Multi-Modality Spatio-Temporal Forecasting (MoSTF) extends traditional spatio-temporal forecasting by incorporating diverse traffic modalities. Despite significant recent strides in spatio-temporal modeling, existing approaches often fail to explicitly model the coupling relationships between different modality variables. Accurate MoSTF is challenging, as it requires modeling (1) temporal dynamic heterogeneity under exogenous influences and (2) heterogeneous spatial dependencies alongside complex cross-variable couplings. To address these challenges, we propose the Dual-Domain Spectral Filtering Network (DSFNet). Our framework employs dual-domain spectral filtering to capture heterogeneous spatial patterns and explicitly model the relationships between variables. Unlike graph-based message passing or dense attention over node-modality pairs, DSFNet factorizes space-modality interactions into feature-domain and spatial-domain spectral operators, enabling scalable modeling of nonlocal dependencies and cross-modality couplings. Furthermore, we introduce an external gating mechanism to adaptively regulate temporal dynamics under external influences. We validate our method through extensive experiments on five representative real-world traffic datasets. Compared with the second-best baselines, DSFNet reduces MAE by 3.21

With the development of intelligent transportation systems and the increasing demand for transportation, traffic congestion on highways has become more prominent. So accurate short-term traffic flow prediction on these highways is exceedingly crucial. However, because of the complexity, nonlinearity, and randomness of highway traffic flows, short-term prediction of its flows can be difficult to achieve the desired accuracy and robustness. This article presents a novel architectural model that harmoniously fuses bidirectional long–short-term memory (BiLSTM), bidirectional gated recurrent unit (BiGRU), and multi-head attention (MHA) components. Bayesian optimization (BO) is also used to determine the optimal set of hyperparameters. Based on the PeMS04 dataset from California, USA, we evaluated the performance of the proposed model across various prediction intervals and found that it performs best within a 5-min prediction interval. In addition, we have conducted comparison and ablation studies. This not only proves the effectiveness of the BO strategy but also highlights the advantages of the proposed model in improving predictive accuracy. These results indicate that our model can effectively handle the complexity of highway traffic flows and provide more accurate traffic flow predictions, thereby significantly improving the operational efficiency of highway traffic.

Moving Target Defense (MTD) proactively counters adaptive adversaries by continually changing system properties. This work improves transmission security along three MTD dimensions: routing dispersion, topology mutation, and address obfuscation. We propose the K-Randomized Routing Dispersion (KRRD) strategy, which uses multi-objective optimization to increase path diversity and limit key-node exposure while balancing node load. At packet granularity, our mutation mechanism increases per-packet dispersion and reduces the probability of sequential packet interception. We also present a decentralized topology-mutation algorithm that updates neighbor lists based on QoS, and an SDN-enabled identifier-based addressing scheme that applies per-hop address obfuscation to hide real endpoints. In experiments, under high-frequency attacks our method reduces packet eavesdropping by up to 63% compared with classical routing-mutation schemes. Under a simulated 100 Mbps link with 100 source-destination pairs, KRRD attains up to 1.5 times the throughput of flow-level mutation algorithms while preserving scalability and usability.

Interchanges are junctions where traffic merges, diverges and weaves between highways. Accurate prediction of ramp traffic flow at highway interchanges is essential for proactive traffic management involving flow control and ramp management. In practice, real-time mainline flow can be obtained through Electronic Toll Collection (ETC) systems and ramp flow can be inferred from license plate matching. However, data privacy restrictions and matching latency prevent real-time access to ramp data, leading to a "real-time blind spot". This discrep ancy dictates that while the model can utilize both mainline and ramp flow data during the training phase, the input is strictly limited to the real-time observable mainline flow during actual deployment. To address this chal lenge, this paper proposes a novel two-stage framework, including pre-training and prediction. In the pre-training stage, the Spatio-Temporal Decoupled Autoencoder (STDAE) leverages a proxy reconstruction task to mitigate missing ramp data issues. STDAE learns to reconstruct historical ramp flows exclusively from mainline traffic data, thereby compelling the model to capture the intrinsic spatio-temporal relationship between the mainline and ramp traffic flows. The unique decoupled architecture of STDAE, consisting of parallel spatial (SAE) and temporal (TAE) autoencoders, efficiently extracts the temporal and spatial dependency features of traffic flow. In the downstream prediction stage, the learned representations are integrated with specific forecasting models such as GWNet to enhance prediction accuracy. Comprehensive experiments conducted on three real-world in terchange datasets (QiLin, DanYangXinQu, and XueBu) across multiple sampling intervals (3, 5, and 10 minutes) demonstrate that our proposed combined model, STDAEGWNET, consistently outperforms thirteen state-of-the-art baselines. Specifically, STDAEGWNET achieves the best overall average ranks of 1.00, 2.11, and 1.44 across the three respective datasets evaluated by multiple metrics. At the 3-minute sampling interval, the model attains the lowest Mean Absolute Errors (MAE) of 4.89, 5.61, and 4.58. Furthermore, the model demonstrates strong ro bustness against missing mainline data scenarios, achieving an average MAE reduction of approximately 2.23% compared to baseline models without the STDAE module. Importantly, the architecture-agnostic nature enables STDAE to serve as a plug-and-play enhancement module to improve diverse forecasting pipelines. The code and datasets used in this study are publicly available at: https://github.com/ChaochaoSeu/RPaI-STDAE.

Parking reservation systems (PRS) are promoted as smart urban parking tools, yet their continued use remains limited because users face both technological uncertainty and schedule-related uncertainty. This study develops a behavioral analysis framework that combines structural equation modeling (SEM), a stated-preference binary logit model, and Random Forest learning. SEM examines how perceived usefulness, perceived ease of use, perceived risk, social influence, and behavioral attitude shape intention to reuse. The binary logit model examines whether users retain their reserved lot under 10 reservation mechanisms and three arrival scenarios. Random Forest is then used to test nonlinear prediction and interaction effects, with intention to reuse measured as the average of the two reuse-intention items and model performance evaluated by the conventional coefficient of determination (R2), mean squared error, and mean absolute error. The results show that perceived risk suppresses perceived usefulness and behavioral attitude, early and especially late arrival sharply reduce reservation retention, and discount intensity is the strongest positive operational lever. Random Forest additionally shows that the effect of perceived risk depends on perceived ease of use: a more intuitive interface buffers the negative effect of risk on predicted reuse intention. These findings indicate that behavioral uncertainty in PRS is simultaneously perceptual, situational, and interactive. PRS design should therefore combine flexible time management, transparent real-time information, and low-friction user interfaces.

The rapid expansion of the gig economy has increased instant delivery crashes (IDCs), yet the determinants of rider disability remain unclear. This study analyzed 633 adjudicated rider injury cases (2015–2023) from Chinese legal judgments to identify risk factors for the Highest Assessed Disability Grade (HADG). We integrated sociodemographic, vehicular, spatiotemporal, and crash-mechanism variables into ordered logit models to quantify risk determinants. Results indicate that while most adjudicated injuries were non-permanent, lower-limb fractures constituted the primary mechanism of established disability. Significant risk factors for severe disability included rider age over 50, migrant status, and collisions involving automobiles. Notably, a counter-intuitive “liability paradox” emerged: riders bearing secondary liability faced the highest risk of severe disability (OR = 1.943), surpassing those with primary liability (OR = 1.732) due to kinetic asymmetry. Addressing these distinct mechanisms requires a multi-layered framework. We recommend accelerating physical infrastructure segregation to decouple riders from high-energy motor traffic. Conversely, to curb the behavioral precursors of rider-at-fault crashes, we propose a “traffic safety credit system” combining dispatch suspensions with mandatory re-education. This study validates the utility of forensic data, offering actionable, evidence-based pathways to mitigate the severe human costs of the on-demand sector.

Accurate prediction of regional parking demand is crucial for optimizing intelligent parking guidance and management systems in urban intelligent transportation systems. This study introduces a deep learning framework based on a global-to-local attention mechanism, which integrates the strengths of Gated Recurrent Units, Graph Convolutional Networks, and attention mechanisms into a Multi-Graph Convolutional Gated Recurrent Unit Network. The framework is designed to capture complex spatiotemporal patterns in multi-step parking demand forecasting. The model consists of two primary components: the Global Demand Prediction Unit, utilizing gated recurrent unit layers to capture global temporal patterns, and the Local Distribution Prediction Unit, which combines graph convolutional network and gated recurrent unit layers to model spatial and temporal features at the individual parking lot level. To further enhance prediction accuracy, an attention mechanism decoder is incorporated to address multi-step prediction challenges and account for spatiotemporal variations in demand. The model was evaluated using real-world parking data from 43 roadside parking lots in Nanshan District, Shenzhen. Experimental results indicate that the proposed model outperforms baseline models, especially in multi-step parking demand prediction tasks. Specifically, the model achieves a 0.60% improvement in one-step prediction compared to models such as MGCN-GRU, and improvements of 1.97% and 2.10% in two-step and three-step prediction tasks, respectively. These results demonstrate its potential to support proactive parking guidance, dynamic parking management, and network-level parking resource allocation.

Ride-hailing drivers often face queueing challenges at airports owing to mismatches between passenger demand and vehicle supply, which can affect urban mobility efficiency. Queue length and waiting time are key performance metrics reflecting the interaction between airport passengers and ride-hailing services within the broader urban transport system. This study explores the impacts of multi-source exogenous factors on the queueing performance of airport ride-hailing vehicles and develops a generalizable analytical framework to support adaptive operation. The proposed optimal segmentation for circular samples (OSCS) algorithm enables time segmentation based on queueing dynamics while bypassing traditional calendar-based divisions. This segmentation allows precise regression and temporally interpretable analysis. Within each time segment, generalized additive models (GAMs) capture relationships between influencing factors and queueing metrics. Using Hangzhou Airport in China as a case study, we apply the OSCS-GAM framework, which divides the daily operational timeline into three statistically distinct segments, morning, afternoon, and night, revealing differentiated queueing characteristics. Regression results reveal that air passenger volume predominantly drives vehicle queueing dynamics, while other factor categories, including weather, urban traffic conditions, and weekday-weekend patterns, independently exhibit temporally heterogeneous impacts across the three daily segments. Our approach captures the temporal and contextual dependencies of airport ride-hailing vehicle queues, offering insights into optimizing resource utilization within urban transport systems. These findings inform municipal and airport authorities, as well as ride-hailing companies, supporting the development of adaptive operational strategies that promote efficient and sustainable urban transportation.

To optimize the freight distribution structure of ports and reduce carbon emissions from freight transportation, this paper develops a bi-level programming model for freight traffic shifting between roadway and waterway networks that incorporates carbon emissions. First, a complex freight network based on the roadway-water transport system is constructed, comprising roadway networks, inland waterway networks, maritime networks, and transshipment nodes. A traffic impedance model is then formulated within this complex network framework, integrating the roadway BPR function, the M/M/1 queuing model for lock passage time on inland waterways, and the M/M/c queuing model for port cargo handling into the impedance function. This allows micro-level congestion effects to be combined with macro-level traffic assignment. Next, a bi-level programming model for freight traffic shifting in the roadway-water network system is established, with carbon emissions incorporated. The NSGA-II algorithm is employed to determine the optimal carbon subsidy level, based on which the traffic distribution in the complex freight network is analyzed. Finally, the proposed model is applied to the roadway-waterway bimodal network in the Hangzhou Bay port area of Cixi. The results indicate that without subsidies, the waterway transport share is only 1.74%. The optimal subsidy efficiency frontier is identified at CNY 350,000/day, where the waterway share increases to 22.7% and carbon emissions decrease by 33.27 tons/day. The subsidy strategy evolves through three stages: first, prioritizing maritime shipping; second, jointly promoting inland and maritime shipping; and finally, shifting focus to infrastructure investment once subsidies reach saturation. This study offers a quantitative analytical tool for designing differentiated carbon subsidy policies to facilitate the road-to-waterway modal shift under fiscal constraints.

To meet the requirement of refined airspace risk assessment under the autonomous operation mode, this paper proposes an airspace risk calculation method based on potential field theory. First, an aircraft ellipsoidal risk zone is established to define the risk influence range. Based on the airspace grid and potential field function, a risk assessment model is constructed that considers the real-time impacts of distance, speed, heading, and other factors to model the radiation effect of aircraft operation risk. Second, two simulation scenarios are set up to validate the model's effectiveness and analyze differences in the spatio-temporal distribution of airspace risk. Additionally, multiple conflict scenarios are established. The correlation coefficients between conflict indicators and average airspace risk values for different flights are calculated to demonstrate the model's reasonableness. Finally, an airspace hotspot grid marking method is developed. Hotspot distributions are visualized across various grid scales, realizing multi-granularity awareness of the airspace safety situation from a spatial geographic perspective. Simulation results show the model effectively characterizes the real-time spatio-temporal distribution of airspace risk. As the number of simulated aircraft increases, the correlation between conflict indicators and average risk values becomes increasingly strong and stable, verifying the model's validity. Furthermore, multi-granularity grids reveal hotspot distribution details at different spatial resolutions, providing a reference for on-board auxiliary decision-making systems to make trajectory adjustments.

Predicting individual daily trip chain (IDTC) provides an important basis for information recommendation and transportation organization. This paper decomposes IDTC into six subproblems: whether to travel (P1), number of trips (P2), origin of travel (P3), departure time (P4), destination of travel (P5), and arrival time (P6). Using residual networks (ResNet), six submodels (M1–M6) are designed to predict these problems sequentially, which are then integrated into a full IDTC model. The method was applied to public transportation data from Weinan City, China. Input features from transit, weather, and POI data over 8 weeks were selected via random forest. Using 49 days of data, IDTC was predicted for the following 7 days. Results show the serial ResNet model predicts IDTC effectively. Key influences include weather on P1, travel time on P2, POI on P3 and P5, arrival time on P4, and departure time on P6.

Shared autonomous vehicles (SAVs) combine autonomous driving and sharing mobility, offering potential to reshape future travel modes. Their self-driving attribute is expected to significantly reduce parking demand and alter the parking landscape. Previous studies have primarily focused on the quantity of parking demand, with parking choice based solely on parking prices. The spatial distribution of parking demand remains unclear, especially during the transition period with both conventional vehicles and SAVs mixed on roads. An agent-based simulation model was developed to evaluate the impact of SAVs on parking demand from quantity and spatial distribution perspectives. A generalized cost function considered not only parking prices, but also the road toll and energy fee, was developed to alleviate the negative effect of SAVs’ endless cruising. We also explored the trend of such effects with varying SAVs’ market penetration rates. The results indicated a substantial decrease in parking demand (nearly 80%), spreading from the central business district (CBD) to the periphery, leading to a significant increase in vehicle miles traveled (VMT) within the entire network. We also found that the additional VMT is mainly due to SAVs’ empty travel during providing continuous services. To relieve the poor utilization of parking lots in CBD due to outspreading demand, a parking policy which can dynamically adjust parking price based on its real-time occupancy rate was proposed and verified. Results suggested that it would increase of the average utilization rate of overall parking lots by 8% and reduce the VMT caused by SAVs’ parking by 29.5% if the policy applied with a large pricing change coefficient.

Frequent light-to-moderate rainfall may influence metro demand differently across rainfall stages and station ridership rhythms. Using automatic fare collection data from 80 Hangzhou Metro stations and station-matched hourly meteorological observations, this study examines weekday entry-ridership associations with rainfall onset, continuation, and the first post-rain hour. Three station ridership-rhythm types were identified from non-rain weekday entry–exit profiles: destination-oriented, balanced bidirectional, and origin-oriented. A Bayesian hierarchical negative binomial model estimated stage-specific associations while accounting for station-hour baseline ridership, hourly temperature, peak periods, and station- and date-level heterogeneity. During off-peak periods, rainfall onset and continuation were associated with 4.95% and 6.82% reductions in expected entry ridership at balanced bidirectional stations. Continuation associations were more negative in the primary model, but this difference was substantially attenuated after accounting for cumulative precipitation and elapsed episode duration. Origin-oriented stations generally showed more negative onset and continuation associations, while negative first-post-rain associations were observed across all three station types. These findings demonstrate that routine rainfall responses vary across both event stages and station ridership rhythms, supporting more stage-specific and station-sensitive passenger-flow monitoring.

Instant delivery crashes (IDCs) pose a growing public health challenge. Standard crash databases rarely capture the fine-grained variables necessary to model severe pedestrian disability. To address this gap, we analyzed a sample of 732 adjudicated court verdicts involving pedestrian-IDC collisions. This specific data source captures a highly selected subset of litigated, severe events rather than the broader crash population. We applied a Hierarchical Generalized Ordered Probit (HGOP) model. This framework accommodates the ordinal nature of disability grades and unobserved age-related heterogeneity. To achieve model convergence, we consolidated fatalities and severe disabilities into a single highest-severity category. Our modeling reveals that older pedestrian age (65-74 years) and motorbike involvement strongly predict higher-grade disability. Conversely, female pedestrians and winter conditions correlate with lower injury severity. We also identified a negative monotonic association between rider liability and pedestrian injury grade. As legal rider responsibility increased, the predicted severity of pedestrian injuries systematically decreased. Lacking direct kinematic data, we hypothesize this liability association reflects distinct conflict typologies rather than direct physical causation. Translating these findings into effective harm reduction requires a two-tiered approach. Our statistical estimates directly support physical interventions, including age-proofed infrastructure and targeted motorbike regulations. Simultaneously, our descriptive behavioral data align with existing literature to advocate for platform governance reforms addressing algorithmic deadlines. Ultimately, these observational findings carry strict inferential boundaries. They apply exclusively to litigated crash populations and highlight the critical need for integrated clinical and kinetic data in future research.

This investigation explores the application of pre-trained language models (PLMs) in predicting multimodal travel behavior using data derived from a stated preference (SP) survey. The study primarily examines the capacity of PLMs to assimilate complex linguistic inputs and their effectiveness in modeling transportation behaviors over various travel distances. By converting numerical data into textual information and utilizing the sophisticated processing capabilities of PLMs, particularly the BERT and DistilBERT models, this research outlines marked enhancements in predictive accuracy compared to three traditional numerical-based models. Comparative analysis of evaluation results emphasizes the superior performance of PLMs in managing complex classification tasks that involve textual data, demonstrating their effectiveness over traditional models in all evaluated metrics. This finding indicates that PLMs is highly proficient in managing complex decision-making scenarios in transportation research, presenting a potent alternative to conventional modeling techniques.

This study addresses the issues of low efficiency and poor flexibility in conventional public transport services in low-density areas and proposes an integrated network optimization method for regular transit (RT) and demand-responsive transit (DRT) in an effort to achieve a balance between the operational supply costs and passenger travel demand costs in public transport operations. Considering the fleet size required for the lines, passenger walking distances, and the impact of transfers, the cost components of the joint service are derived, and a multi-objective joint optimization model is established. The model optimizes the routes of RT, the service areas of DRT, and the overall fleet size and identifies three different benefit preference plans: service priority, cost priority, and balanced approach for RT line schemes. Based on the spatial distribution of passenger travel and the desired time windows, clustering is performed to determine the optimal clustering points as temporary stops. The Multiple-Chain Chromosome Genetic Algorithm (MCGA) is then applied to further solve for the optimal service routes of DRT. Finally, the effectiveness and adaptability of the proposed method are verified using the Qilitang neighborhood area in Hefei as a case study. The results indicate that compared to the initial scheme, the optimized scheme under the balanced approach reduces passenger travel costs by 13%, lowers operational supply costs by 5%, and comprehensively enhances service levels while improving operational economy.

High-speed railways (HSR) are susceptible to disruptions due to a variety of factors such as extreme weather. Improving the resilience of HSR is crucial for minimizing losses and improving operation efficiency. This paper aims to strengthen the resilience of HSR by reducing network vulnerability and enhancing network reliability. An HSR spatio-temporal network (HSRSN) model is constructed to simulate trains' operation on railways. The model is grounded in the train timetable, combining infrastructure networks and train operations. Critical trains and critical nodes are components that exhibit reduced resilience when the network is subjected to disruptions. Percolation theory is used to identify the critical trains and the information entropy algorithm is introduced for identifying critical nodes. Additionally, a typhoon occurrence is chosen as the disruption for analyzing network vulnerability and connectivity. As for recovery post-disruptions, a strategy is proposed that utilizes timetable adjustments to mitigate the delays caused by disturbances. The performance of the proposed methods has been demonstrated in the case of the HSR network in Jiangsu Province, China. Results show that suspending critical trains during 13:00-15:00 and 17:00-19:00 would significantly reduce the network's connectivity. Network vulnerability is correlated with both the information entropy of nodes and the timing of link occurrences.

There is a lack of systematic research on the behavioral design of charging decision-making for Shared Autonomous Electric Vehicles (ASEVs), and the thresholds of “when to charge and where to charge” have not been clarified. Therefore, this paper investigates the optimization of charging decisions of SAEVs and the impact of different decision-making objectives to provide theoretical support and practical guidance for intelligent operation. A multi-agent simulation model (which accurately simulates complex interaction systems) is constructed to simulate the operation and charging behavior of SAEVs. Four charging decision optimization objective functions are defined, and a weighted multi-objective optimization method is adopted. A comprehensive solution process combining the multi-agent simulation model and genetic algorithm (efficiently solving complex objective optimization problems) is applied to approximate the global optimal solution among 35 scenarios through 100 iterative runs. In this paper, factors such as passenger demand (e.g., average remaining battery power, demand response time) and operator demand (e.g., empty vehicle mileage, charging cost) are considered, and the impacts of different objectives and decision variables are analyzed. The optimization results show that (1) when a single optimization objective is selected, minimizing the total charging cost effectively balances the overall fleet operation; (2) there are trade-offs between different objectives, such as the conflict between the remaining battery power and charging cost, and the balance between the demand response time and the empty vehicle mileage; and (3) in order to satisfy the operational requirements, the weight distribution, charging probability, stopping probability, and recommended battery power should be adjusted. In conclusion, this study provides optimal charging decision strategies for the intelligent operation of SAEVs in different scenarios, which can optimize target weights and charging parameters, and achieve dynamic, balanced fleet management.