Communications in Transportation Research最新文献_第6页

MetaSSC: Enhancing 3D semantic scene completion for autonomous driving through meta-learning and long-sequence modeling MetaSSC：通过元学习和长序列建模增强自动驾驶的3D语义场景完成

IF 12.5 Q1 TRANSPORTATION

Communications in Transportation Research

Pub Date : 2025-05-28 DOI: 10.1016/j.commtr.2025.100184

Yansong Qu , Zixuan Xu , Zilin Huang , Zihao Sheng , Sikai Chen , Tiantian Chen

Semantic scene completion (SSC) plays a pivotal role in achieving comprehensive perceptions of autonomous driving systems. However, existing methods often neglect the high deployment costs of SSC in real-world applications, and traditional architectures such as three-dimensional (3D) convolutional neural networks (3D CNNs) and self-attention mechanisms struggle to efficiently capture long-range dependencies within 3D voxel grids, limiting their effectiveness. To address these challenges, we propose MetaSSC, a novel meta-learning-based framework for SSC that leverages deformable convolution, large-kernel attention, and the Mamba (D-LKA-M) model. Our approach begins with a voxel-based semantic segmentation (SS) pretraining task, which is designed to explore the semantics and geometry of incomplete regions while acquiring transferable meta-knowledge. Using simulated cooperative perception datasets, we supervise the training of a single vehicle's perception via the aggregated sensor data from multiple nearby connected autonomous vehicles (CAVs), generating richer and more comprehensive labels. This meta-knowledge is then adapted to the target domain through a dual-phase training strategy—without adding extra model parameters—ensuring efficient deployment. To further enhance the model's ability to capture long-sequence relationships in 3D voxel grids, we integrate Mamba blocks with deformable convolution and large-kernel attention into the backbone network. Extensive experiments show that MetaSSC achieves state-of-the-art performance, surpassing competing models by a significant margin while also reducing deployment costs.

语义场景完成（SSC）在实现自动驾驶系统的全面感知方面发挥着关键作用。然而，现有的方法往往忽略了SSC在实际应用中的高部署成本，而传统的架构，如三维卷积神经网络（3D cnn）和自关注机制，难以有效地捕获三维体素网格内的远程依赖关系，限制了它们的有效性。为了应对这些挑战，我们提出了MetaSSC，这是一种新的基于元学习的SSC框架，它利用了可变形卷积、大核注意和Mamba （D-LKA-M）模型。我们的方法从基于体素的语义分割（SS）预训练任务开始，该任务旨在探索不完整区域的语义和几何，同时获取可转移的元知识。使用模拟的协同感知数据集，我们通过来自多个附近连接的自动驾驶汽车（cav）的聚合传感器数据来监督单个车辆的感知训练，生成更丰富和更全面的标签。然后，通过双阶段训练策略（不添加额外的模型参数）使元知识适应目标领域，从而确保有效部署。为了进一步增强模型在3D体素网格中捕获长序列关系的能力，我们将具有可变形卷积和大核关注的Mamba块集成到骨干网络中。大量的实验表明，MetaSSC达到了最先进的性能，大大超过了竞争模型，同时也降低了部署成本。

{"title":"MetaSSC: Enhancing 3D semantic scene completion for autonomous driving through meta-learning and long-sequence modeling","authors":"Yansong Qu , Zixuan Xu , Zilin Huang , Zihao Sheng , Sikai Chen , Tiantian Chen","doi":"10.1016/j.commtr.2025.100184","DOIUrl":"10.1016/j.commtr.2025.100184","url":null,"abstract":"<div><div>Semantic scene completion (SSC) plays a pivotal role in achieving comprehensive perceptions of autonomous driving systems. However, existing methods often neglect the high deployment costs of SSC in real-world applications, and traditional architectures such as three-dimensional (3D) convolutional neural networks (3D CNNs) and self-attention mechanisms struggle to efficiently capture long-range dependencies within 3D voxel grids, limiting their effectiveness. To address these challenges, we propose MetaSSC, a novel meta-learning-based framework for SSC that leverages deformable convolution, large-kernel attention, and the Mamba (D-LKA-M) model. Our approach begins with a voxel-based semantic segmentation (SS) pretraining task, which is designed to explore the semantics and geometry of incomplete regions while acquiring transferable meta-knowledge. Using simulated cooperative perception datasets, we supervise the training of a single vehicle's perception via the aggregated sensor data from multiple nearby connected autonomous vehicles (CAVs), generating richer and more comprehensive labels. This meta-knowledge is then adapted to the target domain through a dual-phase training strategy—without adding extra model parameters—ensuring efficient deployment. To further enhance the model's ability to capture long-sequence relationships in 3D voxel grids, we integrate Mamba blocks with deformable convolution and large-kernel attention into the backbone network. Extensive experiments show that MetaSSC achieves state-of-the-art performance, surpassing competing models by a significant margin while also reducing deployment costs.</div></div>","PeriodicalId":100292,"journal":{"name":"Communications in Transportation Research","volume":"5 ","pages":"Article 100184"},"PeriodicalIF":12.5,"publicationDate":"2025-05-28","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144154446","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

Closed-loop coordination of connected and automated vehicles in structured road networks 结构化道路网络中联网和自动驾驶车辆的闭环协调

IF 12.5 Q1 TRANSPORTATION

Communications in Transportation Research

Pub Date : 2025-05-02 DOI: 10.1016/j.commtr.2025.100182

Shijie Cong , Mingyu Zheng , Wei Zhang , Yiming Bie , Emilian Szczepański

引用次数: 0

Survey of research on autonomous driving testing with large models 大型模型自动驾驶测试研究综述

IF 12.5 Q1 TRANSPORTATION

Communications in Transportation Research

Pub Date : 2025-04-30 DOI: 10.1016/j.commtr.2025.100179

Songyan Liu , Shijie Cong , Lan Yang

引用次数: 0

Physical enhanced residual learning (PERL) framework for vehicle trajectory prediction 用于车辆轨迹预测的物理增强残差学习框架

IF 12.5 Q1 TRANSPORTATION

Communications in Transportation Research

Pub Date : 2025-04-14 DOI: 10.1016/j.commtr.2025.100166

Keke Long, Zihao Sheng, Haotian Shi, Xiaopeng Li, Sikai Chen, Soyoung Ahn

While physics models for predicting system states can reveal fundamental insights owing to their parsimonious structure, they may not always yield the most accurate predictions, particularly for complex systems. As an alternative, neural network (NN) models usually yield more accurate predictions; however, they lack interpretable physical insights. To articulate the advantages of both physics and NN models while circumventing their limitations, this study proposes a physics-enhanced residual learning (PERL) framework that adjusts a physics model prediction with a corrective residual predicted from a residual learning NN model. The integration of the physics model preserves interpretability and tremendously reduces the amount of training data compared with pure NN models. We apply PERL to a vehicle trajectory prediction problem with real-world trajectory data of both a human-driven vehicle (HV) and an autonomous vehicle (AV), using an adapted Newell car-following model as the physics model and four kinds of neural networks (Gated Recurrent Unit (GRU), Convolution long short-term memory (CLSTM), Variational Autoencoder (VAE), and the Informer model) as the residual learning model. We compare this PERL model with pure physics models, NN models, and other physics-informed neural network (PINN) models. The results reveal that PERL yields the best prediction when the training data are small. The PERL model converges quickly during training. Moreover, compared with the NN and PINN models, the PERL model requires fewer parameters to achieve similar predictive performance. A sensitivity analysis revealed that the PERL model consistently outperforms the physics models, NN models and PINN models with different physics and residual learning models given a small training dataset. Among these, the PERL model based on CLSTM achieved the most accurate predictions.

尽管用于预测系统状态的物理学模型因其结构简洁而可以揭示基本的洞察力，但它们并不总能产生最准确的预测，尤其是对复杂系统而言。作为替代方案，神经网络（NN）模型通常能得出更准确的预测结果，但它们缺乏可解释的物理洞察力。为了阐明物理模型和神经网络模型的优势，同时规避它们的局限性，本研究提出了一个物理增强残差学习（PERL）框架，用残差学习神经网络模型预测的修正残差来调整物理模型预测。与纯粹的 NN 模型相比，物理模型的整合保留了可解释性，并大大减少了训练数据量。我们将 PERL 应用于一个车辆轨迹预测问题，该问题包含人类驾驶车辆（HV）和自动驾驶车辆（AV）的真实轨迹数据，我们使用了一个改编的 Newell 汽车跟随模型作为物理模型，并使用四种神经网络（门控循环单元（GRU）、卷积长短期记忆（CLSTM）、变异自动编码器（VAE）和 Informer 模型）作为残差学习模型。我们将 PERL 模型与纯物理模型、NN 模型和其他物理信息神经网络 (PINN) 模型进行了比较。结果表明，当训练数据较少时，PERL 预测效果最好。PERL 模型在训练过程中收敛很快。此外，与 NN 和 PINN 模型相比，PERL 模型需要更少的参数就能达到类似的预测性能。灵敏度分析表明，在训练数据集较小的情况下，PERL 模型的性能始终优于物理模型、NN 模型和采用不同物理和残差学习模型的 PINN 模型。其中，基于 CLSTM 的 PERL 模型的预测结果最为准确。

{"title":"Physical enhanced residual learning (PERL) framework for vehicle trajectory prediction","authors":"Keke Long, Zihao Sheng, Haotian Shi, Xiaopeng Li, Sikai Chen, Soyoung Ahn","doi":"10.1016/j.commtr.2025.100166","DOIUrl":"10.1016/j.commtr.2025.100166","url":null,"abstract":"<div><div>While physics models for predicting system states can reveal fundamental insights owing to their parsimonious structure, they may not always yield the most accurate predictions, particularly for complex systems. As an alternative, neural network (NN) models usually yield more accurate predictions; however, they lack interpretable physical insights. To articulate the advantages of both physics and NN models while circumventing their limitations, this study proposes a physics-enhanced residual learning (PERL) framework that adjusts a physics model prediction with a corrective residual predicted from a residual learning NN model. The integration of the physics model preserves interpretability and tremendously reduces the amount of training data compared with pure NN models. We apply PERL to a vehicle trajectory prediction problem with real-world trajectory data of both a human-driven vehicle (HV) and an autonomous vehicle (AV), using an adapted Newell car-following model as the physics model and four kinds of neural networks (Gated Recurrent Unit (GRU), Convolution long short-term memory (CLSTM), Variational Autoencoder (VAE), and the Informer model) as the residual learning model. We compare this PERL model with pure physics models, NN models, and other physics-informed neural network (PINN) models. The results reveal that PERL yields the best prediction when the training data are small. The PERL model converges quickly during training. Moreover, compared with the NN and PINN models, the PERL model requires fewer parameters to achieve similar predictive performance. A sensitivity analysis revealed that the PERL model consistently outperforms the physics models, NN models and PINN models with different physics and residual learning models given a small training dataset. Among these, the PERL model based on CLSTM achieved the most accurate predictions.</div></div>","PeriodicalId":100292,"journal":{"name":"Communications in Transportation Research","volume":"5 ","pages":"Article 100166"},"PeriodicalIF":12.5,"publicationDate":"2025-04-14","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143825944","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

A causality-based explainable AI method for bus delay propagation analysis 基于因果关系的可解释人工智能总线延迟传播分析方法

IF 12.5 Q1 TRANSPORTATION

Communications in Transportation Research

Pub Date : 2025-04-10 DOI: 10.1016/j.commtr.2025.100178

Qi Zhang , Zhenliang Ma , Zhiyong Cui

Public transportation networks are highly interconnected, where disruptions like traffic congestion propagate bus delays and impact performance. Identifying delay causes is crucial, yet most studies rely on correlation-based methods rather than causal analysis. Attribution methods like the Shapley value quantify factor contributions but often overlook causal dependencies, leading to potential bias. This study uses a causal discovery model to uncover causal relationships between bus delays and various factors (e.g., operational factors, calendar, and weather). Based on this causal graph, an explainable Artificial Intelligence (AI) method quantifies each factor's contribution to delays, focusing on how these contributions vary at different stops along a route. By integrating scheduled route data and real-time vehicle locations, we analyze factor contributions over time and space, exploring various scenarios along the route. Cross-validation is conducted by comparing the importance ranking of factors with the Seemingly Unrelated Regression Equations (SURE). Results show significant variations in factors contributing to delays along the route. Delays at upstream stops propagate downstream, indicating a cascading effect. Operational factors dominate, accounting for 50%–83% of delays. Notably, delays from the preceding two to three stops have a larger impact than just the immediately preceding one stop, and origin delays strongly affect the first half of the route.

公共交通网络高度互联，交通拥堵等中断会导致公交车延误，影响性能。确定延迟原因是至关重要的，但大多数研究依赖于基于相关性的方法，而不是因果分析。像Shapley值这样的归因方法量化了因素的贡献，但往往忽略了因果关系，导致潜在的偏差。本研究使用因果发现模型来揭示公共汽车延误与各种因素（如运营因素、日历和天气）之间的因果关系。基于这张因果图，一种可解释的人工智能（AI）方法量化了每个因素对延误的影响，重点关注这些影响在路线上不同站点的变化。通过整合预定路线数据和实时车辆位置，我们分析了时间和空间上的因素影响，探索了路线上的各种场景。通过比较各因素的重要性排序与看似不相关回归方程（SURE）进行交叉验证。结果显示，导致沿线延误的因素存在显著差异。上游站点的延迟向下游传播，表明级联效应。运营因素占主导地位，占延误的50%-83%。值得注意的是，前两到三站的延误比前一站的延误影响更大，始发点的延误对路线的前半段影响很大。

{"title":"A causality-based explainable AI method for bus delay propagation analysis","authors":"Qi Zhang , Zhenliang Ma , Zhiyong Cui","doi":"10.1016/j.commtr.2025.100178","DOIUrl":"10.1016/j.commtr.2025.100178","url":null,"abstract":"<div><div>Public transportation networks are highly interconnected, where disruptions like traffic congestion propagate bus delays and impact performance. Identifying delay causes is crucial, yet most studies rely on correlation-based methods rather than causal analysis. Attribution methods like the Shapley value quantify factor contributions but often overlook causal dependencies, leading to potential bias. This study uses a causal discovery model to uncover causal relationships between bus delays and various factors (e.g., operational factors, calendar, and weather). Based on this causal graph, an explainable Artificial Intelligence (AI) method quantifies each factor's contribution to delays, focusing on how these contributions vary at different stops along a route. By integrating scheduled route data and real-time vehicle locations, we analyze factor contributions over time and space, exploring various scenarios along the route. Cross-validation is conducted by comparing the importance ranking of factors with the Seemingly Unrelated Regression Equations (SURE). Results show significant variations in factors contributing to delays along the route. Delays at upstream stops propagate downstream, indicating a cascading effect. Operational factors dominate, accounting for 50%–83% of delays. Notably, delays from the preceding two to three stops have a larger impact than just the immediately preceding one stop, and origin delays strongly affect the first half of the route.</div></div>","PeriodicalId":100292,"journal":{"name":"Communications in Transportation Research","volume":"5 ","pages":"Article 100178"},"PeriodicalIF":12.5,"publicationDate":"2025-04-10","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143808766","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

CASAformer: Congestion-aware sparse attention transformer for traffic speed prediction CASAformer：用于交通速度预测的拥塞感知稀疏注意力转换器

IF 12.5 Q1 TRANSPORTATION

Communications in Transportation Research

Pub Date : 2025-04-10 DOI: 10.1016/j.commtr.2025.100174

Yifan Zhang , Qishen Zhou , Jianping Wang , Anastasios Kouvelas , Michail A. Makridis

Accurate and efficient traffic speed prediction is crucial for improving roaDongguand safety and efficiency. With the emerging deep learning and extensive traffic data, data-driven methods are widely adopted to achieve this task with increasingly complicated structures and progressively deeper layers of neural networks. Despite the design of the models, they aim to optimize the overall average performance without discriminating against different traffic states. However, the fact is that predicting the traffic speed under congestion is normally more important than under free flow since the downstream tasks, such as traffic control and optimization, are more interested in congestion rather than free flow. Most of the State-Of-The-Art (SOTA) models unfortunately do not differentiate between the traffic states during training and evaluation. To this end, we first comprehensively study the performance of the SOTA models under different speed regimes to illustrate the low accuracy of low-speed prediction. We further propose and design a novel Congestion-Aware Sparse Attention transformer (CASAformer) to enhance the prediction performance under low-speed traffic conditions. Specifically, the CASA layer emphasizes the congestion data and reduces the impact of free-flow data. Moreover, we adopt a new congestion adaptive loss function for training to make the model learn more from the congestion data. Extensive experiments on real-world datasets show that our CASAformer outperforms the SOTA models for predicting speed under 40 mph in all prediction horizons.

准确、高效的交通速度预测对于提高道路通行和安全效率至关重要。随着深度学习的兴起和交通数据的丰富，数据驱动的方法被广泛采用，神经网络的结构越来越复杂，层次越来越深。尽管模型的设计，他们的目标是优化整体平均性能，而不区分不同的交通状态。然而，事实是，在拥塞情况下预测交通速度通常比在自由流情况下更重要，因为下游任务，如交通控制和优化，对拥塞而不是自由流更感兴趣。不幸的是，大多数最先进的（SOTA）模型在训练和评估期间不能区分交通状态。为此，我们首先综合研究了SOTA模型在不同速度范围下的性能，以说明低速预测的低精度。我们进一步提出并设计了一种新的拥塞感知稀疏注意变压器（CASAformer），以提高低速交通条件下的预测性能。具体来说，CASA层强调拥塞数据，减少自由流数据的影响。此外，我们采用新的拥塞自适应损失函数进行训练，使模型从拥塞数据中学习更多。在实际数据集上进行的大量实验表明，在所有预测范围内，CASAformer在预测40英里/小时以下的速度方面都优于SOTA模型。

{"title":"CASAformer: Congestion-aware sparse attention transformer for traffic speed prediction","authors":"Yifan Zhang , Qishen Zhou , Jianping Wang , Anastasios Kouvelas , Michail A. Makridis","doi":"10.1016/j.commtr.2025.100174","DOIUrl":"10.1016/j.commtr.2025.100174","url":null,"abstract":"<div><div>Accurate and efficient traffic speed prediction is crucial for improving roaDongguand safety and efficiency. With the emerging deep learning and extensive traffic data, data-driven methods are widely adopted to achieve this task with increasingly complicated structures and progressively deeper layers of neural networks. Despite the design of the models, they aim to optimize the overall average performance without discriminating against different traffic states. However, the fact is that predicting the traffic speed under congestion is normally more important than under free flow since the downstream tasks, such as traffic control and optimization, are more interested in congestion rather than free flow. Most of the State-Of-The-Art (SOTA) models unfortunately do not differentiate between the traffic states during training and evaluation. To this end, we first comprehensively study the performance of the SOTA models under different speed regimes to illustrate the low accuracy of low-speed prediction. We further propose and design a novel Congestion-Aware Sparse Attention transformer (CASAformer) to enhance the prediction performance under low-speed traffic conditions. Specifically, the CASA layer emphasizes the congestion data and reduces the impact of free-flow data. Moreover, we adopt a new congestion adaptive loss function for training to make the model learn more from the congestion data. Extensive experiments on real-world datasets show that our CASAformer outperforms the SOTA models for predicting speed under 40 mph in all prediction horizons.</div></div>","PeriodicalId":100292,"journal":{"name":"Communications in Transportation Research","volume":"5 ","pages":"Article 100174"},"PeriodicalIF":12.5,"publicationDate":"2025-04-10","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143808765","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

Counterfactual explanations for deep learning-based traffic forecasting 基于深度学习的交通预测反事实解释

IF 12.5 Q1 TRANSPORTATION

Communications in Transportation Research

Pub Date : 2025-04-09 DOI: 10.1016/j.commtr.2025.100176

Rushan Wang , Yanan Xin , Yatao Zhang , Fernando Perez-Cruz , Martin Raubal

Deep learning models are widely used in traffic forecasting and have achieved state-of-the-art prediction accuracy. However, their black-box nature presents challenges for interpretability and usability, particularly when predictions are significantly influenced by complex urban contextual features. This study aims to leverage an explainable artificial intelligence (AI) approach, counterfactual explanations, to enhance the explainability of deep learning-based traffic forecasting models and elucidate their relationships with various contextual features. We present a comprehensive framework that generates counterfactual explanations for traffic forecasting. The study first implements a graph convolutional network (GCN) to predict traffic speed based on historical traffic data and contextual variables. Counterfactual explanations are generated through a multi-objective optimization process, with four objectives, validity, proximity, sparsity, and plausibility, each emphasizing different aspects of optimization. We investigated the impact of contextual features on traffic speed prediction under varying spatial and temporal conditions. The scenario-driven counterfactual explanations integrate two types of user-defined constraints, directional and weighting constraints, to tailor the search for counterfactual explanations to specific use cases. These tailored explanations benefit machine learning practitioners who aim to understand the model's learning mechanisms and traffic domain experts who seek insights for necessity factors to alter traffic condition. The results showcase the effectiveness of counterfactual explanations in revealing traffic patterns learned by deep learning models and explaining the relationship between traffic prediction and contextual features, demonstrating its potential for interpreting black-box deep learning models.

深度学习模型被广泛应用于交通预测，并达到了最先进的预测精度。然而，它们的黑箱性质对可解释性和可用性提出了挑战，特别是当预测受到复杂城市环境特征的显著影响时。本研究旨在利用一种可解释的人工智能（AI）方法，即反事实解释，来增强基于深度学习的流量预测模型的可解释性，并阐明它们与各种上下文特征的关系。我们提出了一个全面的框架，为交通预测产生反事实解释。该研究首先实现了基于历史交通数据和上下文变量的图形卷积网络（GCN）来预测交通速度。反事实解释是通过多目标优化过程产生的，有四个目标：有效性、接近性、稀疏性和合理性，每个目标都强调优化的不同方面。在不同时空条件下，研究了上下文特征对交通速度预测的影响。场景驱动的反事实解释集成了两种类型的用户定义约束，定向约束和加权约束，以定制针对特定用例的反事实解释的搜索。这些量身定制的解释有利于旨在了解模型学习机制的机器学习从业者和寻求改变交通状况的必要因素见解的交通领域专家。结果表明，反事实解释在揭示由深度学习模型学习的交通模式和解释交通预测与上下文特征之间的关系方面是有效的，证明了其在解释黑箱深度学习模型方面的潜力。

{"title":"Counterfactual explanations for deep learning-based traffic forecasting","authors":"Rushan Wang , Yanan Xin , Yatao Zhang , Fernando Perez-Cruz , Martin Raubal","doi":"10.1016/j.commtr.2025.100176","DOIUrl":"10.1016/j.commtr.2025.100176","url":null,"abstract":"<div><div>Deep learning models are widely used in traffic forecasting and have achieved state-of-the-art prediction accuracy. However, their black-box nature presents challenges for interpretability and usability, particularly when predictions are significantly influenced by complex urban contextual features. This study aims to leverage an explainable artificial intelligence (AI) approach, counterfactual explanations, to enhance the explainability of deep learning-based traffic forecasting models and elucidate their relationships with various contextual features. We present a comprehensive framework that generates counterfactual explanations for traffic forecasting. The study first implements a graph convolutional network (GCN) to predict traffic speed based on historical traffic data and contextual variables. Counterfactual explanations are generated through a multi-objective optimization process, with four objectives, validity, proximity, sparsity, and plausibility, each emphasizing different aspects of optimization. We investigated the impact of contextual features on traffic speed prediction under varying spatial and temporal conditions. The scenario-driven counterfactual explanations integrate two types of user-defined constraints, directional and weighting constraints, to tailor the search for counterfactual explanations to specific use cases. These tailored explanations benefit machine learning practitioners who aim to understand the model's learning mechanisms and traffic domain experts who seek insights for necessity factors to alter traffic condition. The results showcase the effectiveness of counterfactual explanations in revealing traffic patterns learned by deep learning models and explaining the relationship between traffic prediction and contextual features, demonstrating its potential for interpreting black-box deep learning models.</div></div>","PeriodicalId":100292,"journal":{"name":"Communications in Transportation Research","volume":"5 ","pages":"Article 100176"},"PeriodicalIF":12.5,"publicationDate":"2025-04-09","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143800692","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

LC-LLM: Explainable lane-change intention and trajectory predictions with Large Language Models LC-LLM：大语言模型的可解释变道意图和轨迹预测

IF 12.5 Q1 TRANSPORTATION

Communications in Transportation Research

Pub Date : 2025-04-01 DOI: 10.1016/j.commtr.2025.100170

Mingxing Peng , Xusen Guo , Xianda Chen , Kehua Chen , Meixin Zhu , Long Chen , Fei-Yue Wang

To ensure safe driving in dynamic environments, autonomous vehicles should possess the capability to accurately predict lane change intentions of surrounding vehicles in advance and forecast their future trajectories. Existing motion prediction approaches have ample room for improvement, particularly in terms of long-term prediction accuracy and interpretability. In this study, we address these challenges by proposing a Lane Change-Large Language Model (LC-LLM), an explainable lane change prediction model that leverages the strong reasoning capabilities and self explanation abilities of Large Language Models (LLMs). Essentially, we reformulate the lane change prediction task as a language modeling problem, processing heterogeneous driving scenario information as natural language prompts for LLMs and employing supervised fine-tuning to tailor LLMs specifically for lane change prediction task. Additionally, we finetune the Chain-of-Thought (CoT) reasoning to improve prediction transparency and reliability, and include explanatory requirements in the prompts during the inference stage. Therefore, our LC-LLM not only predicts lane change intentions and trajectories but also provides CoT reasoning and explanations for its predictions, enhancing its interpretability. Extensive experiments based on the large-scale highD dataset demonstrate the superior performance and interpretability of our LC-LLM in lane change prediction task. To the best of our knowledge, this is the first attempt to utilize LLMs for predicting lane change behavior. Our study shows that LLMs can effectively encode comprehensive interaction information for understanding driving behavior.

为了保证动态环境下的安全驾驶，自动驾驶汽车应该具备提前准确预测周围车辆变道意图并预测其未来轨迹的能力。现有的运动预测方法有很大的改进空间，特别是在长期预测精度和可解释性方面。在本研究中，我们提出了一个车道变化大语言模型（LC-LLM）来解决这些挑战，这是一个可解释的车道变化预测模型，利用了大型语言模型（llm）强大的推理能力和自我解释能力。从本质上讲，我们将变道预测任务重新表述为一个语言建模问题，将异构驾驶场景信息处理为llm的自然语言提示，并采用监督微调来定制专门用于变道预测任务的llm。此外，我们对思维链（CoT）推理进行了微调，以提高预测的透明度和可靠性，并在推理阶段的提示中包含解释性要求。因此，我们的LC-LLM不仅可以预测变道意图和轨迹，还可以为其预测提供CoT推理和解释，增强了其可解释性。基于大规模高d数据集的大量实验证明了LC-LLM在车道变化预测任务中的优越性能和可解释性。据我们所知，这是第一次尝试利用llm来预测变道行为。我们的研究表明，llm可以有效地编码全面的交互信息，以理解驾驶行为。

{"title":"LC-LLM: Explainable lane-change intention and trajectory predictions with Large Language Models","authors":"Mingxing Peng , Xusen Guo , Xianda Chen , Kehua Chen , Meixin Zhu , Long Chen , Fei-Yue Wang","doi":"10.1016/j.commtr.2025.100170","DOIUrl":"10.1016/j.commtr.2025.100170","url":null,"abstract":"<div><div>To ensure safe driving in dynamic environments, autonomous vehicles should possess the capability to accurately predict lane change intentions of surrounding vehicles in advance and forecast their future trajectories. Existing motion prediction approaches have ample room for improvement, particularly in terms of long-term prediction accuracy and interpretability. In this study, we address these challenges by proposing a Lane Change-Large Language Model (LC-LLM), an explainable lane change prediction model that leverages the strong reasoning capabilities and self explanation abilities of Large Language Models (LLMs). Essentially, we reformulate the lane change prediction task as a language modeling problem, processing heterogeneous driving scenario information as natural language prompts for LLMs and employing supervised fine-tuning to tailor LLMs specifically for lane change prediction task. Additionally, we finetune the Chain-of-Thought (CoT) reasoning to improve prediction transparency and reliability, and include explanatory requirements in the prompts during the inference stage. Therefore, our LC-LLM not only predicts lane change intentions and trajectories but also provides CoT reasoning and explanations for its predictions, enhancing its interpretability. Extensive experiments based on the large-scale highD dataset demonstrate the superior performance and interpretability of our LC-LLM in lane change prediction task. To the best of our knowledge, this is the first attempt to utilize LLMs for predicting lane change behavior. Our study shows that LLMs can effectively encode comprehensive interaction information for understanding driving behavior.</div></div>","PeriodicalId":100292,"journal":{"name":"Communications in Transportation Research","volume":"5 ","pages":"Article 100170"},"PeriodicalIF":12.5,"publicationDate":"2025-04-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143738664","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

Urban rail transit resilience under different operation schemes: A percolation-based approach 不同运营方案下的城市轨道交通弹性：基于渗流的方法

IF 12.5 Q1 TRANSPORTATION

Communications in Transportation Research

Pub Date : 2025-04-01 DOI: 10.1016/j.commtr.2025.100177

Tianlei Zhu , Xin Yang , Yun Wei , Anthony Chen , Jianjun Wu

To assess the resilience of urban rail transit (URT) systems under various operational conditions accurately and enhance their operation, this study develops a percolation model for nonfree flow transportation networks on the basis of percolation theory, which integrates multisource information and operational characteristics. Our model accounts for the state evolution of different hierarchical structures within the network and identifies nonlinear features. Specifically, we observed significant percolation transitions in the URT network, with distinct differences in critical percolation thresholds at different times, leading to multistate behavior. Network bottlenecks spatially shift with network phase transitions, exhibiting power-law frequency characteristics. On the basis of the full-day resilience assessment results, we analyzed the impact of different operational schemes on network resilience during the morning peak, the period with the lowest resilience. The results demonstrate that our resilience analysis framework effectively evaluates URT network resilience, providing theoretical support for enhancing operational management efficiency and accident prevention measures.

为准确评估城市轨道交通系统在不同运行条件下的弹性，提高其运行能力，基于渗流理论，综合多源信息和运行特征，建立了非自由流交通网络渗流模型。我们的模型考虑了网络中不同层次结构的状态演化，并识别了非线性特征。具体而言，我们观察到URT网络中存在显著的渗透转变，不同时间的关键渗透阈值存在明显差异，导致多状态行为。网络瓶颈在空间上随网络相变而变化，表现出幂律频率特性。在全天弹性评估结果的基础上，分析了在弹性最低的早高峰时段，不同运行方案对网络弹性的影响。结果表明，本文提出的弹性分析框架能够有效地评价轨道交通网络弹性，为提高运营管理效率和事故预防措施提供理论支持。

{"title":"Urban rail transit resilience under different operation schemes: A percolation-based approach","authors":"Tianlei Zhu , Xin Yang , Yun Wei , Anthony Chen , Jianjun Wu","doi":"10.1016/j.commtr.2025.100177","DOIUrl":"10.1016/j.commtr.2025.100177","url":null,"abstract":"<div><div>To assess the resilience of urban rail transit (URT) systems under various operational conditions accurately and enhance their operation, this study develops a percolation model for nonfree flow transportation networks on the basis of percolation theory, which integrates multisource information and operational characteristics. Our model accounts for the state evolution of different hierarchical structures within the network and identifies nonlinear features. Specifically, we observed significant percolation transitions in the URT network, with distinct differences in critical percolation thresholds at different times, leading to multistate behavior. Network bottlenecks spatially shift with network phase transitions, exhibiting power-law frequency characteristics. On the basis of the full-day resilience assessment results, we analyzed the impact of different operational schemes on network resilience during the morning peak, the period with the lowest resilience. The results demonstrate that our resilience analysis framework effectively evaluates URT network resilience, providing theoretical support for enhancing operational management efficiency and accident prevention measures.</div></div>","PeriodicalId":100292,"journal":{"name":"Communications in Transportation Research","volume":"5 ","pages":"Article 100177"},"PeriodicalIF":12.5,"publicationDate":"2025-04-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143738665","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

FedAV: Federated learning for cyberattack vulnerability and resilience of cooperative driving automation FedAV：协同驾驶自动化网络攻击脆弱性和弹性的联邦学习

IF 12.5 Q1 TRANSPORTATION

Communications in Transportation Research

Pub Date : 2025-04-01 DOI: 10.1016/j.commtr.2025.100175

Guanyu Lin , Sean Qian , Zulqarnain H. Khattak

Cooperative driving automation (CDA) has gained attention over the years because of its cooperative driving capability that provides solution to individual automated driving challenges. Although reliance on communication and automation enables cooperative driving, it also introduces new cybersecurity threats. This study introduces a federated learning concept for autonomous and connected vehicles, known as the federated agents on vehicle platooning (FedAV) framework, which is designed to address the challenges of cyberattack simulations and anomaly detection in cooperative vehicle platooning systems. The federated learning approach was adopted because of its decentralized nature, which allows each vehicle to learn independently with the ability to overcome adversarial attacks. First, FedAV employs a mixed cyberattack simulation approach to capture complex attack patterns effectively. We tested the scalability of our approach against several attacks, including spoofing, message falsification, and replay attacks, as well as on anomalies, including short anomalies, noise anomalies, bias anomalies, and gradual shifts. In addition, our approach integrates federated learning for decentralized anomaly detection, ensuring data privacy and reducing communication overhead. The anomaly detection performance was enhanced by average and weighted aggregation strategies. Real-world scenarios from cooperative driving experiments and simulations validated the framework's effectiveness and demonstrated its potential to improve the safety, privacy, and efficiency of CDA.

近年来，协作驾驶自动化（CDA）因其协作驾驶能力，为个人自动驾驶挑战提供了解决方案而备受关注。尽管对通信和自动化的依赖使合作驾驶成为可能，但它也带来了新的网络安全威胁。本研究引入了自动驾驶和联网车辆的联邦学习概念，称为车辆队列上的联邦代理（FedAV）框架，旨在解决协作车辆队列系统中网络攻击模拟和异常检测的挑战。采用联邦学习方法是因为它的分散性，它允许每个车辆独立学习，并具有克服对抗性攻击的能力。首先，FedAV采用混合网络攻击模拟方法，有效捕获复杂的攻击模式。我们测试了我们的方法针对几种攻击的可扩展性，包括欺骗、消息伪造和重放攻击，以及异常，包括短异常、噪声异常、偏差异常和逐渐变化。此外，我们的方法集成了用于分散异常检测的联邦学习，确保了数据隐私并减少了通信开销。采用平均聚合和加权聚合策略提高了异常检测性能。来自协作驾驶实验和模拟的真实场景验证了该框架的有效性，并展示了其在提高CDA的安全性、隐私性和效率方面的潜力。

{"title":"FedAV: Federated learning for cyberattack vulnerability and resilience of cooperative driving automation","authors":"Guanyu Lin , Sean Qian , Zulqarnain H. Khattak","doi":"10.1016/j.commtr.2025.100175","DOIUrl":"10.1016/j.commtr.2025.100175","url":null,"abstract":"<div><div>Cooperative driving automation (CDA) has gained attention over the years because of its cooperative driving capability that provides solution to individual automated driving challenges. Although reliance on communication and automation enables cooperative driving, it also introduces new cybersecurity threats. This study introduces a federated learning concept for autonomous and connected vehicles, known as the federated agents on vehicle platooning (FedAV) framework, which is designed to address the challenges of cyberattack simulations and anomaly detection in cooperative vehicle platooning systems. The federated learning approach was adopted because of its decentralized nature, which allows each vehicle to learn independently with the ability to overcome adversarial attacks. First, FedAV employs a mixed cyberattack simulation approach to capture complex attack patterns effectively. We tested the scalability of our approach against several attacks, including spoofing, message falsification, and replay attacks, as well as on anomalies, including short anomalies, noise anomalies, bias anomalies, and gradual shifts. In addition, our approach integrates federated learning for decentralized anomaly detection, ensuring data privacy and reducing communication overhead. The anomaly detection performance was enhanced by average and weighted aggregation strategies. Real-world scenarios from cooperative driving experiments and simulations validated the framework's effectiveness and demonstrated its potential to improve the safety, privacy, and efficiency of CDA.</div></div>","PeriodicalId":100292,"journal":{"name":"Communications in Transportation Research","volume":"5 ","pages":"Article 100175"},"PeriodicalIF":12.5,"publicationDate":"2025-04-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143748613","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0