Monday, November 16, 2026
Science
No Result
View All Result
  • Login
  • HOME
  • SCIENCE NEWS
  • CONTACT US
  • HOME
  • SCIENCE NEWS
  • CONTACT US
No Result
View All Result
Scienmag
No Result
View All Result
Home Science News

Dynamical Evolution Sequence Boosts Decision Tree Ship Motion Predictions to Ten Seconds

November 16, 2026
in Science News
Gavin Prescott
By Gavin Prescott Scienmag Editorial Profile - Ecology and Ecosystem Dynamics
Reading Time: 6 mins read
0
Dynamical Evolution Sequence Boosts Decision Tree Ship Motion Predictions to Ten Seconds

Dynamical Evolution Sequence Boosts Decision Tree Ship Motion Predictions to Ten Seconds

65
SHARES
587
VIEWS
Share on FacebookShare on Twitter
ADVERTISEMENT

Maritime researchers have developed a machine learning framework capable of calculating six-degree-of-freedom ship motion responses in irregular seas in approximately ten seconds, replacing conventional numerical hydrodynamic simulations that typically require hours of processing time per operating condition. Operating ships at sea routinely experience continuous oscillatory motions across all six degrees of freedom due to environmental winds and waves, which can severely constrain critical marine offshore drilling, cargo transfers, and lifting operations. Advance forecasting of vessel behavior in specific wave environments enables operators to identify safe operational time windows and choose appropriate staging areas, but traditional physics-based simulation methods remain too computationally demanding to deploy directly for rapid onboard evaluation. To resolve this operational constraint, the study proposes a surrogate modeling approach driven by a gradient boosting decision tree regression algorithm enhanced through a specialized sampling method known as a dynamical evolution sequence.

Traditional short-term seakeeping predictions rely heavily on physics-based computational tools such as three-dimensional linear potential flow theory or viscous computational fluid dynamics. While linear potential flow calculations determine radiation and diffraction velocity potentials around the hull to establish hydrodynamic matrices, and fluid dynamics solvers resolve local viscous interactions directly, both approaches consume substantial computing time and resources. Commercial seakeeping suites often take considerable time to execute complete parameter sweeps across incident wave frequencies, headings, and floating states, preventing real-time evaluation when weather conditions shift rapidly. Previous research has explored surrogate models, such as Kriging methods, support vector regression, and random forests, to approximate hydrodynamic responses quickly. However, earlier surrogate frameworks frequently suffered from limited training scopes, prolonged data generation periods spanning multiple weeks, or narrow applicability confined to upright hulls and single-degree-of-freedom roll scenarios.

The novel framework addresses the limitations of prior surrogate models by targeting the initial distribution of training samples across the design space, which governs the predictive fidelity of machine learning approximations. Conventional surrogate workflows frequently adopt Latin hypercube sampling to populate multidimensional parameter spaces, yet the resulting distributions can exhibit uneven spacing that restricts regression accuracy. The researchers replaced Latin hypercube sampling with a dynamical evolution sequence, an approach that mathematically conceptualizes candidate sample points as moving celestial bodies within an s-dimensional space. Under this formulation, gravitational interactions governed by Lagrangian mechanics and Hamiltonian variation draw and repel the points until an applied damping force settles them into a stable, highly uniform distribution across the multi-parameter domain. This uniform distribution ensures that subsequent regression algorithms capture nonlinear hydrodynamic variations without localized data clustering.

Before constructing vessel hydrodynamic models, the team evaluated the performance of the dynamical evolution sequence against conventional Latin hypercube sampling using four standard two-dimensional analytical test functions. For each benchmark function, identical sample counts of one hundred points were generated using both sampling techniques to train standard gradient boosting decision trees and enhanced dynamical evolution sequence decision trees. When evaluated against an independent test set of two hundred randomly generated points using the coefficient of determination, root mean square error, and relative average absolute error, the dynamically sampled model demonstrated consistent performance advantages across every benchmark. In the most demanding test case, the dynamical evolution sequence boosted the coefficient of determination by 8.72 percent, reduced root mean square error by 46.32 percent, and lowered relative average absolute error by 38.09 percent, confirming that superior sample uniformity directly translates into heightened regression precision.

Applying this sampling methodology to marine hydrodynamics, the investigators constructed surrogate models for an offshore support vessel operating at zero forward speed in deep water. Under three-dimensional linear potential flow theory, a vessel’s hydrodynamic characteristics vary according to hull geometry, draft, heel angle, trim, wave encounter frequency, and wave heading relative to the bow. Across the vessel’s operational design envelope, the input domain spanned drafts from 4.4 to 6.6 meters, heel angles from negative 5 degrees to positive 5 degrees, and trim variations between negative 0.6 and positive 0.6 meters, alongside incident wave frequencies spanning 0.05 to 1.7 radians per second and directional headings from 0 to 360 degrees. Using the dynamical evolution sequence, the team generated one thousand parameter combinations and performed batch hydrodynamic simulations in the commercial solver Hydrostar, creating a comprehensive database without requiring feature normalization.

To capture all fluid-structure interactions required for six-degree-of-freedom motion prediction, the researchers built a comprehensive network of eighty-four distinct surrogate submodels. These encompass seventy-two individual models representing the full directional matrices of added mass and radiation damping coefficients across all six spatial degrees of freedom, alongside twelve surrogate models characterizing the force amplitudes and phase angles of wave excitation. To optimize the architecture of each individual submodel, Bayesian optimization was conducted over thirty iterations to fine-tune the tree ensemble size between fifty and five hundred iterations and the learning rate parameter between 0.01 and 0.3. For instance, tuning the longitudinal added mass coefficient model converged on an optimal setting of 244 boosting iterations and a learning rate of 0.027 within nine optimization rounds, maximizing the cross-validation score.

The predictive reliability of all eighty-four surrogate models was validated through rigorous fivefold cross-validation, monitoring both the coefficient of determination and normalized root mean square error against the total sample range. The evaluation revealed that seventy-two submodels achieved average determination coefficients exceeding 0.8, with the majority surpassing 0.9, while keeping normalized root mean square errors safely below 10 percent. Only twelve submodels exhibited determination coefficients falling below 0.8, comprising six cross-coupling radiation damping coefficients—specifically those coupling surge with heave, sway with heave, heave with surge, heave with sway, heave with roll, and roll with heave—as well as the phase angles for all six wave excitation force components. Further frequency-segregated analysis showed that the lower scores in radiation damping occurred primarily within an intermediate frequency window between 0.8 and 1.1 radians per second, where sharp nonlinear coefficient fluctuations occurred.

Similarly, the excitation force phase models exhibited strong performance across lower wave frequencies from 0.05 to 0.9 radians per second, but experienced marked dispersion at higher frequencies between 0.9 and 1.7 radians per second. In addition, the yaw excitation moment phase suffered numerical ambiguity because low-frequency values clustered near 0 degrees and 360 degrees, which represent physically identical orientations that the decision tree algorithm treated as distant extremes. However, the researchers emphasized that ocean wave energy is predominantly concentrated within low-frequency regimes rather than high frequencies where phase scatter emerged. Furthermore, for vessels exhibiting port and starboard geometric symmetry operated at minimal heel angles, the six poorly performing cross-coupling damping terms naturally approach zero, exerting negligible influence on actual seakeeping dynamics while yaw deviations are readily managed by automated dynamic positioning systems.

To substantiate overall operational feasibility, the researchers tested the complete prediction framework under two distinct loading states of the offshore support vessel subjected to irregular seas described by the JONSWAP wave spectrum. The sea state evaluated featured a significant wave height of 2.5 meters, peak wave periods ranging from 5 to 14 seconds in one-second increments, and wave headings varied from 0 to 180 degrees in 45-degree increments. Condition 1 simulated a light operational profile with 10 percent fuel and water capacity with an offshore crane extended 15 meters to starboard, whereas Condition 2 evaluated the identical deck configuration under 100 percent liquid capacity. Hydrodynamic coefficients and resulting motion amplitude operators generated by the surrogate models were compared directly against full hydrodynamic calculations executed using the commercial Hydrostar software package.

The comparison confirmed that the surrogate prediction pipeline reproduced six-degree-of-freedom motion amplitudes with high physical accuracy across all wave periods and operational headings. For heave motion under the heavy loading condition, the largest relative error reached 34.6 percent at a short peak period of 5 seconds, yet this corresponded to an absolute dimensional deviation of only 0.1034 meters. In sway motion, the peak relative discrepancy was 16.3 percent under Condition 2 at 5 seconds, representing an absolute error of just 0.0364 meters, while the maximum absolute error observed was 0.1310 meters at 11 seconds under Condition 1. Roll predictions demonstrated a maximum relative error of 31.74 percent at 5 seconds with an absolute discrepancy of 0.207 degrees, and an absolute peak deviation of 3.3113 degrees at 12 seconds representing a 25 percent relative error, satisfying established naval architectural engineering tolerances.

Yaw motions exhibited a peak relative error of 33.27 percent at 11 seconds under the light loading condition, but because the vessel experienced very small total yaw excursions, the corresponding absolute error was restricted to 0.1544 degrees. The motions exhibiting the highest numerical agreement with commercial software—surge, heave, and pitch—directly corresponded to the wave excitation force components that demonstrated the highest surrogate training accuracy, specifically vertical force, pitch moment, and longitudinal force. These close alignments verify that the framework accurately mirrors full-scale commercial solver outputs for primary seakeeping motions while remaining robust against minor numerical deviations in weakly coupled hydrodynamic parameters.

The primary advantage of the validated machine learning framework lies in its dramatic compression of computational turnaround time during practical deployment. Generating the mesh and solving full hydrodynamic coefficients through conventional software chains required approximately 48 minutes per sea state after an initial one-hour manual meshing phase in ANSYS AQWA, culminating in roughly two hours of total processing per condition. By contrast, once the one-time offline investment of 2.5 days for batch sample generation and two hours for training all eighty-four models was completed, the surrogate framework evaluated input operational parameters and delivered complete six-degree-of-freedom motion amplitudes in approximately ten seconds. Although verified specifically on an offshore support vessel, the underlying data-driven modeling methodology operates independently of specific hull geometries, offering a viable foundation for rapid onboard motion prediction across tankers, bulk carriers, and container ships.

Subject of Research: Rapid short-term prediction of six-degree-of-freedom ship motions using machine learning surrogate models

Article Title: A novel framework for rapid short-term prediction of ship motions using an enhanced gradient boosting decision tree with dynamical evolution sequence

Article References: Li, H., Wu, F., Gui, M., Zhu, L., & Zou, K. (2026). A novel framework for rapid short-term prediction of ship motions using an enhanced gradient boosting decision tree with dynamical evolution sequence. Intelligent Marine Technology and Systems, 4(1), Article 32. https://doi.org/10.1007/s44295-026-00119-1

Image Credits: AI Generated

DOI: 10.1007/s44295-026-00119-1

Keywords: ship motion prediction, gradient boosting decision tree, dynamical evolution sequence, six degrees of freedom, hydrodynamic coefficients, surrogate modeling, potential flow theory, offshore support vessel, added mass, radiation damping, wave excitation force, Bayesian optimization, Latin hypercube sampling, seakeeping analysis, Hydrostar, marine technology, JONSWAP spectrum

Cite Scienmag News

Gavin Prescott. (November 16, 2026). Dynamical Evolution Sequence Boosts Decision Tree Ship Motion Predictions to Ten Seconds. Scienmag. https://scienmag.com/dynamical-evolution-sequence-boosts-decision-tree-ship-motion-predictions-to-ten-seconds/

Gavin Prescott. "Dynamical Evolution Sequence Boosts Decision Tree Ship Motion Predictions to Ten Seconds." Scienmag, 16 November 2026, https://scienmag.com/dynamical-evolution-sequence-boosts-decision-tree-ship-motion-predictions-to-ten-seconds/. Accessed 16 November 2026.

Gavin Prescott. "Dynamical Evolution Sequence Boosts Decision Tree Ship Motion Predictions to Ten Seconds." Scienmag. November 16, 2026. https://scienmag.com/dynamical-evolution-sequence-boosts-decision-tree-ship-motion-predictions-to-ten-seconds/

Tags: added massBayesian optimizationdynamical evolution sequencegradient boosting decision treehydrodynamic coefficientsHydrostarJONSWAP spectrumLatin hypercube samplingmarine technologyoffshore support vesselpotential flow theoryradiation dampingseakeeping analysisship motion predictionsix degrees of freedomsurrogate modelingwave excitation force
Share26Tweet16
Previous Post

Robotic Telepresence Guides Common Bile Duct Exploration During Index Cholecystectomy

Next Post

Continuous Pediatric Antibiotic Infusions Reduce Preparation Time Without Altering Perceived Workload

Related Posts

Updated Neuropathy Guidelines Revise Treatment Strategies for CIDP and Guillain-Barré Syndrome
Medicine

Updated Neuropathy Guidelines Revise Treatment Strategies for CIDP and Guillain-Barré Syndrome

November 16, 2026
Multimodality Imaging Transforms Cardiac Amyloidosis Diagnosis and Longitudinal Disease Monitoring
Science News

Multimodality Imaging Transforms Cardiac Amyloidosis Diagnosis and Longitudinal Disease Monitoring

November 16, 2026
Atractylodes Rhizome Sesquiterpenoid Dimers Show PDE4 Inhibition and Anti-Fibrotic Potential
Science News

Atractylodes Rhizome Sesquiterpenoid Dimers Show PDE4 Inhibition and Anti-Fibrotic Potential

November 16, 2026
Molecular Welding Embeds Carbon Nanotubes into Cotton for Washable Smart Textiles
Science News

Molecular Welding Embeds Carbon Nanotubes into Cotton for Washable Smart Textiles

November 16, 2026
Early Myocardial Strain Deterioration Associates With Cardiotoxicity During Immune Checkpoint Therapy
Medicine

Early Myocardial Strain Deterioration Associates With Cardiotoxicity During Immune Checkpoint Therapy

November 16, 2026
Distillation Improves Flight Control Model Accuracy in Cloud-Edge Environments
Technology and Engineering

Distillation Improves Flight Control Model Accuracy in Cloud-Edge Environments

November 16, 2026
Next Post
Continuous Pediatric Antibiotic Infusions Reduce Preparation Time Without Altering Perceived Workload

Continuous Pediatric Antibiotic Infusions Reduce Preparation Time Without Altering Perceived Workload

  • Mothers who receive childcare support from maternal grandparents show more optimized

    Mothers who receive childcare support from maternal grandparents show more parental warmth, finds NTU Singapore study

    27656 shares
    Share 11059 Tweet 6912
  • University of Seville Breaks 120-Year-Old Mystery, Revises a Key Einstein Concept

    1061 shares
    Share 424 Tweet 265
  • Bee body mass, pathogens and local climate influence heat tolerance

    682 shares
    Share 273 Tweet 171
  • Researchers record first-ever images and data of a shark experiencing a boat strike

    546 shares
    Share 218 Tweet 137
  • Groundbreaking Clinical Trial Reveals Lubiprostone Enhances Kidney Function

    531 shares
    Share 212 Tweet 133
Science

Embark on a thrilling journey of discovery with Scienmag.com—your ultimate source for cutting-edge breakthroughs. Immerse yourself in a world where curiosity knows no limits and tomorrow’s possibilities become today’s reality!

RECENT NEWS

  • Updated Neuropathy Guidelines Revise Treatment Strategies for CIDP and Guillain-Barré Syndrome
  • Multimodality Imaging Transforms Cardiac Amyloidosis Diagnosis and Longitudinal Disease Monitoring
  • Atractylodes Rhizome Sesquiterpenoid Dimers Show PDE4 Inhibition and Anti-Fibrotic Potential
  • Molecular Welding Embeds Carbon Nanotubes into Cotton for Washable Smart Textiles

Categories

  • Agriculture
  • Anthropology
  • Archaeology
  • Athmospheric
  • Biology
  • Biotechnology
  • Blog
  • Bussines
  • Cancer
  • Chemistry
  • Climate
  • Earth Science
  • Editorial Policy
  • Marine
  • Mathematics
  • Medicine
  • Pediatry
  • Policy
  • Psychology & Psychiatry
  • Science Education
  • Science News
  • Social Science
  • Space
  • Technology and Engineering

Subscribe to Blog via Email

Enter your email address to subscribe to this blog and receive notifications of new posts by email.

Join 5,150 other subscribers

© 2026 Scienmag - Science Magazine

Welcome Back!

Login to your account below

Forgotten Password?

Retrieve your password

Please enter your username or email address to reset your password.

Log In
No Result
View All Result
  • HOME
  • SCIENCE NEWS
  • CONTACT US

© 2026 Scienmag - Science Magazine

Discover more from Science

Subscribe now to keep reading and get access to the full archive.

Continue reading