Unsupervised Machine Learning Method for Stationary Sea State Clustering Based on Gaussian Wave Surface Elevation

Paper i proceeding, 2025

This study proposes a novel unsupervised machine learning framework to identify stationary sea states based on Gaussian wave surface elevation simulated using the Pierson-Moskowitz spectrum for fully developed seas. The proposed methodology consists of two main steps. First, the wavelet scattering transform is employed to extract features from the wave elevation time series, while multiple change point detection algorithms are used to estimate the expected number of stationary sea states (clusters). Next, a comparative analysis is conducted using various clustering algorithms approaches to determine their effectiveness in this specific application.