IET Intelligent Transport Systems最新文献

High-speed railway operations rely on various crew members. Among them, onboard mechanics are responsible for monitoring train systems and handling in-transit faults, bearing long working hours and critical safety responsibilities. However, they have received limited attention in existing crew rostering research, particularly those residing outside the base depot city. Additionally, the crew rostering plan often requires adjustments during execution, due to foreseeable disruptions such as leave requests, training courses, and additional tasks. Yet, these practical issues are often ignored, with adjustments mainly made manually and without a systematic correction approach. To address these gaps, this paper first proposes a hybrid service pattern and a mathematical model to design more reasonable crew routes and provide more convenient task arrangements for onboard mechanics. Following this, a rolling optimisation model is introduced to address foreseeable disruptions during the execution of the rostering plan, aiming to minimise plan modifications while maintaining balanced workloads. The relationship between the two models is sequential, with the crew route information obtained from the first model serving as partial input for the second model. A real-world case study based on data from the Qingdao EMU Depot in China is conducted to validate the effectiveness of the proposed models, with Gurobi used for solving. The computational results show that the crew routes are well assessed and balanced in terms of both operational costs and work convenience. Furthermore, the rolling optimisation model achieves a significantly lower modification-to-disruption ratio than the manual approach, ensuring greater robustness against disruptions. It further improves workload balance, reduces commuting costs, and prevents overwork, offering a practical and efficient solution for real-world planning.

Merging behaviours in work zones pose significant traffic management challenges due to lane closures that create bottlenecks and increase accident risk. In mixed traffic environments involving emergency vehicles (EVs), connected automated vehicles (CAVs), and human-driven vehicles, addressing these challenges becomes more complex. EVs, in particular, require priority movement to minimize response times, but their movement is often hindered by traditional traffic management strategies. While previous merging strategies have shown promising results in managing traffic flow, they fail to account for the unique challenges posed by different vehicle types in work zone environments. In this paper, we propose a new reward weight adjustment method for multi-agent proximal policy optimization based CAVs in work zone for emergency vehicles (MAPPO-WEV). The proposed method improves EV's response times without compromising travel time of other vehicles. MAPPO-WEV introduces a dynamic reward weighting mechanism that adjusts the importance weight of speed, headway, and merging behaviour based on the type and number of vehicles. This approach allows EVs to travel more freely while maintaining safety in mixed traffic conditions. The simulation results of MAPPO-WEV show significant improvements in both travel times and waiting times of EVs by 25% and 33% respectively compared to the Baseline method.

The safety decisions of autonomous driving systems rely on the accurate understanding of 3D scenes, and the existing 3D occupancy prediction (OCC) models are difficult to meet the requirements of in-vehicle deployment due to their high computational complexity and a large number of parameters. Traditional methods (e.g., OccWorld, FlashOcc) rely on full-precision floating-point operations and dense 3D convolution, resulting in hundreds of millions of model parameters. In this paper, we propose a lightweight two-branch binarization network, TBBOcc, to break through the bottleneck of ‘efficiency-accuracy’ trade-off through multi-technology co-optimization. First, we design two-branch binarized feature extraction, using channel compression and hyperbolic tangent relaxation activation function to alleviate the problem of vanishing binarized gradient, which reduces the computation amount while retaining the key geometrical information; second, we improve the EfficientViM module by integrating state space modeling and a two-dimensional normalization strategy, which enhances the ability of global temporal feature modeling; and lastly, we introduce a dynamic temporal fusion mechanism, combining binocular depth estimation with deformable BEV pooling to capture the spatio-temporal evolution laws. Experiments show that TBBOcc achieves 39.1% mean intersection over union (mIoU) on the Occ3D-nuScenes validation set with 32.8 M parameter counts and 164.8 G FLOPs, which reduces the amount of parameters by 26.6%, computation by 33.7%, and improves the accuracy by 3.3% compared with the baseline model FlashOcc. Especially, it performs well in dynamic obstacles (e.g., pedestrians, traffic cones) and complex scenes. In this paper, binarization computation is introduced into the 3D OCC task for the first time, which provides an efficient and reliable technical path for real-time environment sensing for autonomous driving.

To enhance the robustness and interpretability of road-accident prediction under sparse, noisy, and dynamically evolving conditions, this study proposes a Multimodal Grey-Markov and Adversarial Meta-Learning (MGMC-AML) framework. The model integrates grey relational analysis for adaptive modality weighting, dynamic Markov state modelling for temporal-spatial transition learning, and adversarial meta-optimisation for rapid recovery from perturbations. A reinforcement-enhanced dynamic partitioning mechanism is introduced to maintain structural consistency of road-network topology during fluctuating traffic states. Extensive experiments on multi-source traffic and weather datasets demonstrate that the proposed framework achieves superior predictive performance, with a Mean Absolute Error (MAE) of 2.34 incidents per day, a Segmentation Consistency Index (SCI) of 0.85, and a Modular Q value of 0.78, while maintaining an attack-recovery time of 162 s. These results surpass state-of-the-art baselines, including LSTM, GNN, and Transformer models, confirming both its precision and resilience under adversarial or uncertain sensing conditions. The proposed approach offers a unified and reproducible framework that bridges uncertainty modelling, adaptive fusion, and robust optimisation, providing a theoretical and practical foundation for resilient traffic-risk forecasting in intelligent-transportation systems.

Given that constant-parameter model predictive control (MPC)-based automatic emergency braking (AEB) systems are unable to simultaneously balance comfort, safety, and computational efficiency, this paper proposes an adaptive optimization strategy for MPC that adjusts weights and sampling times to address this issue. First, a three-level warning strategy based on a safety distance model is introduced. This strategy assesses the risk level during vehicle operation using an emergency coefficient and adaptively adjusts the sampling time. By analysing the correlation between relative distance, speed, and injury risk, weight adjustment rules based on injury risk are determined. Second, a fuzzy regulator is developed with the emergency coefficient, injury risk, and relative distance as inputs is developed to enable dynamic adjustment of weights and sampling time in response to operational conditions. Finally, an AEB control strategy is designed based on hierarchical control principles: the upper layer uses MPC to achieve multi-objective optimization, while the lower layer employs PID correction to track the desired acceleration. In the test scenarios, joint simulation experiments were conducted using CarSim and MATLAB/Simulink, and the results under four scenarios and operating conditions were analysed and compared. The results show that the proposed control strategy enhances comfort while ensuring AEB safety control, reducing the average braking distance deviation by 11.80% and the average acceleration deviation by 48.91%. These improvements are significant for enhancing AEB performance without hardware modifications.

Traditional methods of vehicle trajectory reconstruction heavily rely on the data of cross-sectional traffic detectors, but the effectiveness of existing methods is limited by insufficient information of the cross-sectional data. In response to this, this study proposes a novel trajectory reconstruction method based on long-range traffic detectors. It builds upon Newell's car-following model and its derived inverse following model. By taking driver heterogeneity into account through individualised calibration of the key parameter, namely the spatial shift, which is optimised by the whale optimisation algorithm, the accuracy of trajectory reconstruction is enhanced. Furthermore, to connect trajectories of the same vehicle in the same area that are reconstructed by adjacent long-range detectors, a particle filter-based trajectory fusion method is developed. It can fuse overlapped reconstructed trajectories and smoothly connect sectional trajectories into a complete and seamlessly connected one. The performance of the trajectory reconstruction method is evaluated on the NGSIM I-80 dataset, while the trajectory fusion method was tested on both the I-80 and the TRJD TS datasets. Results show that the reconstruction method generates complete vehicle trajectories across various traffic flow conditions, achieving an average of 28.65% reduction in mean absolute error compared to methods that do not account for driver heterogeneity. The mean absolute error of the fused trajectories was reduced by 49.23% and 59.69% on average for two datasets, respectively, compared to reconstructed trajectories using a single detector. The trajectory reconstruction accuracy of the proposed method also outperforms that of a deep convolutional neural network and an improved adaptive smoothing method.

In this study, we introduced a multi-task deep learning framework that concurrently forecasts traffic accident risk and severity by integrating convolutional neural networks (CNNs), bidirectional long short-term memory (BiLSTM) units, and a self-attention mechanism. Unlike conventional single-task approaches, our model leverages shared spatiotemporal representations to capture complex patterns in traffic data, thereby enhancing both predictive accuracy and generalizability. Evaluations on large-scale datasets from New York City and Chicago demonstrate that our approach achieves high accuracy (up to 92% for accident risk and 89% for severity) and remains robust across diverse urban contexts. Moreover, an enhanced SHAP-based interpretability module provides granular insights into the influence of both observable and latent factors, such as driver behaviour or road surface conditions, on prediction outcomes. The self-attention mechanism further mitigates unobserved heterogeneity by highlighting critical time steps and feature interactions. With competitive real-time performance and throughput, our framework offers a practical solution for dynamic traffic safety applications. Future work will focus on extending evaluations to broader urban settings and integrating latent variable models to better quantify unobserved influences, ultimately advancing the development of safer, more efficient transportation systems.

Accurate modelling of acceleration and deceleration decisions is vital for replicating car-following behaviour, as these govern longitudinal control, traffic stability, and safety. This study introduces ArrowNet81, a 200-layer novel convolutional neural network (CNN) architecture designed to model the asymmetric nature of these decisions using multivariate time-series data. The architecture leverages modular Ribs and Linkers to enhance depth while minimizing complexity. A perceived-risk variable derived from nominal time-to-collision (NomTTC) addresses the effect of vehicle heterogeneity in car following behaviour. Trajectories extracted from UAV video footage support neighbourhood vehicle identification via the queens move system (QMS), and a novel generalized data arrangement technique ensures compatibility with deep learning (DL) inputs. To prove its generalizability and superiority, the proposed architecture is trained, tested, and validated in three diverse traffic conditions against different statistical, machine learning (ML), and DL techniques. And the developed models using ArrowNet81 architecture outperform all of them. Future research may adapt ArrowNet81 for classification problems such as mode choice, accident severity, or spatial risk mapping, and integrate it into connected and autonomous vehicles (CAVs) to emulate human-like behaviours. The rib-based architecture also permits the development of lighter variants, facilitating real-time deployment in resource-constrained environments without compromising predictive performance.Accurate modelling of acceleration and deceleration decisions is vital for replicating car-following behaviour. ArrowNet81, a 200-layer novel CNN architecture designed to model the asymmetric nature of these decisions using multivariate time-series data. The rib-based architecture permits the development of lighter variants, facilitating real-time deployment in resource-constrained environments without compromising predictive performance.

Multi-bottleneck highway segments present significant challenges in traffic management due to the propagation of congestion waves between closely spaced bottlenecks. This study proposes a linear model predictive control (MPC)-based ramp metering strategy designed specifically for continuous multi-bottleneck corridors. The approach incorporates a macroscopic linear traffic flow model that discretizes the roadway into interconnected cells, allowing real-time prediction of traffic states on both mainline and ramp segments. The control problem is formulated as a constrained quadratic programming task aimed at minimizing vehicle accumulation and enhancing overall throughput. A key innovation of this strategy lies in its predictive, multi-input, multi-output architecture, which enables proactive, corridor-wide coordination of ramp inflows based on anticipated traffic conditions and inter-bottleneck interactions. To ensure real-time computational feasibility, a linearized cell transmission model is used and efficiently solved via the CPLEX optimizer. Simulation experiments demonstrate the effectiveness of the proposed method, with reductions in total travel time of 23.7%, 11.6% and 2.4% and corresponding reductions in total delay of 74.6%, 54.1% and 8.9%, compared to no-control, PI-ALINEA and regional MPC ramp metering strategies, respectively. These results highlight the strategy's superiority in improving system-wide traffic efficiency under complex congestion scenarios.