Reproduce “Learning Response-Statistic Shifts and Parametric Roll Episodes from Wave--Vessel Time Series via LSTM Functional Models” on OpenResearch

Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.

Universal app for Apple silicon and Intel. Requires macOS 11 or later.

About this paper

Jose del Aguila Ferrandis

An LSTM functional model is developed to map incident wave histories to vessel motion histories, designed to predict parametric roll episodes and their associated statistical shifts. This model successfully reproduces the temporal evolution of these large-amplitude events and accurately represents changes in roll probability density functions, including heavy-tail behavior, particularly when trained with tail-emphasizing loss functions.

Key insights

  • Stacked LSTM networks can learn complex, nonlinear causal mappings from raw wave histories to vessel motions, reproducing both parametric roll episodes and associated statistical shifts without explicit conditioning on sea-state parameters.
  • The choice of loss function significantly impacts the fidelity of predictions in risk-critical regions; relative-entropy-inspired and amplitude-weighted objectives substantially improve the preservation of heavy-tail behavior in roll probability density functions compared to standard MSE.
  • The functional model can infer changes in underlying dynamics and statistical behavior directly from the raw wave elevation history, demonstrating robust learning capabilities across different forcing conditions.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.