The performance of granular subbase (GSB) layers in flexible pavements critically depends on their permeability and mechanical properties, which are influenced by minerology, particle shape, crushing, and fines content. Despite their importance, the interrelationship between compaction, particle breakage and permeability remains underexplored. This study examines conventional aggregates used in GSBs, focusing on shape characteristics, permeability and relative breakage through laboratory tests including gradation analysis, compaction, California bearing ratio (CBR) and permeability assessments. Compaction tests show that dense-graded mixtures achieve higher maximum dry unit weight and optimum moisture content compared to open-graded mixtures. CBR values range from 94%–129% for dense-graded mixtures and 52%–82% for open-graded mixtures, indicating better load-bearing capacity for dense-graded GSBs. Permeability tests are conducted under saturated conditions using a custom-built permeameter that closely simulates in-situ field environments. Initially, the flow regime during permeability testing is characterized using Reynolds number (Re) to distinguish between laminar and transitional flow conditions. Dense graded mixtures predominantly exhibited laminar flow regimes (Re < 1), while open graded mixtures showed higher Re values indicating the onset of inertial effects. Horizontal permeability (kH) increased with cross slope up to 4.5% for dense-graded mixtures, beyond which the flow became turbulent. Open-graded mixtures exhibited higher kH values due to larger void spaces. Vertical permeability (kV) values are generally higher than kH values, attributed to gravitational effects and the structural arrangement of the aggregate matrix. All gradations met the minimum permeability requirement of 300 m/d when the flow is vertical demonstrating the superior hydraulic performance. Dense-graded mixtures experienced lower relative breakage (0.83%–3.18%) compared to open-graded mixtures (4.30%–7.20%). The findings suggest that designing GSB mixtures close to the lower limit of gradation specifications can enhance permeability and reduce breakage. In conclusion, the research highlights the critical role of aggregate characteristics and gradation in determining the permeability and overall performance of GSBs.
This study presents the development of a game engine-based traffic microsimulation model that enables users to test and analyze unique transportation scenarios in an interactive 3D environment. The model combines Unity 3D, and SketchUp and scripts to implement built environment, pedestrian and driving behaviors within a game engine simulation framework. The game engine comprises three components: a 3D model for built environment, vehicle game object, and pedestrian game object. C# programming language is utilized to develop scripts within unity to create 3D game-objects, construct infrastructures, and train both static and dynamic game-objects to exhibit required behaviors. This study also deploys the developed model to enhance the understanding of interactions between human-driven vehicles (HVs) and autonomous vehicles (AVs) in mixed traffic scenarios. By examining human drivers’ responses in fully human-driven, fully autonomous, and mixed traffic scenarios, this study highlights the variations in behavior, interaction dynamics, and impacts on safety and performance. Results show that participants perceive higher safety and comfort when following AVs. With higher uptake of AVs, smoother and more predictable driving was observed, reducing near-miss incidents with pedestrians and enhancing overall traffic flow. The game engine developed in this study can potentially facilitate collaboration with researchers and stakeholders, providing design inputs and expert opinions to address planning and engineering challenges in AV operations.
Road cracks must be promptly and accurately detected to protect public safety and maintain traffic infrastructure. Conventional image processing and typical convolutional neural network (CNN) techniques have limited efficacy in dynamic environmental conditions, particularly when segmenting faint, fragmented, and low-contrast fractures. This paper presents a unique lightweight dual attention network (DAN-Net) that incorporates both spatial and channel attention within a U-Net-based architecture, using residual blocks to address these problems. The proposed model dynamically highlights contextually meaningful spatial regions and feature channels, thereby enhancing segmentation accuracy across complex backdrops and variable lighting conditions. We assess DAN-Net on the publicly available CRACK500 dataset and demonstrate consistent improvements over established models, including U-Net, UNet++, ResUNet, SE-Net, and SCA-Net, across critical metrics. DAN-Net attains an F1 score of 70.97, a Jaccard index (JI) of 58.75, a precision of 88.76%, and a recall of 62.96%, surpassing all comparative approaches in F1, JI, and recall metrics. Additionally, it has a reduced parameter count (9.276 million), which enhances its suitability for real-time implementation. Experimental findings validate its resilience across various contexts and its capacity to markedly diminish false positives and manual review efforts.
This study evaluates the safety performance of urban roundabouts in non-lane-based, heterogeneous traffic environments typical of low- and middle-income countries, focusing on identifying vulnerable locations within roundabout areas, classifying conflict types across different zones, and providing reproducible, easy-to-interpret visualizations for practitioners and stakeholders. High-resolution drone-recorded data from 23 Indian roundabouts were processed for trajectory extraction. Traffic conflicts were detected using time-to-collision (TTC) (≤ 1.5 seconds) and post-encroachment time (PET) (≤2.5 seconds) and classified (rear-end, lane-changing, crossing) according to conflicting angles. A conflict frequency prediction model was developed using the XGBoost (extreme gradient boosting) algorithm. Sensitivity analysis was performed using the Shannon Entropy-based method alongside SHapley Additive exPlanations (SHAP) to rank the influencing factors. Results showed significant clustering of critical conflicts (low TTC and PET) in weaving zones, entries, and exits. Spatial heatmaps and conflict-maps pinpointed high-risk hotspots and illustrated the distribution of conflict types. The conflict frequency prediction model demonstrated high predictive accuracy (root mean squared error (RMSE) = 178.15, mean absolute error (MAE) = 139.42, R2 = 0.85). Sensitivity analysis identified circulating and approaching traffic volumes, average speed, pedestrian volume, traffic composition, occupancy time, circulatory roadway width, inscribed circle diameter, entry, exit, and weaving dimensions as key factors influencing conflict frequency. The integrated framework provides data-driven guidance for targeted infrastructural improvements and traffic management strategies for roundabout safety enhancement in heterogeneous settings.
The creation and evaluation of effective traffic demand management (TDM) methods is crucial for addressing traffic congestion in metropolitan areas. One important component of these strategies is a thorough understanding of individual travel patterns, especially route choice. While existing literature acknowledges the promise of random parameters logit models for route choice analysis, this study advances the field by leveraging extensive, high-resolution trajectory data from passenger vehicles to investigate the intricate impacts of route characteristics, real-time traffic conditions, individual trip attributes, and traveler demographics on weekday morning and evening route choice behaviors. We propose an advanced route choice model that simultaneously addresses route overlap and captures random parameter heterogeneity in both means and variances, enabling a more accurate and comprehensive characterization of traveler behavior. When contrasted with more traditional models, our model shows a better statistical fit; crucially, the research reveals significant heterogeneity in route choice behavior between household-used commuting and non-commuting vehicle users, and further uncovers distinct and variable effects of influencing factors on route choices across morning and evening travel scenarios. These empirical insights offer invaluable guidance for developing personalized TDM strategies that can more effectively encourage desired changes in departure times and route choices.
The complexity and randomness of driving behavior pose significant challenges to traffic safety. Analyzing driver behavior characteristics and heterogeneity is crucial for enhancing traffic safety and designing effective interventions. This study develops a graph-based feature analysis method that uses a driving‑behavior graph to characterize intra‑driver heterogeneity across multiple segments of the same driver and inter‑driver heterogeneity across single segments drawn from different drivers. By analyzing differences in driving behavior segments, a microscopic driving behavior graph was constructed to identify abnormal driving segments. Abnormal behavior segments were extracted and classified to refine intra-driver heterogeneity labels. To explore inter-driver heterogeneity, a multi-view method was employed to analyze single behavior segments across drivers, excluding intra-driver variability. Key node data were extracted from the constructed microscopic driving behavior graph to analyze spatiotemporal dynamics. Results show that the driving behavior graph effectively reveals critical features of both intra- and inter-driver heterogeneity. An ensemble of convolutional neural networks (CNNs) and XGBoost achieves 91% accuracy for binary classification (normal vs. abnormal driving segments) and 89% for a ternary classification task (normal, conservative‑leaning abnormal, aggressive‑leaning abnormal), representing improvements of 14%–19% and 10%–16% over conventional statistical models, respectively. Drivers exhibited significant variability in behavior, with some showing greater sensitivity to longitudinal operations (e.g., speed and acceleration) and others to lateral operations (e.g., lane changes and risk indicators). These findings underscore the complexity of driver behavior and provide a scientific basis for developing targeted traffic safety interventions.
Driving on curved roads under black ice can be difficult and risky due to a lack of visibility, unanticipated loss of traction, and the curvature effect. This greatly impacts driving behavior and often requires drivers to brake and accelerate, causing extra energy consumption and skidding. To address this issue, in this paper, we propose an energy-efficient and safe (eco-safe) driving strategy (EDS) for a host car employing nonlinear model predictive control (NMPC) that incorporates information on road curvatures and black ice surface conditions. An objective function is formulated considering parameters that affect fuel economy and driving safety, and solved using a nonlinearly constrained optimization technique with a finite prediction horizon. The EDS produces optimum acceleration and velocity trajectories for the host car by utilizing its motion dynamical model, preceding vehicle states, and information on road curvatures and icy surface conditions. Microscopic traffic simulations evaluate the performance of the EDS on typical freeways. Numerical results demonstrate that the proposed NMPC-based EDS significantly mitigates the host car’s fuel usage and emissions compared to the conventional driving strategy (CDS) whilst maintaining dynamic safety gaps. Specifically, for transition curves and S-curves, the proposed EDS enhances fuel efficiency by 6.76% and 7.82%, respectively, while reducing emissions (CO2, HC, CO, and NOx) by 6.54%–21.06% and 7.29%–34.97%, respectively. Likewise, the time to collision (TTC) with the proposed EDS remains below 2.5 s for both icy roadways, assuring safety. Moreover, the proposed EDS reduces the host car’s braking force, thus enhancing skidding safety by reducing skidding risk on icy surfaces. The proposed system is deployable as an advanced driving assistance system (ADAS) for semi-autonomous mobility. Finally, we recommend a pertinent policy for the proposed system, specific to these critical road conditions.
Effective maintenance of roads is fundamental to transportation infrastructure with significant impact on safety and mobility. Notably, timely and precise pothole detection is essential for preventing road hazards and maintaining driving comfort. Traditional pothole detection methods are time- consuming and labor-intensive, prompting exploration of deep learning (DL)-based approaches for real-time and efficient pothole identification. In this study, three advanced “you only look once” (YOLO) models—YOLOv10n, YOLO11n, and YOLO11s—were evaluated on a dataset comprising of 13 767 images. Model performance was assessed using mean average precision (mAP) at an intersection over union (IoU) threshold of 0.5 (mAP@50). At this threshold, detections with an IoU ≥ 0.5 were considered true positives. In addition, precision, recall, and F1-score (the harmonic mean of precision and recall) were also reported. Experimental results indicate that YOLO11n outperforms the other models, achieving a precision rate of 100%. Further-more, YOLO11n, with its compact size of 5.6 MiB and an inference time of 1.1 ms, demonstrated an optimal balance between detection accuracy and computational efficiency, making it suitable for real-time deployment. To enhance the system’s effectiveness, global positioning system (GPS) tagging using geographic information systems (GISs) was integrated for accurate pothole mapping, and depth cameras were utilized to improve detection reliability. The model was deployed on Streamlit Cloud with an intuitive interface that allows users to upload road videos with GPS data, detect potholes, visualize results on maps and heatmaps, and download processed outputs. Overall, the findings establish YOLO11n as a highly effective solution for automated pothole detection, providing a fast, scalable, and efficient tool for proactive road maintenance on edge networks.
End-to-end autonomous driving has drawn significant attention for its ability to unify perception, decision-making, and control into a single learning framework. However, existing methods often struggle in complex and dynamic environments due to ineffective multi-modal feature fusion and inflexible decision-making strategies. To overcome these challenges, we proposed a novel Transformer-based multi-modal feature fusion framework that integrates red–green–blue (RGB) images and depth data to generate robust vehicular control commands. Our approach employed a four-stage pyramid vision transformer (PVT) backbone to extract multi-scale features and introduced a dual-attention feature fusion network to capture both intra-modal and cross-modal dependencies, yielding a more robust and context-aware environmental representation. Furthermore, we proposed a dynamic trajectory-control fusion strategy (Traj-CtrlFuser), which utilizes a learnable loss estimator to adaptively balance the outputs of trajectory and control branches based on real-time driving conditions. Extensive evaluations on the CARLA Town05 Long benchmark demonstrated that our model outperformed the current state-of-the-art multi-modal method, DriveInsight, achieving a 5.6% improvement in driving score and a 7.6% improvement in infraction score. These results underscore the framework’s potential to enhance the robustness and safety of autonomous vehicles in complex urban scenarios, providing a reliable and scalable solution for future end-to-end driving systems.
Traffic anomaly detection (TAD) is essential for highway operational safety but remains challenging due to limitations of single-modality visual methods or costly sensor reliance. Current systems exhibit high false positives and missed detections in adverse conditions. To overcome these engineering challenges, we pioneer the integration of spatiotemporal graphs with abnormal object detection (AOD), utilizing dynamic vehicle behavior (e.g., abrupt lane changes, speed variations) to enhance detection robustness in complex highway environments. We propose a comprehensive framework, introducing (i) the visual-and-spatiotemporal traffic anomaly detection (VTAD) task, which uses spatiotemporal trajectory maps as auxiliary cues to improve detection accuracy, (ii) the VTAD-highway dataset, comprising 6435 video sequences paired with spatiotemporal graphs, carefully curated from real-world highway surveillance footage, and (iii) VTAD-FL, a dual-stage network that combines a linked memory token Turing machine (LMTTM) for spatiotemporal trajectory modeling and a contrastive learning head to optimize feature discriminability. VTAD-FL integrates spatiotemporal trajectory features with visual cues through adaptive multi-scale integration, achieving superior temporal coherence and intra-class compactness. Extensive experiments show that VTAD-FL significantly outperforms existing AOD methods across all metrics, establishing a new practical benchmark for unified visual-and-spatiotemporal traffic anomaly detection in intelligent transportation systems (ITS). The dataset and code are available at https://github.com/hongkai-wei/VTAD.
Over the past decade, airlines in Asia-Oceania have faced the dual challenge of expanding operational capacity while addressing pressing environmental concerns. This study evaluates how major carriers in the region have balanced these competing priorities through eco-efficiency initiatives from 2012 to 2023. Using a Slack-based measure (SBM), we assess 13 airlines’ sustainability performance by examining their capacity utilization alongside environmental indicators including fuel consumption and CO2 emissions. The findings reveal significant disparities in how airlines manage the capacity-environment tradeoff. While leaders like Singapore Airlines and Qantas demonstrate that capacity growth can coexist with emissions reduction through strategic fleet modernization and high load factors, other carriers struggle with inefficient operations that compromise both productivity and sustainability. The COVID-19 pandemic unexpectedly served as a natural experiment, showing how temporary capacity reductions led to improved environmental metrics, though these gains proved uneven across airlines. A subsequent regression analysis identified newer fleet age and higher load factors as statistically significant drivers of eco-efficiency. The study highlights how the region’s lack of coordinated environmental policies creates an uneven playing field for sustainable capacity expansion. As demand rebounds, our analysis suggests that airlines adopting integrated approaches to network optimization, focusing on improving load factors, and clean technology investments will be best positioned to achieve both operational growth and sustainability targets. These findings provide critical insights for airlines navigating the complex interplay between market expansion pressures and decarbonization commitments in one of the world’s fastest-growing aviation markets.
High surface temperatures in asphalt pavements, primarily caused by solar radiation absorption, accelerate material degradation and increase urban heat load. Incorporating steel slag (SS), a high-thermal-conductivity by-product of the steel industry, as a substitute aggregate offers both environmental and functional benefits. Its high conductivity can promote rapid heat transfer to the subgrade and underlying pavement layers, effectively acting as a thermal sink. This study investigates the thermal performance of SS asphalt mixtures for road cooling through laboratory testing and finite element (FE) heat transfer modelling validated under various material configurations and environmental conditions. Laboratory results show that increasing SS content improves the thermal conductivity of asphalt mixtures. In particular, mixtures containing 40% slag exhibited a 30.8% increase in conductivity and lowered daytime surface temperatures by up to 6 °C at a depth of 1.5 cm. However, higher conductivity also increased nighttime heat retention by slowing the release of stored heat. Simulations indicate that this drawback can be mitigated through several strategies, including placing the slag layer in lower pavement courses, using a lower-conductivity material at the surface, and providing a thermal buffer. Furthermore, it can also be addressed by integrating SS asphalt with other cool pavement technologies, such as heat-reflective coatings (HRCs) and permeable pavements (PPs). Reflective coatings with SS asphalt yielded daytime surface temperature reductions exceeding 20 °C compared to conventional pavements, while also limiting nighttime heat retention by enhancing radiative cooling. Permeable surfaces further moderated both daytime and nighttime temperatures by utilising moisture for evaporative cooling and buffering subsurface heat release. Overall, this study demonstrates the potential of SS asphalt mixtures to enhance road cooling, with performance strongly influenced by subsurface heat storage and surface emissivity and further improved when integrated with other cool pavement technologies.
Deciding whether to implement cross-line operation or maintain independent operation between two suburban railway lines with compatible systems requires quantitative decision-making support. This study focuses on two interconnected suburban railway lines and compares the optimal train operation schemes under the two service modes, taking into account total passenger demand, the weighting of passenger benefits, the proportion of cross-line passengers, and the share of passenger flows at low-level stations. The train operation scheme optimization model adopts average passenger-perceived travel time and train load factor as objective functions. Passenger route choice and flow assignment are performed using a C-logit model combined with the method of successive averages, while the non-dominated sorting genetic algorithm II (NSGA-II) is employed to identify Pareto-optimal solutions. Taking the cross-line operation project in Guangzhou, China, as a case study, the results indicate that independent operation demonstrates advantages under scenarios such as large total passenger flow, high cross-line passenger proportion, or significantly low-level station passenger flow ratio. Compared with independent operation, cross-line operation can improve passenger travel benefits by reducing perceived travel time by 3.51 min and transfer frequency by 0.21, but this comes at the expense of increased operational costs for railway operators, including a 3.25% reduction in train load factor and an additional 138.28 km of operating distance. The cross-line operation mode is most advantageous when the cross-line passenger rate remains below 70% and total passenger volume is approximately 70–80% of the line’s design capacity.
Currently, there exists an imbalance in the utilization of the mixed road network formed by urban expressways and regular surface roads. This phenomenon is not only related to road attributes but also influenced by drivers’ route choice preferences. To improve the balance of utilization in the mixed road network, understanding the road route choice behavior between urban expressways and regular surface roads is crucial. Introducing travel preferences into the road route choice model for mixed road networks facilitates more accurate route choice prediction and guidance.To incorporate the specific route preferences of different drivers, this study establishes a route choice model framework for a mixed network. Based on survey data of Shanghai, China, it first employs the CatBoost classification prediction algorithm to predict drivers’ road type preferences based on their preference classification. Subsequently, the road type preference is set as a variable, combined with route attribute variables, to model route choice behavior in the mixed road network. The improved random regret minimization model (RRM-P), which incorporates the scale effect and road type preferences, demonstrates the best fitting performance compared to the other models, achieving a McFadden’s R2 value exceeding 0.2 and an overall hit rate exceeding 0.87. This model can be used for traffic prediction in mixed road networks and provides a basis for macroscopic-level route guidance.
Traffic safety education has proved to be an effective means to reduce the number of traffic violations; however, few studies have attempted to quantify how its safety effect sustains over time, that is, the decay mechanism of the safety effect. In this paper, we design an “overlap number of violations before and after traffic safety education” as an evaluation indicator for the impact of traffic safety education. With this, we built a decay model for the effects of traffic safety education using non-linear differential equations, showing it to follow a logistic curve. We identified two critical time points from the decay model: the “optimal safety education frequency point” and the “minimum safety education frequency point,” offering a diversified approach to forming traffic safety education policies. Finally, we applied our model to long-term tracking data from 30 088 traffic violators in Hangzhou, China. Findings suggest that the optimal and minimum safety education frequency points for violations are the 10th and 23rd weeks, respectively. The immediate effect of an education program is higher for reducing motorized vehicle-related violations than for non-motorized vehicle-related violations; however, the long-term impact of education is more profound for non-motorized vehicle-related violations than for motorized vehicles. This research reveals the decay mechanism of traffic safety education effects, which has significant implications for developing effective programs and policies.
Implementing collaborative governance in the road transportation of dangerous goods encounters severe challenges due to the lack of trust among stakeholders. This study proposes a collaborative governance framework for the road transportation of dangerous goods based on blockchain technology. A tripartite evolutionary game model was developed to investigate conflicts and cooperative dynamics among three key stakeholders, including carriers, transportation management authorities (TMA), and emergency management authorities (EMA). The stability analysis and numerical simulations were used to identify stability conditions of stakeholders’ strategies. The influences of blockchain platform maturity, illegal excess profits, and regulatory expenses were also analyzed. This study proposes three innovative management strategies: (1) maturing blockchain platforms with a standardized evaluation system to enhance stakeholders’ collaboration and data credibility; (2) curbing illegal dangerous goods transportation via legislative coordination and data-sharing; (3) automating regulatory processes with smart contract technology for cost reduction and efficiency improvement.The proposed model shows the potential of introducing the blockchain platform for collaborative governance for road transportation of dangerous goods.
Deep learning methods have been widely applied in urban traffic flow prediction and have achieved promising results. However, these methods rely on large amounts of training data. In reality, due to the scarcity of traffic data in some cities, deep learning methods struggle to achieve optimal predictive performance. Recently, transfer learning has been introduced into traffic flow prediction to alleviate data scarcity through knowledge transfer. Nevertheless, most existing studies primarily focus on modeling spatial adjacency within traffic networks, overlooking the inherently multidimensional spatial structures of urban transportation and the distribution discrepancies between cities. In this paper, we propose a multi-view spatio-temporal graph convolutional network (MSTGCN) with domain-adversarial learning to address the challenges of traffic flow prediction under data-scarce conditions. Specifically, MSTGCN constructs a multi-view spatio-temporal graph based on the physical distance, functional similarity, and dynamic correlation of traffic nodes. On this basis, a parameter-sharing spatio-temporal graph convolution is designed to extract spatio-temporal features of traffic flow. Additionally, a domain-adversarial learning approach is integrated to extract domain-invariant spatio-temporal features and reduce distribution differences. Extensive experiments on five real-world traffic datasets demonstrate that MSTGCN significantly outperforms state-of-the-art transfer learning methods, achieving up to 9.7% lower mean absolute error on average.
In recent years, deep reinforcement learning (DRL) has been widely applied to urban traffic signal control in mixed traffic environments where human-driven vehicles (HDVs) and connected and automated vehicles (CAVs) coexist. Most existing studies optimize signal timing based on mixed traffic flow characteristics but overlook CAVs’ control strategies. To achieve cooperative control, some approaches design both traffic signals and CAVs as agents within a multi-agent reinforcement learning framework. However, such methods often suffer from high model complexity and computational costs, making real-world deployment challenging. To address these issues, we propose a novel unified control framework, Uni-Light, which introduces an invalid-action masking mechanism within a DRL-based multidimensional action space. This design constrains frequent signal phase switching while filtering out infeasible vehicle actions under red phases, enabling Uni-Light to generate both low-frequency signal control and high-frequency CAV speed control with only the traffic signal as an agent. The CAV control is applied exclusively to the leading CAV in each approach lane, resulting in a simplified yet efficient cooperative strategy. Moreover, a grid-based microscopic state representation is fused with macroscopic traffic features to enhance the agent’s perception capability. Simulation results demonstrate that Uni-Light significantly improves network throughput and reduces energy consumption compared with existing methods, while maintaining stability and ensuring safe vehicle operations under varying CAV penetration rates. This study provides a high-performance and novel paradigm for cooperative control between traffic signals and CAVs, further facilitating the deployment of cooperative control strategies in mixed traffic and enhancing the operational safety of autonomous vehicles.
In rapidly urbanising regions, public transport systems play a vital role in ensuring mobility, but they may also cause stress for users due to adverse conditions along the journey. Despite a growing body of research, limited attention has been given to the heterogeneity of public bus commuter experiences and how these vary across different transit environments. Understanding the diversity of commuter stress during transit is critical for improving rider experience and advancing equitable transport planning. This paper examines the variability in transport stress patterns among bus commuters and explores their associations with socio-demographic characteristics and environmental stressors using latent class analysis (LCA) on survey data collected from over 5000 commuters in Hong Kong, China. A four-class solution was identified, revealing distinct commuter stress experiences across bus stops, interchanges, and onboard environments. Findings show that the most stressed class accounts for approximately 21% of all commuters. Members of this class reported frequent and significant stress across multiple journey stages. Key stressors include congested, undersized waiting areas, exposure to extreme weather, unclear queueing systems, and the lack of real-time information at bus stops. Additional stressors include washroom facilities, seating at interchanges and onboard, conditions within bus compartments, uncomfortable bus temperatures, and inadequate travel information. In contrast, the smallest class (roughly 15%) comprises commuters with largely positive, stress-free experiences. The study concludes with practical policy recommendations, particularly targeting younger commuters and individuals in specific occupational sectors who are more vulnerable to elevated transport-related stress.
Individual weekly activity patterns are essential for enhancing activity-based mobility models, yet existing methods mainly focus on single-day behaviors, while overlooking temporal dependencies across days, and struggle to balance predictive accuracy with interpretability. To address the gap, this study proposes AP-WASG, a three-layer, activity-pattern-based weekly activity sequence generation framework grounded in natural language processing. The proposed method first casts an individual weekly timeline as text, leveraging latent Dirichlet allocation (LDA) to distill complex sequences into a concise topic space and then grouping them into seven activity patterns. Secondly, an XGBoost classifier forges a clear link between these patterns and individuals’ socio-demographics and built environment, making it possible to predict one’s dominant activity pattern in new contexts. A Transformer-based architecture is then proposed to generate individual weekly activity sequences with the socio-demographics-enhanced texts, achieving cosine similarity scores of 0.74–0.92 with real-world data, representing 12–33% improvement over pattern-unconstrained approaches. The hybrid framework addresses traditional limitations in big data applications while maintaining interpretability through pattern constraints. The results provide practical support for personalized urban planning, transportation management, and policy analysis, including cold-start and scenario applications in unseen neighborhoods and planning years.
