Marvelocity Pdf May 2026

\subsection{Future Work} \begin{enumerate} \item Extension to **fuel‑consumption** prediction via a joint multi‑task network. \item Incorporation of **ship‑maneuvering** dynamics for autonomous docking. \item Open‑source **benchmark suite** for maritime speed prediction (datasets, evaluation scripts). \end{enumerate}

\subsection{Learning the Residual} Define the residual speed: \begin{equation} \Delta V = V_{\text{SOG}} - V_{\text{HM}}, \end{equation} where $V_{\text{SOG}}$ is the measured speed over ground from AIS. We train a Gradient‑Boosted Regression Tree (XGBoost \cite{Chen2016}) to predict $\Delta V$ from the feature vector $\mathbf{x}$: \[ \mathbf{x} = \bigl[\,\underbrace{L, B, D, C_B}_{\text{design}};\, \underbrace{V_{\text{HM}}}_{\text{baseline}};\, \underbrace{U_{10}, \theta_{\text{wind}}}_{\text{wind}};\, \underbrace{H_s, \theta_{\text{wave}}}_{\text{wave}};\, \underbrace{U_c, \theta_{\text{current}}}_{\text{current}}\,\bigr]. \] marvelocity pdf

\section{Methodology} \label{sec:method} \subsection{Data Acquisition} \begin{itemize} \item \textbf{AIS}: 2.3 M messages (2018–2023) from the Global Fishing Watch and MarineTraffic APIs. \item \textbf{Oceanographic Reanalysis}: ERA5 \cite{Hersbach2020} providing 10‑m wind vectors, significant wave height, and surface currents at 0.25° resolution. \item \textbf{Ship Catalog}: Technical specifications (length overall, beam, draft, block coefficient, engine power) extracted from the Lloyd’s Register database. \end{itemize} All timestamps are aligned to UTC and interpolated to a 10‑minute cadence. 15 \% for validation

The final **MarVelocity** prediction is: \begin{equation} V_{\text{MarV}} = V_{\text{HM}} + \hat{\Delta V}(\mathbf{x}). \end{equation} and number of estimators.

\section{Conclusion} \label{sec:conclusion} We presented **MarVelocity**, a hybrid metric that blends classical hydrodynamic resistance modelling with a universal machine‑

\subsection{Training Procedure} \begin{itemize} \item \textbf{Train/validation split}: 70 \% ships for training, 15 \% for validation, 15 \% for test (no ship appears in more than one split). \item \textbf{Hyper‑parameter optimisation}: Bayesian optimisation (Optuna \cite{Akiba2019}) over tree depth, learning rate, and number of estimators. \item \textbf{Loss function}: Mean Absolute Error (MAE) on $\Delta V$. \end{itemize} Model training is performed on a single NVIDIA RTX 4090 GPU (≈ 5 min).