A Machine Learning Approach for Tidal Flow Classification

Sergio Lopez Dubon*, Miguel Angel Valdivia Camacho, Raymond Lam, Brian Sellar

*Corresponding author for this work

Research output: Working paperPreprint

Abstract / Description of output

Due to the rapid growth of the tidal stream energy sector, turbine developers are incentivised to design and manufacture larger tidal blades, which require fatigue testing before deployment. While the recent advances in computational methods allow efficient prediction of unsteady blade loads, it is difficult to consider all flow conditions throughout the turbine lifetime for predicting blade loads that serve as input to a blade fatigue test. To solve this, the following paper proposes a data-driven methodology to classify flow velocities measured at the European Marine Energy Centre (EMEC) tidal test site. A series of wavelet transforms were applied to the measurements to obtain wavelet spectra, giving the flow velocities time series and frequency patterns. These wavelet spectra are treated as an unsupervised image (2D vector) classification problem addressed using neural network autoencoders mixed with clustering algorithms. The reconstructed flow velocities from the centroids of the clusters can then be used as input for load prediction and fatigue testing.
Original languageEnglish
PublisherSocial Science Research Network (SSRN)
Number of pages24
Publication statusSubmitted - 12 May 2023

Keywords / Materials (for Non-textual outputs)

  • neuronal networks
  • Classification
  • Fatigue
  • Tidal Flow
  • Autoencoder


Dive into the research topics of 'A Machine Learning Approach for Tidal Flow Classification'. Together they form a unique fingerprint.

Cite this