workload, and the respiratory cycle, respiratory amplitude and inspiration-exhalation ratio were significantly correlated with physical workload. At a significance level of 0.05, the constructed prediction models for mental workload and physical workload severity demonstrated the satisfactory goodness of fit and exhibited certain detection capabilities, with the overall AUC (Area Under Curve) values reaching 0.679 and 0.753 respectively. The evaluation results of Confusion Matrix showed that the prediction model demonstrated optimal performance in identifying high mental and physical workload conditions, achieving highly prediction accuracies of 88.9% and 83.3% respectively. The research results of this article can provide a certain reference value for monitoring control workload based on respiratory parameters.

Aiming at the tight gate resources and dynamic scheduling requirements in the context of the recovery of the civil aviation industry, a multi-objective optimization model for gate assignment integrating economy and robustness is proposed. First, a comprehensive economic objective function is constructed, which includes the costs of taxiing fuel consumption, gate service, and idle time. Then, a Backtracking-Hybrid Particle Swarm Optimization (BH-PSO) algorithm is proposed to alleviate the problem of traditional Particle Swarm Optimization (PSO) algorithm due to the excessive dependence of parameter initialization, and to enhance the global search ability. In addition, a dynamic robustness model based on the Generalized Extreme Value Distribution is further proposed for the uncertainty of flight schedules. To verify the effectiveness of the proposed model, an experimental analysis is conducted using a case of 142 flights at a regional hub airport in China. The experimental results show that the improved BH PSO algorithm reduces the comprehensive cost by 8590.5 yuan compared with the PSO algorithm, with a decrease of 3.63%, and the convergence speed is increased by about 50%. Through the verification of 100 simulation scenarios, the dynamic robustness model reduces the number of conflicts by about 20% on the premise of a cost increase of 0.62% to 1.28%, which is significantly better than the static optimization scheme. The relevant research conclusions alleviate the static limitations of traditional models to a certain extent, provide decision-making support for smart airports to deal with dynamic flight changes, and have important reference value for improving resource utilization and operational efficiency.

The emission reduction benefits of plug-in hybrid electric vehicles (PHEVs) highly depend on their power mode, and assessing their reduction potential accurately requires a deep understanding of the interplay between various influencing factors. This study, based on over seventy million seconds of second-by-second driving data collected in Beijing, investigates the impacts of driving distance, charging behavior, and battery capacity on PHEV power mode selection. The results indicate that the utility factor (UF) exhibits nonlinear sensitivity to the number of charging piles per vehicle (CPPV). When CPPV is less than 0.6, increasing it by 0.1 can reduce daily CO2 emissions by approximately 637 tons; however, beyond this threshold, the reduction benefit declines to 81 tons per day. Furthermore, the study finds that under adequate charging accessibility, PHEVs with smaller battery capacities can still maintain a high proportion of electric driving. For example, a PHEV with a 20 kWh battery can achieve a UF of 0.82 when CPPV reaches 0.5, which is sufficient for daily electric travel demand. Therefore, deploying low-power charging infrastructure strategically in residential areas can enhance PHEV electrification levels effectively, reduce reliance on high-capacity batteries and high-power charging piles, and alleviate the load pressure on the power grid caused by concentrated charging.

This study aims to reveal the characteristics of vehicle conflict risk during car-following on through-village highways, focusing on risk severity, duration, and influencing factors. Data is collected on a typical second-class through-village highway in Yunnan Province using two-day UAV aerial recordings of car-following maneuvers. A two-dimensional Time-to-Collision (TTC2D ) method is employed, incorporating vehicle collision angles, for a more precise conflict risk assessment. For each vehicle, the 15% quantile of TTC2D serves as a threshold to identify high-risk trajectory segments and calculate their durations. A survival analysis is conducted to assess the impact of key factors on the duration of risky car-following conflicts, which includes car-following maneuvers of preceding and following vehicles, and characteristics of surrounding pedestrians and traffic flow. Results reveal that the classical TTC method underestimated collision time by an average of 1.15 seconds when vehicle collision angles exceed 20 degrees, whereas TTC2D provides more accurate conflict risk estimates. The average duration of risky car-following conflicts on through-village highways is 3.6 seconds, with the Weibull Accelerated Failure Time (AFT) model demonstrating optimal fitting performance for these durations. An increase in the longitudinal speed of the following vehicle by 1 m·s-1 extended the duration of risky car-following conflicts by 30.21%. Conversely, a 1 m·s-1 increase in the longitudinal speed of the preceding vehicle and a 1 m increment in its minimum collision distance reduces the exposure duration by 23.66% and 3.44%, respectively. In situations where vehicles are exposed to risky car-following conflicts for more than 6.13 seconds, had a longitudinal velocity difference exceeding 6.2 m·s-1, and a shortest collision distance of less than 12.53 m, there was a higher likelihood of experiencing a high-risk car-following conflict.

To investigate the influence of the horizontal and vertical alignments of mountainous two-lane highways on drivers' visual workload, this study conducted a naturalistic driving experiment with 28 participants. The experiment utilized the Dikablis eye-tracking glasses and the CTM-8A non-contact multifunctional speedometer to collect visual response data and speed values, which were subsequently validated for accuracy. By analyzing the drivers' pupil area variation rate, fixation duration, and blink frequency across different gradients, horizontal curve radii, and combined curve-slope sections, the study quantitatively assessed the impact of alignment conditions on visual workload. A visual workload model was developed using factor analysis and entropy weight methods, and thresholds for visual workload levels were determined based on clustering algorithms. The results indicate that in sections with gradients ranging from 1.50% to 6.00%, the pupil area variation rate and fixation duration are positively correlated with the gradient, whereas the blink frequency is negatively correlated. Specifically, under the same gradient, the pupil area variation rate is higher in downhill sections compared to uphill sections. Additionally, drivers' visual workload is negatively correlated with the horizontal curve radius. In sections with smaller radii, visual workload is primarily influenced by limited visibility, whereas as the curve radius increases, the pupil area variation rate and fixation duration decrease, while the blink frequency gradually increases. Based on the analysis, visual workload can be classified into three levels: low, medium, and high. It was also found that when the combined curve-slope index is less than 23.4 %·km-1, the visual workload remains at a medium level or below. The study results provide a theoretical foundation for improving the safety and comfort of mountainous highways.