<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>Reliability Modelling and Requirement Derivation for Automated Railway Inspection Systems under Operational Fallback</title>
      <link>https://trid.trb.org/View/2713304</link>
      <description><![CDATA[Automated inspection systems are increasingly deployed in railway maintenance to reduce workshop-based inspection workload and to support availability-driven planning. Yet, there is limited guidance on how to derive verifiable reliability requirements that are explicitly conditioned by the operational and economic consequences of workshop fallback when automation is unavailable. This paper proposes a consequence-conditioned requirement-derivation framework that (i) models inspection-system availability bottom-up from a modular decomposition, (ii) quantifies fallback-induced functional loss through additional workshop workload, and (iii) formulates an economic admissibility constraint that yields a maximum admissible failure-rate threshold (equivalently, a minimum mean time between failures (MTBF)) consistent with predefined operational-economic targets. The structural properties of this admissibility constraint (feasibility and monotonicity) are analytically characterised to support transparent robustness assessment. The admissible threshold is then operationalised via back-propagation, cast as constrained lifetime-tuning to produce auditable element-level procurement targets under alternative allocation policies. The framework is demonstrated through an industrial automated underframe inspection system, showing how an MTBF requirement for a selected inspection subsystem can be derived from a break-even condition between annual benefit and expected fallback cost, and how required lifetime improvements depend on the allocation policy. The approach positions reliability modelling as a decision-support mechanism for validating the operational and economic viability of automated maintenance-inspection architectures.]]></description>
      <pubDate>Mon, 31 Aug 2026 10:38:39 GMT</pubDate>
      <guid>https://trid.trb.org/View/2713304</guid>
    </item>
    <item>
      <title>Reliability-oriented vessel trajectory prediction through graph-aware embedding and selective state modeling</title>
      <link>https://trid.trb.org/View/2713216</link>
      <description><![CDATA[Reliable vessel trajectory forecasting is essential for maritime risk assessment, traffic supervision, and early collision-warning support, but remains difficult because AIS motion is governed by orientation-rich geometry, region-specific navigation constraints, non-stationary vessel behavior, and noisy environmental observations. We present Mamba-AIS, a reliability-oriented forecasting framework that integrates graph-aware maritime priors with selective state-space temporal modeling. First, we construct a region-aware maritime traffic graph from training AIS polylines and introduce cosine-similarity gated interactions to model high-order relations among sea lanes, turning zones, and coastal constraints while avoiding test-set leakage. Second, we develop a Mamba-based temporal encoder that provides linear-time long-range sequence modeling and improves robustness to local behavioral shifts. Third, we design a domain-guided self-supervised objective that forms maritime-consistent positive and negative pairs from trajectory-weather co-variations, thereby improving temporal heterogeneity modeling. Experiments on AIS trajectories with aligned ERA5 weather data show that Mamba-AIS improves multi-attribute and spatial forecasting accuracy over recurrent, Transformer-based, and recent AIS forecasting baselines, while additional ablations, robustness tests, and feasibility checks verify the contribution of graph encoding, weather fusion, discretized decoding, and land-aware inference.]]></description>
      <pubDate>Mon, 31 Aug 2026 10:38:39 GMT</pubDate>
      <guid>https://trid.trb.org/View/2713216</guid>
    </item>
    <item>
      <title>Multi-ship collision risk assessment based on multi-criteria evaluation strategy</title>
      <link>https://trid.trb.org/View/2709839</link>
      <description><![CDATA[Due to the complex and dynamic interactions between ships, the accurate quantification of multi-ship collision risk remains challenging. This study proposes the multi-ship collision risk assessment method based on a multi-criteria evaluation strategy to achieve quantitative analysis of multi-ship collision risk. Firstly, ship latitude, longitude, speed, and course are acquired from the original Automatic Identification System (AIS) data. Then, the Density-Based Spatial Clustering of Applications with Noise (DBSCAN) algorithm is adopted to cluster ship positions. The value of four key indicators between ship pairs within the clusters are calculated and analyzed the importance of each indicator from objective and subjective weights. Afterwards, the distribution functions are selected for fitting the characteristics of different involved indicators. And then Risk Index Function (RIF) is constructed to quantify the collision risk of individual indicators. Finally, combined with the risk values between ship pairs, the collision risk of each ship is determined based on the Shapley value method. To verify the effectiveness of the proposed method, Automatic Identification System (AIS) data from six sea areas along the U.S. Gulf of Mexico and the Danish coast are used to analyze the collision risks of the involved ships. The results verify the effectiveness of our proposed method, providing a scientific basis for ship navigation safety management and control.]]></description>
      <pubDate>Thu, 27 Aug 2026 10:04:51 GMT</pubDate>
      <guid>https://trid.trb.org/View/2709839</guid>
    </item>
    <item>
      <title>A novel resilience evaluation and enhancement strategy of an interdependent transportation-power distribution system for low-altitude intelligent networks</title>
      <link>https://trid.trb.org/View/2709837</link>
      <description><![CDATA[With the rapid growth of low-altitude intelligent networks, the supporting ground transportation and electric vehicle charging infrastructures are increasingly vulnerable to local overload under high-load conditions. These overloads may spread across interdependent transportation and power distribution systems, causing performance degradation and cascading failures. However, a unified framework for modeling interdependent transportation-power distribution systems, quantifying multi-stage resilience, and developing scenario-specific control strategies is still lacking. To address these problems, resilience evaluation and enhancement strategies of interdependent transportation-power distribution networks are proposed. First, this paper develops an interdependent architecture for transportation-power distribution systems for low-altitude intelligent networks. Then, a multi-stage dynamic resilience assessment framework is proposed for the prevention and recovery stages. In addition, an enhancement strategy based on load balancing is proposed for normal operation scenarios, while an enhancement strategy based on the importance measure is proposed for accident disturbance scenarios. Finally, using the real operational data from a district of Shenzhen in China, a case study is conducted with real traffic flow and power distribution network, comprising 179,940 low-altitude infrastructure-related records and 1,961,280 road records. The results show that the proposed strategy improves the comprehensive resilience value of the interdependent network by 18.76%.]]></description>
      <pubDate>Thu, 27 Aug 2026 10:04:51 GMT</pubDate>
      <guid>https://trid.trb.org/View/2709837</guid>
    </item>
    <item>
      <title>A causality-enhanced network framework for analyzing and governing systemic risk in air traffic control</title>
      <link>https://trid.trb.org/View/2709966</link>
      <description><![CDATA[The repeated activation of key causal factors and insufficient preventive control remain major challenges to the reliability and safety of Air Traffic Control (ATC) systems. Traditional safety analysis methods often struggle to capture the complex and nonlinear causal structures underlying such events. To address this issue, this study develops a causality-enhanced analytical framework for systemic risk analysis in ATC. A tailored Human Factors Analysis and Classification System for Air Traffic Controllers (HFACS-ATCo) was developed to extract and encode causal chains from unsafe event reports, based on which a directed and weighted Air Traffic Controller Event Causation Network (ATCoECN) was constructed. The Role-based Entropy-Weighted TOPSIS (R-EW-TOPSIS) model was then used to evaluate the importance of causal nodes according to their structural roles, and a Path Probability–Frequency–Severity–Importance (PFSI) metric was proposed to identify critical causal propagation paths. Results show that the ATCoECN exhibits a sparse but highly heterogeneous causal structure, in which nodes such as incorrect control instructions, delayed issuance of commands, inadequate risk/conflict assessment, inattention, and information confusion are identified as relatively important. Robustness-based comparative analysis further shows that interventions guided by the R-EW-TOPSIS ranking and PFSI-prioritized paths can produce stronger early-stage disruption of causal connectivity than other strategies. Based on the integrated analysis, governance strategies in different directions are proposed for reference in ATC safety management.]]></description>
      <pubDate>Thu, 27 Aug 2026 10:04:51 GMT</pubDate>
      <guid>https://trid.trb.org/View/2709966</guid>
    </item>
    <item>
      <title>A hybrid multi-criteria methodology for ranking leading indicators of liquefied natural gas leak prevention barriers under uncertainty</title>
      <link>https://trid.trb.org/View/2709881</link>
      <description><![CDATA[The present study proposes a systematic approach to weight and rank performance leading indicators for the safety barriers established against liquefied natural gas (LNG) leaks onboard the marine LNG units. To achieve the paper's target, a literature review of publications relevant to a set of issues related to the risk of marine LNG leaks is conducted. Based on that review, a comprehensive list of 51 performance leading indicators is developed. In addition, a hybrid multi-criteria decision-making (MCDM) approach is adopted to weight and rank the proposed indicators. The results indicate that the indicators’ ranking favours those that monitor the performance of aspects whose failure led to historical process leaks. Examples of these aspects are the integrity of the maintenance management system and the efficiency of developing and following safety-critical procedures. In addition, the results of the MCDM model showed stability against slight changes in the criteria’s weights. The study also discusses the possible uses of the indicators developed and weighted in this study within the safety and risk management in the marine LNG field.]]></description>
      <pubDate>Thu, 27 Aug 2026 10:04:51 GMT</pubDate>
      <guid>https://trid.trb.org/View/2709881</guid>
    </item>
    <item>
      <title>An intelligent, multi-stage maintenance strategy optimization method for supply reliability improvement in natural gas pipeline networks</title>
      <link>https://trid.trb.org/View/2709878</link>
      <description><![CDATA[In natural gas pipeline systems, current equipment maintenance strategies often consider the system and its subunits (e.g., centrifugal compressor stations) individually, leading to inefficiencies and higher costs. This paper proposes an intelligent, systematic optimization framework for centrifugal compressor maintenance strategies, considering system-wide interdependencies. First, the reliability thresholds for each centrifugal compressor station are determined using system reliability and physics-informed Reliability Allocation by Heuristic Optimization. The hydraulic characteristics of natural gas pipelines are integrated into system supply reliability, supported by graph theory and Monte Carlo. Furthermore, the degradation process is modeled using Reliability Block Diagrams, treating the evolution of each centrifugal compressor as a Weibull distribution. Finally, to achieve preventive maintenance and eliminate the risk of gas shortages in the natural gas pipeline network at an optimal cost, the study formulates optimal maintenance policies using artificial intelligence methods such as Deep Reinforcement Learning. Case study on typical natural gas pipeline systems shows that the proposed maintenance strategy reduces costs by 40.8% compared to subunit-based maintenance strategies and by 46.2% compared to time-based maintenance strategies. This study supports the improvement of reliability for subunits in evolving natural gas pipeline systems in practice.]]></description>
      <pubDate>Thu, 27 Aug 2026 10:04:51 GMT</pubDate>
      <guid>https://trid.trb.org/View/2709878</guid>
    </item>
    <item>
      <title>A Global Data-Driven Bayesian Network Model for Ship Collision Accident Analysis</title>
      <link>https://trid.trb.org/View/2709877</link>
      <description><![CDATA[Ship collision accidents remain one of the most frequent and severe types of maritime incidents worldwide, often resulting in significant property damage, loss of life, and pollution. Despite extensive regulatory and technological advancements, the underlying risk mechanisms and consequence formation processes of collision accidents are not yet fully understood, particularly from a global and data-driven perspective. To address this gap, this study develops a data-driven Bayesian Network framework to analyze and predict the consequences of ship collision accidents on a global scale. The proposed model is constructed using a Tree-Augmented Naive Bayes (TAN) structure and is trained on a long-term dataset of collision accident investigation reports covering the period 1978-2024, compiled from multiple international maritime bodies. Unlike most previous studies that primarily focus on causal factors or rely on expert judgment, this research simultaneously investigates accident related, environmental, ship related, and bridge operator related risk influencing factors (RIFs), while explicitly accounting for the characteristics of both vessels involved in a collision. In addition, spatial analysis based on accident coordinates reveals distinct geographical clustering patterns in high traffic coastal and port areas. The findings of the study provide practical value for maritime administrations, ship operators, and policy makers by supporting evidence-based decision making in collision prevention, risk informed regulation, targeted training strategies, and the prioritization of proactive safety measures at both operational and strategic levels.]]></description>
      <pubDate>Thu, 27 Aug 2026 10:04:51 GMT</pubDate>
      <guid>https://trid.trb.org/View/2709877</guid>
    </item>
    <item>
      <title>A Multi-Stage Analytical Framework for Probabilistic Prediction and Structural Risk Prioritization of Fatal Traffic Accidents</title>
      <link>https://trid.trb.org/View/2709868</link>
      <description><![CDATA[The pervasive challenge posed by road traffic accidents, with substantial financial and human costs, necessitates the development of robust analytical frameworks for effective risk prediction and subsequent mitigation. A multi-stage methodology was thus developed in this investigation to accurately predict fatal outcomes and prioritize infrastructural safety improvements and recommendations. Feature assessment was conducted using Cramer's V and SHAP analysis, based on an optimized XGBoost model, whereby road location, environmental light, safety equipment, time interval, and road user type were identified as the most salient predictors of fatality. Furthermore, eigenvector analysis applied to Weight of Evidence-transformed data revealed that road configuration and condition constituted the preponderant element of structural risk variance in the subject dataset. Subsequently, sub-factors were partitioned into precise fatality-risk strata using Gaussian Mixture Model clustering. The predictive determination of a stable fatal accident probability, calculated to be 0.16 under the assumption of factor dependence, was then executed by means of a Monte Carlo simulation founded upon Bayes' theorem. This methodology provides an optimized framework for resource disposition and the maximization of fatal accident burden reduction, establishing a critical instrument for evidence-based policy.]]></description>
      <pubDate>Thu, 27 Aug 2026 10:04:51 GMT</pubDate>
      <guid>https://trid.trb.org/View/2709868</guid>
    </item>
    <item>
      <title>Intelligent Prediction of Cylinder Head Cracking in Commercial Vehicle Natural Gas Engines Using Hybrid Modeling Approaches</title>
      <link>https://trid.trb.org/View/2707003</link>
      <description><![CDATA[Cylinder head cracking in commercial vehicle natural gas engines is a critical failure mode leading to costly unplanned downtime. Traditional model-based prognostics are limited by system complexity, while purely data-driven methods often lack sensitivity to incipient, weak fault signatures in "weakly-related systems" characterized by imprecise mechanisms and sparse sensing. To address this, a hybrid mechanism-data fusion modeling framework is proposed for the intelligent prediction of cylinder head cracking. First, Failure Mode Analysis (FMA) systematically identifies the crack initiation/propagation mechanisms and associated noise factors. Subsequently, these qualitative insights are translated into quantifiable feature indicators with clear physical meanings (e.g., derived power performance ratio, conditional coolant temperature anomaly) to inject prior knowledge. A comprehensive feature set encompassing vehicle, system, component, and geographical dimensions is constructed from Internet of Vehicles (IoV) data, and regression labels spanning from fault-free to severe damage are defined. Finally, considering the temporal evolution of the fault, a Long Short-Term Memory (LSTM) network model is developed for Remaining Useful Life (RUL) prediction. The model was trained and validated using a real-world dataset (206 samples, including 148 faulty cases) provided by FAW Group. On an independent validation set, it achieved an overall accuracy of 88.9%. The model successfully provided an early warning by predicting the maintenance threshold (label 1.8) with a lead time of 2 sampling periods (corresponding to a 15-day system latency), demonstrating its sensitivity to early-stage faults.The developed prognostic model has been successfully deployed on an enterprise cloud platform, integrated with a continuous iteration mechanism based on actual maintenance data. This work not only delivers a high-accuracy predictive maintenance solution for cylinder head cracking but, more importantly, the proposed deep mechanism-data fusion paradigm provides a practical and applicable reference framework for the intelligent fault prognostics of complex systems under conditions of incomplete knowledge and limited sensing.]]></description>
      <pubDate>Mon, 24 Aug 2026 09:03:37 GMT</pubDate>
      <guid>https://trid.trb.org/View/2707003</guid>
    </item>
    <item>
      <title>Dynamic vulnerability assessment of urban rail transit systems using spatiotemporal deep learning</title>
      <link>https://trid.trb.org/View/2706996</link>
      <description><![CDATA[A dynamic vulnerability assessment framework that integrates deep learning and complex networks is proposed to better understand the responses of urban rail transit (URT) systems to various disruptions. Specifically, a spatiotemporal deep learning model is developed to forecast network passenger flows over different time periods. A node importance evaluation method considering dynamic travel demands and static network structural characteristics is presented. A comprehensive metric is designed to assess both structural and functional vulnerabilities of the network. The proposed method is validated using Shanghai's URT system as a case study. The results reveal that: (1) The proposed graph convolutional network-temporal Kolmogorov-Arnold network (GCN-TKAN) model accurately predicts passenger flows across different time periods, achieving an average R² of 0.865 on the testing data, an improvement of 0.056 compared to the second-best model (TKAN); (2) When a sequential single-node attack, guided by the proposed node-importance metric, removes the top 5% of stations, overall network performance drops by 45.00%, which is 4.92% higher than the average decline caused by other basic attack strategies. The innovation of this study lies in the joint modeling of time-varying passenger flows and spatial topological properties, which accurately quantifies the network's vulnerability under different disturbance scenarios, providing a theoretical basis for enhancing system resilience.]]></description>
      <pubDate>Mon, 24 Aug 2026 09:03:37 GMT</pubDate>
      <guid>https://trid.trb.org/View/2706996</guid>
    </item>
    <item>
      <title>Robust reinforcement learning with graph neural networks for stochastic urban rail daily maintenance scheduling</title>
      <link>https://trid.trb.org/View/2706973</link>
      <description><![CDATA[Urban rail daily maintenance scheduling is challenged by urgent task insertions, limited time windows, and multi-resource requirements. This paper formulates the problem as a stochastic resource-constrained project scheduling problem on dynamic directed acyclic graphs and proposes a preference-conditioned graph neural network (GNN) and deep reinforcement learning framework for real-time rescheduling. The GNN combines precedence-aware propagation and change-decaying message passing to encode topology shifts after interruptions and insertions, while deterministic action masking guarantees precedence- and resource-feasible decisions. The policy is conditioned on a dispatcher-specified preference vector and trained with an Augmented Tchebycheff objective, enabling online adaptation among makespan, overtime penalty, financial cost, maintenance quality, and resource fairness without retraining. Experiments on PSPLIB show near-optimal static performance, including a 0.29% optimality gap on J120 with linear inference complexity. In stochastic insertion environments, the method achieves strong Pareto-front quality in hypervolume, IGD, and non-dominated solutions, while obtaining the best fixed-preference composite cost with sub-second inference latency. It also demonstrates robust cross-scale and zero-shot transfer to the Shenzhen Metro network. On real-world metro data, the framework achieves an average makespan of 625.03 min in 1.17 s, while additional stress tests verify its robustness, interpretability, and industrial applicability.]]></description>
      <pubDate>Mon, 24 Aug 2026 09:03:37 GMT</pubDate>
      <guid>https://trid.trb.org/View/2706973</guid>
    </item>
    <item>
      <title>Modeling the Port-Shipping Company Game under Extreme Weather Event Impact: A Multi-Agent Reinforcement Learning-based Approach</title>
      <link>https://trid.trb.org/View/2706604</link>
      <description><![CDATA[Extreme weather events (EWEs) significantly disrupt port operations, creating highly uncertain environments that challenge effective recovery planning for both ports and shipping companies. Existing approaches, particularly those based on evolutionary game theory (EGT), typically rely on static payoff structures and myopic decision-making, limiting their ability to capture system dynamics and long-term strategic interactions. To address these limitations, this study proposes a Multi-Agent Reinforcement Learning (MARL) framework based on the Multi-Agent Deep Deterministic Policy Gradient (MADDPG) algorithm. The framework models recovery investment decisions by port authorities and cargo transfer strategies by shipping companies as continuous, state-dependent actions that evolve over time in a stochastic recovery process. Comparative experiments against a state-dependent EGT (s-EGT) benchmark demonstrate that the proposed approach achieves superior performance in terms of cumulative returns and recovery efficiency. Specifically, the MARL framework learns forward-looking policies that explicitly account for how current decisions influence future system states, even when optimal strategies involve short-term sacrifices. In contrast, the s-EGT model fails to capture these intertemporal effects due to its limited representation of state transitions. The results show that the proposed method accelerates system recovery, improving port efficiency recovery rates by approximately 30% and reducing average vessel waiting times by nearly 80%. Furthermore, cross-scenario fine-tuning experiments indicate that the learned policies generalize well to new disruption scenarios while maintaining stable behavioral patterns. The findings highlight the potential of MARL as an effective decision-support tool for adaptive recovery management in port-shipping systems under extreme weather disruptions.]]></description>
      <pubDate>Mon, 24 Aug 2026 09:03:37 GMT</pubDate>
      <guid>https://trid.trb.org/View/2706604</guid>
    </item>
    <item>
      <title>Dual-modality vessel trajectory anomaly detection for maritime risk prevention based on automatic identification system data</title>
      <link>https://trid.trb.org/View/2706589</link>
      <description><![CDATA[Detecting anomalies in vessel trajectories is an effective approach to prevent maritime risks. However, existing vessel trajectory anomaly detection methods fail to utilize multidimensional vessel features and lack systematic consideration of factors such as trajectory risk. This study proposes a dual-modality vessel trajectory anomaly detection framework using Automatic Identification System (AIS) data. In the numerical modality, geometric features—trajectory tortuosity, self-intersection count, and sharp-turn frequency—are used to construct a trajectory risk model, providing criteria for anomaly detection. In the visual modality, a trajectory image generation method incorporating multidimensional feature information is proposed. Subsequently, an Adaptive Conditional Vision Transformer Auto-Encoder (ACViTAE) model is developed, employing adaptive patch embedding and sparse-aware self-attention to focus on image regions, while integrating meteorological conditions into features via Feature-wise Linear Modulation (FiLM) to enhance reconstruction capability. Finally, dual-modality vessel trajectory anomaly detection is achieved through a decision-level fusion strategy. Experiments in the Port of Baltimore demonstrate that the proposed method detects anomalous vessel trajectories, consistent with maritime management experience and the spatial distribution of historical accident locations. Furthermore, the method exhibits superior performance over mainstream approaches in image reconstruction and multi-scenario anomaly detection, demonstrating engineering applicability and cross-scenario adaptability for maritime risk prevention.]]></description>
      <pubDate>Tue, 18 Aug 2026 14:11:38 GMT</pubDate>
      <guid>https://trid.trb.org/View/2706589</guid>
    </item>
    <item>
      <title>An anomalous behavior detection framework for inland vessels using automatic identification system data</title>
      <link>https://trid.trb.org/View/2706421</link>
      <description><![CDATA[To address the challenges of inadequate parameter optimization and the latency caused by reliance on manual rules, this study proposed a BOA-XGBoost framework integrating trajectory clustering and supervised learning. The Butterfly Optimization Algorithm (BOA) is introduced to adaptively optimize key parameters for both DBSCAN clustering and the XGBoost classifier. In addition, an adaptive and hybrid DP-based vessel trajectory compression strategy is designed to reduce noise while preserving navigational semantic features. To detect anomalous behaviors in inland waterways, a vessel behavior representation system is constructed by jointly characterizing trajectory morphology and multidimensional lateral motion features. Comparative experiments conducted in the Hanjiang River Confluence in the Wuhan reach of the Yangtze River (HRC-Wuhan) and the WISCO Transverse Crossing Area (WTCA) demonstrate that the proposed BOA-XGBoost model achieves detection accuracies of 98.06% and 93.40% for two representative anomalous behaviors—illegal downstream navigation and early river-crossing, respectively—outperforming all baseline models across multiple performance metrics. Furthermore, ten-fold cross-validation and SHAP-based interpretability analysis confirm the robustness and interpretability of the proposed model. The proposed framework effectively identifies anomalous vessel behaviors in complex inland waters, providing reliable technical support for intelligent inland waterway traffic supervision.]]></description>
      <pubDate>Tue, 18 Aug 2026 14:11:38 GMT</pubDate>
      <guid>https://trid.trb.org/View/2706421</guid>
    </item>
  </channel>
</rss>