Accurately predicting traffic accident risk is crucial for preventing traffic accidents and enhancing road traffic safety. Extensive approaches have been proposed for this task. However, existing methods face three main challenges: (i) difficulty in capturing complex spatial correlations, especially semantic dependencies; (ii) neglect of differences between daily and weekly temporal patterns; and (iii) equal treatment of heterogeneous semantic and geographical spatio-temporal features, which hampers predictive performance. To address these limitations, a multi-graph spatio-temporal network for predicting traffic accident risk is proposed, referred to as MG-STNET. Specifically, the Multi-Graph Spatial Network (MGSNet) and the Geographical Spatial Network (GeoSNet) are first introduced to capture semantic spatial dependencies across all regions and geographical correlations among adjacent regions, respectively. Temporal Blocks is then employed to model daily and weekly accident patterns separately. Finally, an adaptive channel fusion gate is integrated to automatically balance heterogeneous semantic and geographical spatio-temporal features. Experiments conducted on the NYC and Chicago datasets evaluate model performance under two aspects: all-day and high-frequency traffic accident periods. The results demonstrate that MG-STNET consistently outperforms baseline methods and highlight the importance of each model component.

Physics-informed deep learning (PIDL)-based models have recently garnered remarkable success in Traffic State Estimation (TSE). However, the prior knowledge used to guide regularization training in current mainstream architectures is based on deterministic physical models. The drawback is that a solely deterministic model fails to capture the universally observed traffic flow dynamic scattering effect. Considering the existence of more realistic stochastic physical models that can reproduce the relationship between speed and flow, they can provide better bounds for neural network models with uncertainty. Therefore, this study, for the first time, incorporates stochastic physics information to improve the PIDL architecture and propose stochastic physics-informed deep learning (SPIDL) for traffic state estimation. The idea behind such SPIDL is simple and is based on the fact that a stochastic fundamental diagram provides the entire range of possible speeds for any given density with associated probabilities. Specifically, we select percentile-based fundamental diagram and distribution-based fundamental diagram as stochastic physics knowledge and design corresponding physics-uninformed neural networks for effective fusion, thereby realizing two specific SPIDL models, namely α[jls-end-space/]-SPIDL and B[jls-end-space/]-SPIDL. The main contribution of SPIDL lies in addressing the "overly centralized guidance" caused by the one-to-one speed-density relationship in deterministic models during neural network training, enabling the network to digest more reliable knowledge-based constraints. Experiments on real-world datasets indicate that proposed SPIDL models achieve accurate traffic state estimation in sparse data scenarios. More importantly, as expected, SPIDL models reproduce well the scattering effect of field observations, demonstrating the effectiveness of fusing stochastic physics model knowledge with deep learning frameworks. © 2025 The Author(s). Published by Elsevier Ltd. This is an open access article under the CC BY license. http://creativecommons.org/licenses/by/4.0/

Traffic state estimation (TSE) fundamentally involves solving high-dimensional spatiotemporal partial differential equations (PDEs) governing traffic flow dynamics from limited, noisy measurements. While Physics-Informed Neural Networks (PINNs) enforce PDE constraints point-wise, this paper adopts a physics-informed deep operator network (PI-DeepONet) framework that reformulates TSE as an operator learning problem. Our approach trains a parameterized neural operator that maps sparse input data to the full spatiotemporal traffic state field, governed by the traffic flow conservation law. Crucially, unlike PINNs that enforce PDE constraints point-wise, PI-DeepONet integrates traffic flow conservation model and the fundamental diagram directly into the operator learning process, ensuring physical consistency while capturing congestion propagation, spatial correlations, and temporal evolution. Experiments on the Next Generation Simulation (NGSIM) dataset demonstrate superior performance compared to state-of-the-art baselines. Further analysis reveals insights into optimal function generation strategies and branch network complexity. Additionally, the impact of input function generation methods and the number of functions on model performance is explored, highlighting the robustness and efficacy of proposed framework.

In urban intelligent transportation systems, the real-time acquisition of network-wide traffic states is constrained by limited sensor density and high deployment costs. To address this challenge, this paper proposes a learnable Detection Point Selection Module (DPSM), which adaptively determines the most informative observation points through an end-to-end attention mechanism to support full-map traffic state estimation. Distinct from conventional fixed deployment strategies, DPSM provides an adaptive detector configuration that, under the same number of loop sensors, achieves significantly higher estimation accuracy by intelligently optimizing their placement. Specifically, the module takes normalized spatial and temporal information as input and generates an attention-based distribution to identify critical traffic flow readings, which are subsequently fed into various backbone prediction models, including fully connected networks, convolutional neural networks, and long short-term memory networks. Experiments on the real-world NGSIM-US101 dataset demonstrate that three variants-DPSM-NN, DPSM-CNN, and DPSM-LSTM-consistently outperform their corresponding baselines, with notable robustness under sparse observation scenarios. These results highlight the advantage of adaptive detector placement in maximizing the utility of limited sensors, effectively mitigating information loss from sparse deployments and offering a cost-efficient, scalable solution for urban traffic monitoring and control.

Deep learning-based traffic flow prediction has made remarkable progress, including many efforts utilising decomposition for distinct components prediction. However, the decomposition-based strategy ignores the capture of non-decomposed global information. Besides, relying solely on spatiotemporal attention mechanisms for feature extraction at different scales is not effective enough. Therefore, this paper presents a multi-scale feature-aware spatiotemporal graph convolutional network (MsFaGCN) for highway traffic flow prediction. Each module within MsFaGCN is designed to target specific features of traffic flow at different scale perspectives, tailored to leverage the unique strengths of each module. The MsFaGCN is designed based on an encoder-decoder architecture, where the encoder focuses on the feature scales of decomposed seasonal and trend components, while the decoder focuses on the non-decomposed global feature scale. The experiments are conducted using real-world highway data in Ningde City, Fujian Province, China. The results demonstrate that MsFaGCN delivers accurate predictions and outperforms baseline models.

The rapid expansion of ride-hailing services has profoundly impacted urban mobility and residents' travel behavior. This study aims to precisely identify and quantify how the built environment and socioeconomic factors influence spatial variations in ride-hailing demand using multi-source data from Haikou, China. A multi-scale geographically weighted regression (MGWR) model is employed to address spatial scale heterogeneity. To more accurately capture environmental features around sampling points, the DeepLabv3+ model is used to segment street-level imagery, with extracted visual indicators integrated into the regression analysis. By combining multi-scale geospatial data and computer vision techniques, the study provides a refined understanding of the spatial dynamics between ride-hailing demand and urban form. The results indicate notable spatiotemporal imbalances in demand, with varying patterns across workdays and holidays. Key factors, such as distance to the city center, bus stop density, and street-level features like greenery and sidewalk proportions, exert significant but spatially varied impacts on demand. These findings offer actionable insights for urban transportation planning and the design of more adaptive mobility strategies in contemporary cities.

Accurate traffic prediction is essential for intelligent transportation systems, urban mobility management, and traffic optimization. However, existing deep learning approaches often struggle to jointly capture complex spatial dependencies and temporal dynamics, and they are prone to overfitting when modeling large-scale traffic networks. To address these challenges, we propose the GSF-LLM (graph-enhanced spatio-temporal fusion-based large language model), a novel framework that integrates large language models (LLMs) with graph-based spatio-temporal learning. GSF-LLM employs a spatio-temporal fusion module to jointly encode spatial and temporal correlations, combined with a partially frozen graph attention (PFGA) mechanism to model topological dependencies while mitigating overfitting. Furthermore, a low-rank adaptation (LoRA) strategy is adopted to fine-tune a subset of LLM parameters, improving training efficiency and generalization. Experiments on multiple real-world traffic datasets demonstrate that GSF-LLM consistently outperforms state-of-the-art baselines, showing strong potential for extension to related tasks such as data imputation, trajectory generation, and anomaly detection.

At present, traffic state prediction primarily relies on purely data-driven methods, ignoring the incorporation of physical constraints within the field of traffic flow. Taking this as a starting point, this paper endeavors to embed the physical mechanism of the traffic flow fundamental graph into the deep learning model, and proposes a physics-integrated spatiotemporal graph neural network with fundamental diagram learner (PI-STGnet) for highway traffic flow prediction. The PI-STGnet mainly comprises a semantic enhancement module (SE), a spatiotemporal correlation extraction module (ST-Block), and a fundamental diagram learner module (FD-learner). These modules are strategically designed to sequentially enhance the contextual semantic relationships of the traffic flow and speed inputs, extract intricate spatiotemporal correlations of traffic states, and adaptively learn the dynamic evolution of the physical relationships between traffic flow and speed. The real-world dataset employed to assess the performance of the PI-STGnet is sourced from the monitoring data of highway gantry sensors located in Ningde City, Fujian Province, China. The experimental results demonstrate that the proposed PI-STGnet has the advantages to extract spatiotemporal features and achieve prediction, surpassing the accuracy of cutting-edge baseline models. In summary, as an exploratory work, this paper provides a novel approach that driven by the fusion of deep learning and physical mechanisms for traffic flow prediction to a certain extent.

Reliable and accurate traffic flow prediction is crucial for the construction and operation of smart highways, supporting scientific traffic management and planning. However, accurately predicting spatiotemporal traffic flow in non-stationary and unprecedented traffic patterns scenarios, such as holidays and adverse weather conditions, remains a challenging task. Considering that (1) Koopman theory effectively captures the underlying time-variant dynamics of the non-stationary temporal sequence (2) Graph convolutional network (GCN) effectively extracts complex spatial dependencies, combining the strengths of both is a promising solution. Therefore, this paper proposes a spatiotemporal prediction network that integrates Koopman theory and GCN, named KoopGCN, for predicting non-stationary and inexperienced highway traffic flow. KoopGCN decomposes the input into time-invariant and time-variant components based on Fast Fourier Transform. The dual engine block consisting of KoopGCN InvarEngine and KoopGCN VarEngine is designed to predict two types of components separately. And the dual engine block also passes the residual to the next block for modeling. The experiment is conducted on real monitored highway data in Ningde City, Fujian Province, China. The results indicate that even if there is a significant distribution difference between the training and testing sets, KoopGCN can achieve accurate prediction, significantly outperforms state-of-the-art baselines. © 2024 The Author(s)

Shared autonomous electric vehicles (SAEVs) are predicted to become a significant solution to reduce global emissions and energy consumption resulting from urban transportation. The centralized operation of SAEVs not only allows large-scale travel demand response but also can provide essential ancillary services to the smart grid through the concept of vehicle-to-grid (V2G). With V2G technology, unused electric vehicles can work as a distributed energy storage facility for the electricity grid to smoothen the intermittent demand. Designing and operating a V2G-enabled SAEV system is challenging. This problem involves complicated planning and operational decisions, as well as time-varying electric tariffs. In this work, a flow-based Integer Linear Programming (ILP) model is formulated for determining the optimal configurations (charging infrastructure and fleet size) and daily operation strategies (serving passengers, relocation, charging/discharging). The developed mathematical model allows for maximizing the total profit, comprising investment cost, revenue from serving passengers, and V2G profit through charging/discharging schedules. A two-stage Benders decomposition-based algorithm is proposed to address the sophisticated ILP problem. Via testing instances in the Manhattan network based on real-world and synthetic data, we have demonstrated the feasibility of our approaches and studied the benefits of integrating V2G in the SAEV system. © 2024 Elsevier Ltd