<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>Hybrid approach for enhanced wave spectrum estimation integrating semi-supervised learning and wave buoy analogy</title>
      <link>https://trid.trb.org/View/2693402</link>
      <description><![CDATA[This study presents an innovative approach for estimating wave parameters by integrating a modified wave-buoy analogy with semi-supervised learning techniques, utilizing ship motion responses. The proposed methodology employs a two-step method to accurately determine wave parameters including incoming wave direction, spreading strength, height, and period. In the first step, the primary incoming wave direction is identified using t-Distributed Stochastic Neighbor Embedding (t-SNE) method combined with graph-based semi-supervised learning. This step involves clustering techniques that utilize feature values derived from ship motion response spectra, which exhibit distinct patterns indicative of incoming wave directions. Notably, by clustering these values within a 0 to 360-degree range, the method provides more accurate estimation with limited data compared to supervised learning techniques. Subsequently, transfer functions based on the identified wave direction, are employed to estimate additional wave spectrum attributes such as height and period. This second step addresses the challenges associated with unbounded parameters like wave height and period using a wave-buoy analogy. The validation of the proposed methodology was conducted through ship motion simulation data across various sea states for a large conventional ship. The accuracy of the two-step approach was benchmarked against a brute force algorithm. Furthermore, the study explores how estimation accuracy varies with dataset size, demonstrating robust performance even with constrained data inputs. The analysis concludes by discussing observed errors and suggesting future research directions to refine the estimation process.]]></description>
      <pubDate>Mon, 27 Apr 2026 15:01:23 GMT</pubDate>
      <guid>https://trid.trb.org/View/2693402</guid>
    </item>
    <item>
      <title>Data-driven inverse estimation of ocean wave parameters using field-measured FPSO motion and environmental data</title>
      <link>https://trid.trb.org/View/2631376</link>
      <description><![CDATA[This paper proposes artificial neural network (ANN) models to inversely estimate significant wave height, peak period, and mean wave direction from field-measured motion data of a Floating Production Storage and Offloading (FPSO) vessel. Measured hull motion data over approximately three years were utilized alongside environmental data measured from the Wave Monitoring System II (WaMoS II). Additionally, ECMWF Reanalysis v5 (ERA5) and Hybrid Coordinate Ocean Model (HYCOM) datasets were employed to assess the impact of data resolution on model performance. The ANN models were trained and tested using statistical values of vessel motions and environmental conditions. Various combinations of input variables and time intervals of statistical values were examined to optimize the models’ predictive performance. The results demonstrate that the best ANN models inversely estimate significant wave height with reasonable accuracy, achieving a Root Mean Square Error (RMSE) of 0.185 m. Although the estimation of wave period and direction presented more challenges due to complex nonlinear interactions among multiple environmental factors, the models still show potential for capturing these parameters. Sensitivity analyses indicate that estimation accuracy is heavily influenced by the quality, quantity, and completeness of the input data, emphasizing the need for comprehensive datasets and careful selection of influential variables.]]></description>
      <pubDate>Mon, 27 Apr 2026 15:01:22 GMT</pubDate>
      <guid>https://trid.trb.org/View/2631376</guid>
    </item>
    <item>
      <title>Semi-active actuator-based hybrid model predictive control for vehicle tilting</title>
      <link>https://trid.trb.org/View/2691731</link>
      <description><![CDATA[Typically, vehicle height adjustment during driving can only be achieved using active actuators. However, our research group has proposed a method that uses adjustable damping semi-active actuators to control the raising and lowering of the isolation system. This method leverages asymmetrical damping to adjust the vehicle’s height and also utilizes the vehicle’s vibration energy during driving. In this study, this approach is employed to control the shifting of the left and right sides of the vehicle, thereby implementing tilt control by inclining the vehicle body toward the turning direction. To optimize control for this posture adjustment, nonlinear factors such as the nonlinear constraints of the damper’s output damping force and the logical control conditions of asymmetric damping adjustment are considered, designing this control system as a hybrid system. The system is described using HYSDEL programing language, forming a Mixed Logic Dynamic (MLD) system, and a control logic is constructed for shifting single wheels through damping switching to achieve tilt control. A full vehicle tilt controller is designed using hybrid model predictive control theory, transforming the switching control problem into a continuous receding horizon optimization control problem, integrating single-wheel height adjustment and full vehicle tilt control, and defining the objective function and nonlinear constraint conditions that need to be met. Both simulation and co-simulation calculations have confirmed the effectiveness of this method, which uses asymmetric damping adjustment to shift the vehicle body and implement full vehicle tilt control, providing a new approach for posture control under specific conditions such as short-term obstacle crossing, high-speed cornering, emergency maneuvers, and rollover prevention.]]></description>
      <pubDate>Thu, 23 Apr 2026 13:54:43 GMT</pubDate>
      <guid>https://trid.trb.org/View/2691731</guid>
    </item>
    <item>
      <title>A transformer-based architecture for wave height forecasting within rectangular moonpool and its generalization performance study</title>
      <link>https://trid.trb.org/View/2660800</link>
      <description><![CDATA[Moonpools are widely used in offshore oil and gas drilling and marine energy development. A calm wave environment is required at an operational site for effective functioning. However, affected by incident waves or harmonic ship motions, the fluid inside the moonpool may undergo significant resonant motions that can significantly disrupt operations and compromise offshore safety. Therefore, dependable moonpool wave height forecasting is important for guiding offshore operations. Our study aimed to develop a generalizable model for moonpool wave height forecasting. Initially, we conducted a model test to obtain the wave height within the moonpool and hull motion data. Subsequently, a self-supervised training model called Generalizable Wave height Transformer (GenWT) based on the Transformer framework is proposed. This model utilizes moonpool resonance information to enhance predictive accuracy. Experiments demonstrate the superior performance of GenWT compared to other benchmark models such as Informer, long short-term memory (LSTM) models, achieving 26.15% and 27.98% MSE reduction in two sea conditions, respectively. Owing to the pretrain + fine-tune strategy, the time and space complexity of the GenWT model can be reduced quadratically. Furthermore, the same pre-model can be applied to different ocean engineering forecast tasks while maintaining optimal performance, highlighting GenWT's robustness and potential in ocean engineering applications.]]></description>
      <pubDate>Thu, 23 Apr 2026 09:12:25 GMT</pubDate>
      <guid>https://trid.trb.org/View/2660800</guid>
    </item>
    <item>
      <title>Development of one-week ahead work safety prediction system combining machine learning-based wave prediction and motion response prediction of a workvessel</title>
      <link>https://trid.trb.org/View/2669608</link>
      <description><![CDATA[This study aims to construct a system that evaluates work safety one week in advance by combining wave prediction with motion response prediction of workvessel using machine learning. The authors enhanced the accuracy of wave prediction by utilizing data from NOWPHAS’s GPS wave gauges and meteorological information from ERA5. The Long Short Term Memory (LSTM) algorithm was adopted, and two approaches were undertaken: “point prediction” and “historical prediction”. Point prediction targeted the prediction of wave conditions at specific times, while historical prediction forecast time-series data up to one week ahead. Historical prediction proved advantageous for work planning because it could simultaneously learn and reflect dependencies among multiple future time points. The computational results successfully reduced the probability of prediction failures, specifically underestimations in wave predictions, to approximately 20%. This reduction is critical because underestimating wave heights can lead to unsafe operations. Furthermore, by calculating the prediction accuracy of the motion response of workvessel, the authors developed an operational system that enables safe offshore operations.]]></description>
      <pubDate>Mon, 20 Apr 2026 11:14:50 GMT</pubDate>
      <guid>https://trid.trb.org/View/2669608</guid>
    </item>
    <item>
      <title>Does Height Matter? Analysis of Contributing Factors to Tall-Vehicle/Pedestrian Crashes</title>
      <link>https://trid.trb.org/View/2691021</link>
      <description><![CDATA[With increasing urbanization, interactions between pedestrians and vehicles have become more frequent, raising safety concerns. Now comprising about 40% of the consumer fleet, tall vehicles such as trucks and sport utility vehicles pose unique risks with their elevated front profiles and large blind zones. This “car bloat” trend, combined with distracted and risky driving behaviors, has contributed to an 80% increase in pedestrian fatalities in the U.S. since its record low in 2009. In this study, an in-depth analysis of 15 years of single-vehicle/single-pedestrian crash data (2008–2022) from Wisconsin uncovers that tall vehicles, defined as 5.5 ft (66 in.) or greater in height, are disproportionately involved in crashes during left turns and backing maneuvers, with higher risks across specific pedestrian locations, pedestrian actions, and area type (urban versus rural). The results of a binary logistic model quantify that tall vehicle involvement was significantly associated with specific driver actions, such as left turns, as well as road types, and pedestrian presence in crosswalks, in addition to risk factors such as speed and driver behavior. The findings are instrumental to identify effective countermeasures to improve pedestrian visibility to tall vehicles and prioritize targeted strategies for roadway design, integrated planning, data-driven safety analysis, and targeted driver education addressing tall-vehicle risks.]]></description>
      <pubDate>Mon, 13 Apr 2026 08:41:21 GMT</pubDate>
      <guid>https://trid.trb.org/View/2691021</guid>
    </item>
    <item>
      <title>Performance effects of center of gravity and weight distribution on a single-step planing vessel in regular waves</title>
      <link>https://trid.trb.org/View/2653063</link>
      <description><![CDATA[Dynamic stability in waves is crucial for the performance of planing vessels and crew safety. A key factor is the vessel's weight arrangement, which influences radius of gyration, motions, and accelerations. This study experimentally examined a 2.5m model with a constant 20° deadrise and a flat step (4 % of beam height) located 70 cm from transom. Tests were performed at speeds of 2 m/s, 4 m/s, and 6 m/s, corresponding to Beam Froude numbers (FrB) of 0.9, 1.8, and 2.7. The model weighed 63 kg, with center of gravity (CG) set at 34 % and 36 % of the Length Overall (LOA) from transom. For each CG position, pitch radius of gyration was varied to 23 %, 25 %, and 27 % of the Length Between Perpendiculars (Lₚₚ). Regular waves with a wave height (Hwave) of 5 cm and wave length (λ) of 125 cm were used. Heave, pitch, and acceleration at the CG were measured. Results showed increased speed led to higher heave and CG acceleration. At a CG position of 34 % LOA, a greater pitch radius of gyration ratio to the Lₚₚ (r/Lₚₚ) reduced heave motion. The study also compared hydrodynamic effect of step in calm water and regular waves. This work is vital for optimizing vessel design and improving operational efficiency in maritime applications by enhancing stability through informed weight distribution strategies.]]></description>
      <pubDate>Wed, 08 Apr 2026 13:57:21 GMT</pubDate>
      <guid>https://trid.trb.org/View/2653063</guid>
    </item>
    <item>
      <title>Evaluating classical time-series models ARIMA, SARIMA, ARIMAX, and VAR for wave height forecasting</title>
      <link>https://trid.trb.org/View/2653044</link>
      <description><![CDATA[This study evaluates the suitability of classical time-series models for forecasting significant wave heights using real hydrometeorological observations, with the objective of improving data-driven operational decision-making in ship routing. Accurate short-term wave forecasts can enhance navigational safety, reduce fuel consumption, and minimize emissions. However, many existing Decision Support Systems (DSS) lack integrated forecasting capability to incorporate credible environmental datasets. To address this gap, this research develops a time series–based analytical framework employing four classical forecasting models: Autoregressive Integrated Moving Average (ARIMA), Seasonal ARIMA (SARIMA), ARIMA with Exogenous Variables (ARIMAX), and Vector Autoregressive (VAR). Models were trained and evaluated using observational data from the National Data Buoy Center station 41049 (South Bermuda) under both static and dynamic scenarios. The results indicate limited evidence of stable seasonality in the short-term wave record, and SARIMA showed weaker performance with higher percentage errors. ARIMAX provided comparatively better accuracy due to the influence of exogenous meteorological drivers, while overall findings support the null hypothesis that classical linear models have a restricted capability in capturing rapid wave fluctuations. These insights highlight the practical challenges of applying classical forecasting models to complex sea-state dynamics and underline the need for more advanced or hybrid approaches in DSS development for maritime operations.]]></description>
      <pubDate>Mon, 06 Apr 2026 08:50:03 GMT</pubDate>
      <guid>https://trid.trb.org/View/2653044</guid>
    </item>
    <item>
      <title>TempFusion-Nexus: A multimodal fusion-driven adaptive model for significant wave height forecasting</title>
      <link>https://trid.trb.org/View/2684977</link>
      <description><![CDATA[To enhance wave energy conversion efficiency and address the limited predictive accuracy of significant wave height (SWH) in marine energy applications, we propose a multimodal fusion-driven adaptive model, termed TempFusion-Nexus. The model employs a dual-stream cross-attention block (DSCA-Block) to decouple multi-scale spatiotemporal features, incorporates a convolutional block attention module (CBAM) to enhance the selection of salient discriminative representations, and novelly integrates bidirectional gated recurrent units (BiGRU) with bidirectional long short-term memory networks (BiLSTM) into a hybrid bidirectional recurrent architecture for improved long-range temporal dependency modeling. The experimental dataset consists of hourly SWH records collected from four representative buoy stations across four major ocean basins over a seven-year period (2018-2024). In a 24-hour SWH forecasting task, TempFusion-Nexus reduces the mean absolute error by 5.13% and the root mean square error by 4.87% relative to the second-best baseline. Moreover, the proposed model exhibits superior robustness under extreme wave conditions, achieving R2 improvements of 2.7%-4.0% over the second-best model across the four buoys, thereby validating its capability to accurately characterize complex sea states. This study provides a novel theoretical framework and technical pathway for high-accuracy SWH forecasting in complex marine environments, thereby facilitating the advancement of marine renewable energy toward higher efficiency and precision.]]></description>
      <pubDate>Tue, 31 Mar 2026 16:36:10 GMT</pubDate>
      <guid>https://trid.trb.org/View/2684977</guid>
    </item>
    <item>
      <title>A transformer encoder-only framework for multi-horizon wave forecasting with physical and window-length interpretability</title>
      <link>https://trid.trb.org/View/2685326</link>
      <description><![CDATA[Accurate multi-horizon forecasting of significant wave height (Hₛ) is important for both offshore operations and early warning systems. In this paper, a Transformer encoder-only Hₛ forecasting framework is developed and the impact of input window length on forecasting performance is systematically investigated. Window lengths of 24 h, 48 h, and 72 h are examined through independent hyperparameter optimizations for simultaneous multi-horizon predictions (1–48 h). Results demonstrate that all three window lengths achieve comparable performance. These window lengths allow the model to effectively capture the short-term wind-sea variability and the evolution at the intermediate swell scale, while limiting the influence of weakly informative low-frequency components. The Transformer provides highly accurate predictions in short-term horizons, smoothly degrading in a physically consistent way in the medium-term horizons, and remains stable at longer forecast horizons. Compared to the persistence model, the Transformer performs better at longer horizons. Explainable artificial intelligence investigations indicate that near-term information plays a dominant role, while the model is able to consistently exploit the relevant intermediate temporal variability for wave dynamics. Overall, the effect of input window length is discussed and the capabilities of multi-horizon Transformer-based forecasting are demonstrated for reliable offshore wave prediction.]]></description>
      <pubDate>Fri, 27 Mar 2026 10:14:49 GMT</pubDate>
      <guid>https://trid.trb.org/View/2685326</guid>
    </item>
    <item>
      <title>Shipborne monitoring of significant wave height from BeiDou reflected signal considering satellite elevation and azimuth angles</title>
      <link>https://trid.trb.org/View/2679973</link>
      <description><![CDATA[High-resolution ocean waves are significant for safe navigation and marine disaster monitoring. Hence, a methodology for dynamically monitoring the significant wave height (SWH) of the sea surface using the reflected signal from BeiDou satellites is proposed in this paper. The approach involves employing observables containing elevation and azimuth angles to establish a bivariate field for retrieving SWH, thereby effectively accounting for the coupling among observables induced by variations in observation geometry. In this paper, a receiver for real-time processing of direct signal and reflected signal was carried on shipborne platform. The results of the processing are clearly displayed by delay-Doppler maps (DDMs). According to the dependency of the DDM observables on SWH, two observables [average signal to noise ratio (ASNR) and time delay window (TDW)] are extracted from the generated DDM. Dimensionality reduction and reconstruction are employed in the joint processing of the observables, leading to the establishment of a bivariate correction (BVC) model for retrieving the SWH. Finally, the performance of the proposed BVC model is compared to a single-variable approach, using ERA5 and WAVERYS as reference datasets, and further validated with synchronous satellite altimeter SWH measurements as an independent observation. The inversion results of the experimental route indicate that the minimum root mean square errors (RMSE) of the ERA5 and WAVERYS are 0.167 m and 0.201 m, respectively, and shows overall consistency with the satellite altimeter observations. This work achieves high-resolution monitoring of SWH and provides significant inversion performance.]]></description>
      <pubDate>Wed, 25 Mar 2026 11:44:46 GMT</pubDate>
      <guid>https://trid.trb.org/View/2679973</guid>
    </item>
    <item>
      <title>A parallel hybrid model for very-short-term ship motion prediction: Integrating bidirectional gated recurrent units, gated slot attention mechanism, and informer</title>
      <link>https://trid.trb.org/View/2644035</link>
      <description><![CDATA[Accurate ultra-short-term ship motion prediction is crucial for maritime safety. To solve poor generalization of single models under complex sea conditions caused by ship motion's nonlinear and non-stationary traits, this paper proposes a parallel hybrid model combining Bidirectional Gated Recurrent Unit (BiGRU), Gated Slot Attention (GSA), and Informer, termed GSA-BiGRU-Informer (GBI). It captures global dependencies via the Informer branch and extracts fine-grained temporal features through the BiGRU-GSA branch, integrating multi-scale dynamic information effectively. Ship model test verification shows: 1) Under stable sea conditions, the GBI model demonstrates a significant advantage in prediction accuracy, with the Mean Squared Error (MSE) reduced by up to 32.97 % compared with the Informer model in long-horizon prediction. 2) Under the non-stationary evolution scenarios of multi-condition coupling, the model exhibits strong extrapolation capability, with the coefficient of determination (R²) of single-step prediction approaching 1, and can effectively suppress the error accumulation of multi-step prediction. 3) In cross-sea-state generalization tests, the model trained with data from relatively stable (low wave height) sea conditions achieves an MSE reduction of up to 78.91 % when predicting more severe and unseen (high wave height) complex sea conditions. Results show GBI significantly enhances ultra-short-term ship motion prediction's accuracy, robustness, and generalization ability.]]></description>
      <pubDate>Fri, 20 Mar 2026 14:47:18 GMT</pubDate>
      <guid>https://trid.trb.org/View/2644035</guid>
    </item>
    <item>
      <title>Height3D: A Roadside Visual Framework Based on Height Prediction in Real 3-D Space</title>
      <link>https://trid.trb.org/View/2591221</link>
      <description><![CDATA[In recent years, vision-based roadside 3D object detection has received a great deal of attention, which is an important part of the Intelligent Transportation System (ITS). It extends the perception range beyond the limitations of Autonomous Vehicle (AV) and enhances road safety. While previous work mainly focuses on height prediction in image 2D space, which is limited by the perspective property of near-large and far-small on images, making it difficult for network to understand real dimension of targets in the 3D world. Inspired by this insight, a roadside visual framework Height3D based on height prediction in real 3D space, is proposed. Height Prediction Block (HPB) with explicit height supervision is proposed in real 3D space instead of in image 2D space to predict the height distribution of targets for roadside view transform. Also, Spatial Aware Block (SAB) is used to further extracts spatial context information in BEV space and enhances fine-grained BEV features. The proposed method is applied to two large-scale roadside benchmarks, DAIR-V2X-I and Rope3D. Extensive experiments are performed to verify its effectiveness. The proposed Height3D outperforms the state-of-the-art methods of (1.15, 7.37, 4.03) Average Precision (AP) for Vehicle, Pedestrian and Cyclist categories in 3D object detection task, respectively. Meanwhile, the proposed method achieves 31.55 FPS without using any CUDA or TensorRT acceleration. The code is available at https://github.com/zhangzhang2024/Height3D]]></description>
      <pubDate>Fri, 20 Mar 2026 14:10:52 GMT</pubDate>
      <guid>https://trid.trb.org/View/2591221</guid>
    </item>
    <item>
      <title>Physical modelling of the SeAbacus wave energy converter</title>
      <link>https://trid.trb.org/View/2655822</link>
      <description><![CDATA[The SeAbacus is a new patent for a floating offshore wave attenuator, which essentially consists of a rafted Salter’s Duck. It is modular, suitable also for low-energy seas and for array installation. This paper presents the first physical model tests carried out in the wave tank at the Hydraulic Laboratory of the University of Bologna. The tests focused on the effects of the device shape (by changing the shape of the Salter’s Duck) and of the mooring layout (by testing a Tension Leg Platform, a Catenary Anchor Leg Mooring configuration and a spread mooring system) under wave attacks characterised by different wave height, wave steepness and wave obliquity. The results of the tests highlight the relevance of the shape of the Salter’s Duck and the capability of the device of producing wave energy also in mild seas, provided a moderate wave steepness. Wave obliquity significantly decreases the device pitch motion. The mooring layout affects the device motions because the more rigid the moorings the higher the device pitch due to combined motions of the raft and of the Salter’s Duck. The best compromise between device pitch motions and mooring loads was achieved with the spread mooring system.]]></description>
      <pubDate>Mon, 02 Mar 2026 08:55:13 GMT</pubDate>
      <guid>https://trid.trb.org/View/2655822</guid>
    </item>
    <item>
      <title>A deep learning method for spatiotemporal significant wave height estimation with ship attitude compensation</title>
      <link>https://trid.trb.org/View/2667983</link>
      <description><![CDATA[Significant wave height (SWH) is a critical parameter for ocean state characterization with direct relevance to navigation safety, marine forecasting, and offshore engineering. Traditional SWH sensors are reliable but costly and inflexible. Shipborne vision is cheaper but sensitive to motion and lighting changes. Vision-based techniques offer a low-cost alternative; however, their accuracy degrades under shipborne conditions due to image blur, viewpoint drift, and dynamic disturbances. This paper presents an attitude-aware spatiotemporal deep learning framework for shipborne SWH estimation. Image and attitude encoders are fused with temporal attention to output real-time SWH estimates. The proposed model fuses sequential ocean surface images with synchronized ship attitude data and employs a multi-head self-attention mechanism to jointly capture wave dynamics and compensate for vessel-induced motion. A high-fidelity simulation platform was developed to generate multimodal datasets across 47 sea states, with buoy-derived SWH serving as ground-truth supervision. Experimental results demonstrate stable convergence and high predictive accuracy, with MAE = 0.025 m, RMSE = 0.032 m, and R2 = 0.994. The method outperforms single-frame and image-only baselines by more than 65%, while maintaining robustness under unseen sea states and rough conditions exceeding 2.0 m. These findings confirm that integrating attitude compensation and temporal modeling enables accurate, efficient, and real-time wave height estimation, providing a viable solution for intelligent shipborne perception and maritime safety applications.]]></description>
      <pubDate>Tue, 24 Feb 2026 15:39:13 GMT</pubDate>
      <guid>https://trid.trb.org/View/2667983</guid>
    </item>
  </channel>
</rss>