<rss version="2.0" xmlns:atom="https://www.w3.org/2005/Atom">
  <channel>
    <title>Transport Research International Documentation (TRID)</title>
    <link>https://trid.trb.org/</link>
    <atom:link href="https://trid.trb.org/Record/RSS?s=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" rel="self" type="application/rss+xml" />
    <description></description>
    <language>en-us</language>
    <copyright>Copyright © 2026. National Academy of Sciences. All rights reserved.</copyright>
    <docs>http://blogs.law.harvard.edu/tech/rss</docs>
    <managingEditor>tris-trb@nas.edu (Bill McLeod)</managingEditor>
    <webMaster>tris-trb@nas.edu (Bill McLeod)</webMaster>
    <image>
      <title>Transport Research International Documentation (TRID)</title>
      <url>https://trid.trb.org/Images/PageHeader-wTitle.jpg</url>
      <link>https://trid.trb.org/</link>
    </image>
    <item>
      <title>Machine-Learning-Based Traffic State Prediction in Car–Bicycle Mixed Traffic Using Synthetic Data</title>
      <link>https://trid.trb.org/View/2724773</link>
      <description><![CDATA[This study explores the use of machine learning models to predict traffic conditions in mixed car–bicycle traffic environments. A synthetic dataset was developed from numerical evaluations of traffic flow theory, capturing a wide range of multimodal traffic scenarios. Random forest (RF), multi-layer perceptron (MLP), and linear regression models were trained to estimate key traffic metrics, including output flow, delay, and density. The analysis focuses on model performance under different data splits, especially when sorting by variables such as initial car flow and bicycle flow. Results show that, while RF performs well for previously observed traffic conditions, MLP offers stronger generalization to unseen traffic conditions, particularly in high-flow and high-density regimes. However, prediction performance varies depending on the input variable used for sorting and the distribution of training data. These findings underscore the importance of balanced, diverse datasets and support the use of data-driven models for traffic state estimation in multimodal urban networks.]]></description>
      <pubDate>Fri, 10 Jul 2026 12:12:17 GMT</pubDate>
      <guid>https://trid.trb.org/View/2724773</guid>
    </item>
    <item>
      <title>Integrating subway checkout data for shared micromobility demand forecasting: A novel deep learning framework</title>
      <link>https://trid.trb.org/View/2686723</link>
      <description><![CDATA[This paper proposes an integrated 𝑠ℎ𝑎𝑟𝑒𝑑-𝑚𝑖𝑐𝑟𝑜𝑚𝑜𝑏𝑖𝑙𝑖𝑡𝑦 (𝑠ℎ𝑎𝑟𝑒𝑑-𝜇𝑀) pickups forecasting method for shared 𝜇𝑀 systems based on a 𝑀𝑢𝑙𝑡𝑖-𝐶ℎ𝑎𝑛𝑛𝑒𝑙 𝐴𝑡𝑡𝑒𝑛𝑡𝑖𝑜𝑛 𝑆𝑝𝑎𝑡𝑖𝑜-𝑇𝑒𝑚𝑝𝑜𝑟𝑎𝑙 𝐺𝑟𝑎𝑝ℎ 𝐶𝑜𝑛𝑣𝑜𝑙𝑢𝑡𝑖𝑜𝑛𝑎𝑙 𝑁𝑒𝑡𝑤𝑜𝑟𝑘 (𝑀𝐶-𝐴𝑡𝑡-𝑆𝑇𝐺𝐶𝑁) with a custom 𝑊𝑒𝑖𝑔ℎ𝑡𝑒𝑑 𝑁𝑜𝑟𝑚𝑎𝑙𝑖𝑧𝑒𝑑 𝑀𝑒𝑎𝑛 𝐴𝑏𝑠𝑜𝑙𝑢𝑡𝑒 𝐸𝑟𝑟𝑜𝑟 (𝑊𝑁𝑀𝐴𝐸)  loss function. The method adopts an infrastructure-aware approach for adjacency estimation and systematically compares four strategies for constructing the time-dependent graph: inverse spatial distance, inverse cycling/walking travel time, Pearson demand correlation, and a fully learnable dual local-global attention-based adjacency. Weather inputs are compressed into a single supervised scalar index using an XGBoost-based aggregation scheme to reduce collinearity while preserving predictive information. Additionally, data-related robustness measures and cross-loss diagnostics (MSE, MAE, RMSE, RRSE, oPNBI) are used to assess forecasting quality under data distribution shifts, presence of outliers, and out-of-sample station scenarios.The proposed framework is applied to New York City and Washington, D.C. data demonstrate the interpretability of the model by inspecting multi-channel calibrated weights of various inputs and the learned adjacency matrix, which extracts infrastructure-consistent communities and multimode interaction patterns between stations. Results show enhanced forecasting capabilities while effectively capturing both morning and evening demand peaks at high-volume bike-sharing stations, and maintaining stable performance across off-peak periods. Moreover, at stations with lower demand, our model exhibits superior performance compared to established methods such as 𝐴𝑅𝐼𝑀𝐴, 𝑆𝑇𝐺𝐶𝑁, 𝐺𝑁𝑁-𝐿𝑆𝑇𝑀, 𝐴𝑆𝑇𝐺𝐶𝑁 and 𝑆𝑇𝐴𝐸𝑓𝑜𝑟𝑚𝑒𝑟, achieving on average a double-digit percentage reduction in Mean Absolute Error (MAE), and effectively accommodating weekday-weekend distribution shifts and holiday patterns. Furthermore, incorporating weather (𝑊) data through the compressed scalar index and exploiting the 𝜇𝑀-PT-W setting of the model yield further reductions in error during adverse meteorological conditions, while the explicit inclusion of public transport (PT) checkouts consistently improves forecasts at multimode hubs. These findings underscore the informative value of PT checkout signals and aggregated weather features for accurate and robust bike-sharing pickup forecasting, and highlight attention-based multimode graphs as an effective architecture for operational decision support in urban shared-𝜇𝑀 systems.]]></description>
      <pubDate>Thu, 09 Jul 2026 13:31:03 GMT</pubDate>
      <guid>https://trid.trb.org/View/2686723</guid>
    </item>
    <item>
      <title>Real-time Traffic Prediction for Smart Logistics Using Hybrid Deep Learning-based Traffic Prediction on Cloud Platform</title>
      <link>https://trid.trb.org/View/2685699</link>
      <description><![CDATA[Maximizing the efficiency of the transport and reducing the time of delivery are supported by real-time traffic forecasting during the era of smart logistics. AI and Big Data make intelligent traffic forecasting possible, which allows making a better decision aiding logistics company. The majority of the existing approaches lack the means to address the dynamic traffic conditions effectively, as most of them are likely to absorb both the temporal correlations and the spatial dependencies at an unsatisfactory level. This translates to wrong predictions and inefficient route planning. To address these issues, we propose a Hybrid Deep Learning-Based Traffic Prediction on cloud platform Model (HDL-TPM), which is a version of Long Short-Term Memory (LSTM) networks and Graph Neural Networks (GNNs). STM takes into account changes in traffic flows in sequence, whereas GNN captures spatial networks of roads. Even while cloud-based learning models are increasingly ubiquitous, this research is unique as it uses a unified cloud-native spatiotemporal pipeline that combines distributed data collection, GNN-based spatial modeling, and LSTM-driven temporal forecasting. It allows for scalable, parallel, real-time traffic prediction from a range of urban sensor sources, growing beyond traditional batch-driven cloud systems. The hybrid strategy provides scalable traffic forecasts through big data processing in real-time. The model handles large traffic datasets, predicting congestion precisely and optimizing logistics routes. Through the learning process of past and current traffic data, HDL-TPM improves traffic flow forecasting accuracy, preventing delay in intelligent logistics. The experimental results show that HDL-TPM performs better compared to traditional models, providing greater prediction accuracy and better efficiency in intelligent logistics operations.]]></description>
      <pubDate>Thu, 09 Jul 2026 13:29:26 GMT</pubDate>
      <guid>https://trid.trb.org/View/2685699</guid>
    </item>
    <item>
      <title>Trip Generation and Traffic Prediction: Evaluating Alternative Traffic Prediction Methods for Traffic Impact Analysis</title>
      <link>https://trid.trb.org/View/2721744</link>
      <description><![CDATA[ Traffic impact analysis (TIA) forecasts how a proposed development will affect the surrounding transportation system and what improvements, if any, are needed to improve safety and efficiency.  In addition to trips generated by the proposed development (typically based on rates published by the Institute of Transportation Engineers [ITE], a TIA will include “background traffic” which is traffic on the roadway generated from other sources. The assumption is that background traffic growth is not driven by the proposed development, nor by other parcels approved but not yet built within the study area.  The purpose of this research is twofold: (1) to determine the extent to which this assumption is valid and (2) if the assumption is not valid, to identify best practices to account for this.  This research will thus (1) review practices from other states regarding ways to combine background traffic growth and site-specific trip generation in the TIA process; (2) compare forecast and observed trip generation for a selected set of developments; and (3) conduct an additional case study to compare the traditional approach of doing a TIA (based on ITE rates) and an approach based on a travel demand model.  Lessons learned from this effort may inform Virginia Department of Transportation (VDOT)’s Traffic Impact Analysis guidelines.  This research need tied with another for being the top-ranked research need by the Transportation Planning Research Advisory Committee (TPRAC).]]></description>
      <pubDate>Thu, 02 Jul 2026 11:02:16 GMT</pubDate>
      <guid>https://trid.trb.org/View/2721744</guid>
    </item>
    <item>
      <title>A multiscale decomposition and semantics-guided feature reconstruction LLM framework for traffic flow forecasting</title>
      <link>https://trid.trb.org/View/2684466</link>
      <description><![CDATA[Traffic flow forecasting plays a critical role in intelligent transportation systems (ITS), particularly in urban congestion management and resource allocation. Although large language models (LLMs) have demonstrated powerful capabilities in sequence modeling and reasoning, their application to the traffic domain remains challenging due to inherent modality differences. Specifically, how to effectively transform numerical traffic time series into semantic representations that can be understood by language models. Moreover, traffic flow data often exhibits complex and varying temporal patterns across different sampling scales. The inherent multiscale nature and multidimensional heterogeneity of such data impose high demands on models to achieve structural alignment and semantic fusion, posing significant challenges for conventional LLMs that are primarily designed for text-based tasks. To address these issues, this paper proposes a novel traffic flow forecasting framework, Reconstructed Multiscale Forecasting for Traffic (ReMFT), which integrates multiscale temporal decomposition with semantic integration. The framework first incorporates a multiscale decomposition module to separate and reconstruct trend and seasonal components from raw traffic series, thereby revealing latent temporal structures across different time scales. It then introduces a semantics-guided feature reconstruction module that semantically aligns and embeds the decomposed features in a non-linguistic manner, making them compatible with the LLM’s input representation preferences. This design enables a smooth transition from structural time series modeling to semantic representation learning, enhancing the LLM’s generalization and reasoning capabilities when dealing with the spatiotemporal heterogeneity and dynamic patterns of traffic flow. Experimental results on multiple real-world traffic datasets demonstrate that the proposed method achieves competitive or superior performance compared to several state-of-the-art (SOTA) forecasting models. The code is available at: https://github.com/IansSUn/ReMFT.]]></description>
      <pubDate>Tue, 30 Jun 2026 17:02:35 GMT</pubDate>
      <guid>https://trid.trb.org/View/2684466</guid>
    </item>
    <item>
      <title>A multi-dimensional feature-optimized deep learning framework with self-adaptive attention for container throughput forecasting in sea-rail intermodal systems</title>
      <link>https://trid.trb.org/View/2684786</link>
      <description><![CDATA[Accurate forecasting of container throughput in sea-rail intermodal transport can effectively reduce cargo turnover time and energy consumption, achieving cost reduction and efficiency improvement in multimodal transport. Despite deep learning models being an effective means of solving container throughput prediction, they struggle with some significant challenges, such as difficulty in learning complex nonlinear relationships among multidimensional data. Therefore, this paper constructs a systematic indicator selection framework based on statistics and proposes a novel hybrid deep learning prediction model that integrates bayesian optimization (BO), convolutional neural networks (CNN), long short-term memory (LSTM), and self-attention (SA), which is further equipped with a weight assignment method (WA) to optimize the feature extraction of the convolutional layer. Using real-world data, we constructed the BO-WACNN-LSTM-SA model and conducted a comprehensive comparative analysis against benchmark models. Results indicate that the proposed model achieves the lowest prediction error among the compared baselines. Additionally, we further validate the robustness and generalizability of the proposed architecture through transferability tests, ablation studies, and time-series cross-validation. The results indicate that WA and SA optimize the model by operating on feature importance at different stages. They act on the input and inside the model, respectively, forming a multi-level information filtering mechanism and complementing each other, thereby significantly improving the accuracy of regression prediction. The research results can provide a reference for the preparation of container transportation plans and allocation of transportation resources in sea-rail intermodal transport, which is advantageous for the early layout and scheduling of multimodal transport.]]></description>
      <pubDate>Tue, 30 Jun 2026 17:02:35 GMT</pubDate>
      <guid>https://trid.trb.org/View/2684786</guid>
    </item>
    <item>
      <title>Predicting Demand and Supply in a Real-Time Traffic Management Framework</title>
      <link>https://trid.trb.org/View/2580108</link>
      <description><![CDATA[Predicting the future supply and demand of a transport network are challenging and important problems in real-time traffic management systems that are essential to enhance the decision-making process for deploying adequate traffic strategies under different conditions (e.g., road works, accidents). In the context of the TANGENT H2020 project, simulation-based and data-driven methodologies are developed focusing on the real-time demand and supply prediction problems. This paper focuses on the development and integration of the demand and supply models as well as incident detection methods into traffic simulation environments for network-wide traffic predictions. The role of each component of the framework and their interoperability is explained in the paper, using as testbed the network of Athens, Greece.]]></description>
      <pubDate>Tue, 30 Jun 2026 09:45:14 GMT</pubDate>
      <guid>https://trid.trb.org/View/2580108</guid>
    </item>
    <item>
      <title>Smart City Traffic Flow Prediction and Signal Dynamic Optimization Algorithm Based on Deep Learning</title>
      <link>https://trid.trb.org/View/2711125</link>
      <description><![CDATA[With urbanization and the rapid growth in the number of motor vehicles, urban traffic congestion has intensified. The traditional clock signal control system is difficult to meet the dynamic needs of traffic. Effective traffic flow forecasting and signal optimization are critical to alleviating congestion and improving efficiency. Currently, traffic flow forecasting models are vulnerable to complex scenarios, such as sudden accidents and extreme weather, resulting in poor accuracy. Signal optimization algorithms are generally based on static rules, which are difficult to integrate into real-time prediction, resulting in delays in signal adjustment and it is difficult to achieve optimal dynamic allocation of traffic resources. This work begins with building a deep learning prediction model that integrates data from multiple sources (traffic, weather data, and spatio-temporal characteristics). It uses an improved LSTM to capture the spatio-temporal dependence of traffic flow and improve prediction accuracy. Secondly, an algorithm for optimizing dynamic signals is designed based on reinforcement learning with prediction results as input. Establish a reward function to reduce average vehicle delay and improve road efficiency. Finally, the model was formed and verified using real traffic data sets from cities such as Beijing and Shenzhen. Then, a simulation platform is built to simulate the effect of signal control in different scenarios. Experiments showed that during peak hours in the morning, the proposed algorithm reduced the average delay by 43% compared to the fixed timing scheme and by 25% compared to the scoot algorithm. During night rush hours, the proposed algorithm reduces the average delay by 42% compared to the fixed timing scheme and by 24% compared to the scoot algorithm, far exceeding the traditional scheme.]]></description>
      <pubDate>Tue, 30 Jun 2026 09:45:05 GMT</pubDate>
      <guid>https://trid.trb.org/View/2711125</guid>
    </item>
    <item>
      <title>A Kernel-Based Approach for Simulating Synthetic Travel Patterns</title>
      <link>https://trid.trb.org/View/2710962</link>
      <description><![CDATA[Modeling dynamic traffic processes requires detailed traveler-level demand unavailable in aggregated national forecasting models. We present a kernel-based approach for generating synthetic disaggregate travel demand using travel survey data. The method links survey respondents to agents in a synthetic population through similarity kernels over socio-demographics, schedule characteristics, and activity–location feasibility, and embeds these in a Metropolis–Hastings sampling framework to generate complete daily plans. The approach provides a modular way to combine heterogeneous information sources and to transfer observed behaviour to new populations without requiring exact matches between individuals. We demonstrate the approach on a stylized grid-world and on real data, where survey records are mapped to a national synthetic population. Results show that the method reproduces key marginal travel patterns, preserves individual variability, and avoids collapse toward modal behaviour, illustrating its potential as a demand synthesis and data-fusion tool in dynamic traffic modelling.]]></description>
      <pubDate>Tue, 30 Jun 2026 09:45:05 GMT</pubDate>
      <guid>https://trid.trb.org/View/2710962</guid>
    </item>
    <item>
      <title>Behaviorally informed joint optimization of charger placement and dynamic spatio-temporal pricing for electric vehicle networks</title>
      <link>https://trid.trb.org/View/2692392</link>
      <description><![CDATA[The rapid growth of electric vehicles calls for efficient strategies to design charging networks that are both profitable and accessible. This paper develops an integrated framework that jointly optimizes charger placement and dynamic pricing by combining column generation with reinforcement learning. The column generation master problem governs the selection of charger configurations under budget and accessibility constraints, while pricing is modeled as a sequential decision-making problem and solved using reinforcement learning. To address the intractability of the column generation pricing problem, in which reduced costs depend on reinforcement-learning-based pricing outcomes and therefore admit no closed-form expression, we introduce a set of heuristics to generate promising charger configurations. These candidate configurations are subsequently evaluated using reinforcement learning to estimate their expected profitability. Computational experiments on a synthetic urban grid demonstrate that spatio-temporal pricing yields significantly higher expected profit than uniform pricing strategies, and that extending the behavioral model from spatial-only user charging relocation to joint spatial and temporal charging relocation yields statistically significant revenue gains. The results also highlight the sensitivity of network profitability to the presence of competitors, as well as the robustness of the proposed approach under imperfect demand forecasts. Finally, scalability tests show that heuristic-guided column generation enables efficient solutions for larger networks. These findings underscore the importance of integrating user choice modeling, dynamic pricing, and adaptive optimization in the planning of future EV charging infrastructures.]]></description>
      <pubDate>Tue, 30 Jun 2026 09:45:02 GMT</pubDate>
      <guid>https://trid.trb.org/View/2692392</guid>
    </item>
    <item>
      <title>Interpretable deep learning for traffic prediction: Revealing the role of urban morphology in spatially heterogeneous cities</title>
      <link>https://trid.trb.org/View/2704144</link>
      <description><![CDATA[Urban morphological diversity strongly shapes traffic dynamics, yet conventional prediction systems treat spatially heterogeneous cities uniformly, producing systematic geographic disparities. Morphologically complex areas – dense mixed-use districts with irregular street networks – are associated with 42%–62% higher prediction errors than uniform suburbs across three metropolitan regions, concentrated in areas with higher population densities and greater transit dependency. This disparity persists across all model architectures, indicating morphological complexity as a persistent property of urban traffic systems rather than a model-specific artifact. To address this, we develop MorphNet, an interpretable deep learning framework integrating urban form into spatiotemporal traffic prediction through three innovations: (1) Interpretable morphological encoding extracts three spatial indices – heterogeneity, autocorrelation, and complexity – from 47-dimensional geographic data via variational autoencoders; (2) Dynamic graph construction enables knowledge transfer between morphologically similar areas regardless of geographic distance, connecting analogous urban contexts across metropolitan boundaries; (3) Dual-path neural architecture combines spatiotemporal prediction with morphology-aware corrections, generating location-specific adjustments with full interpretability. Cross-metropolitan validation across Los Angeles, San Francisco Bay Area, and Tokyo reveals a prediction hierarchy strongly correlated with morphological complexity (r=0.67, p<0.001), confirming urban form as a significant spatial determinant. MorphNet achieves 15%–30% error reduction with spatially differentiated gains: 28% improvement in morphologically complex areas versus 8% in uniform areas, with larger gains in higher-complexity areas that tend to serve denser, more transit-dependent populations. This work demonstrates that effective intelligent transportation systems must incorporate spatial context alongside temporal patterns, enabling geographically adaptive deployment across diverse urban environments.]]></description>
      <pubDate>Tue, 30 Jun 2026 08:55:58 GMT</pubDate>
      <guid>https://trid.trb.org/View/2704144</guid>
    </item>
    <item>
      <title>Deep structured Gaussian modeling for Real-Time probabilistic bus travel time prediction</title>
      <link>https://trid.trb.org/View/2681773</link>
      <description><![CDATA[Recent advancements in statistical and machine learning models have substantially enhanced bus travel time forecasting accuracy. However, these studies primarily rely on deterministic models and fail to quantify forecasting uncertainty, which is crucial for travelers and transit operators. Moreover, existing probabilistic methods are either computationally prohibitive or restricted to local dependencies, preventing them from modeling the joint travel time correlation among multiple running buses and links along the route. To address these issues, this paper introduces a probabilistic deep structured Gaussian model that performs joint forecasting of link travel times for all running buses on a route by explicitly capturing both intra-bus and inter-bus correlations. We model the travel time of each bus on every link as a random variable, whose joint correlations are captured using a time-varying multivariate Gaussian mixture distribution. Efficient high-dimensional parameter estimation is achieved through a novel deep neural network that incorporates a Kronecker product-based structure for the covariance matrix, with the multivariate Gaussian mixture likelihood as the loss function. This specialized architecture enables the network to effectively learn dynamic intra-bus and inter-bus correlations by fusing spatiotemporal features encoded from the travel times of the preceding buses. Probabilistic forecasting is then conducted by computing the conditional distribution of downstream bus link travel times based on partially observed upstream travel times, facilitating real-time predictions for all remaining links of all running buses. We evaluate the proposed model with three routes from two bus systems. Compared with other baseline models, results show that our approach achieves an average improvement of 0.44% in MAPE, 1.12 in RMSE, 0.36 in CRPS, 0.20 in 0.5-risk, and 0.24 in 0.9-risk. Furthermore, the model provides interpretable operational insights, capturing time-varying inter-bus correlations with complex short-range and long-range intra-bus dependencies.]]></description>
      <pubDate>Thu, 25 Jun 2026 15:57:06 GMT</pubDate>
      <guid>https://trid.trb.org/View/2681773</guid>
    </item>
    <item>
      <title>TransFM: A foundation model for cross-city multimodal transportation demand prediction</title>
      <link>https://trid.trb.org/View/2681772</link>
      <description><![CDATA[Accurate multimodal transportation demand forecasting is crucial for seamless urban mobility and proactive resource allocation. However, existing multimodal demand prediction approaches face a fundamental dilemma. Specialized models are typically tailored to specific cities/modes and thus struggle with cross-city generalization. By contrast, universal models emphasize transferable representations across cities, but inadequately capture inter-mode distinctions and urban heterogeneity. To transcend these limitations, we propose TransFM, a foundation model that advances multimodal transportation demand forecasting from the traditional single-city focus to cross-city scenarios. TransFM resolves two core challenges: (1) reconciling mode-specific attributes with cross-mode interaction at the mode-level; (2) bridging heterogeneity and universality at the city scale. For the first challenge, we design a Mode-Specialized Memory-Augmented Mixture-of-Experts (MoE), which employs dedicated experts with private memory banks to encode distinct mode attributes, while enabling cross-mode interaction via inter-mode memory interactions. For the second challenge, a Cross-City Universal Spatial-Temporal MoE is proposed. This module features two key components: heterogeneity-aware experts that capture the demand variations inherent to each city-mode pair, and universal prototype memory banks that distill transferable spatiotemporal patterns across diverse urban contexts. Extensive experiments on eleven datasets across six cities and four modes show that TransFM achieves an average performance gain of 5.63% over advanced baselines. More notably, in few-shot cross-city scenarios, it obtains superior performance by fine-tuning fewer than 6.50% of parameters, confirming its exceptional capability to transfer knowledge from data-rich to data-scarce cities.]]></description>
      <pubDate>Thu, 25 Jun 2026 15:57:06 GMT</pubDate>
      <guid>https://trid.trb.org/View/2681772</guid>
    </item>
    <item>
      <title>Resilient Capacity Management: Predicting General Aviation Traffic Counts with Deep Learning</title>
      <link>https://trid.trb.org/View/2681226</link>
      <description><![CDATA[An accurate forecast of general aviation traffic is crucial for air navigation service providers, as it affects the overall efficiency of air traffic management and capacity planning. This paper presents a deep learning methodology for predicting general aviation traffic, combining calendar and meteorological sources through a detailed feature-engineering procedure. The approach is rigorously evaluated using historical data from the Nice Cote D’Azur Terminal Control Center sectors, resulting in a significant 32% enhancement in global prediction performance with recurrent neural network models compared to existing operational tools. The paper explores additional analysis techniques to gain a deeper understanding of the predictions that are produced by each model.]]></description>
      <pubDate>Thu, 25 Jun 2026 09:40:59 GMT</pubDate>
      <guid>https://trid.trb.org/View/2681226</guid>
    </item>
    <item>
      <title>Spatiotemporal Graph Mixture of Experts for Highway Traffic Flow Prediction</title>
      <link>https://trid.trb.org/View/2672987</link>
      <description><![CDATA[With excellent learning ability, the pretrained large model is challenging the mainstream traffic prediction paradigm. However, the pretraining process of large spatiotemporal models still faces the problems of high training cost and fixed graph size limitation, which hinders the practical application of more flexible prediction models in intelligent transportation systems. To address these challenges, this article proposes the Spatiotemporal Graph Mixture of Experts (STGMoE), a novel framework that integrates dynamic graph message passing with an MoE mechanism. The proposed STGMoE framework takes multivariate time-series data as input and enables efficient conditional computation and flexible topological adaptation, ultimately facilitating accurate spatiotemporal feature extraction for downstream traffic prediction tasks. Experiments on the California PeMS and Beijing datasets demonstrate that the model outperforms mainstream methods in fully supervised prediction and zero-shot prediction, validating its generalization capability in complex and dynamic traffic environments.]]></description>
      <pubDate>Wed, 24 Jun 2026 13:22:06 GMT</pubDate>
      <guid>https://trid.trb.org/View/2672987</guid>
    </item>
  </channel>
</rss>