Predicting Ocean Energy Harvesting Dynamics using LSTM Deep Learning Network


Creative Commons License

Alan A. R. , Bayındır C.

The 34th Asian-Pacific Technical Exchange and Advisory Meeting on Marine Structures ITU TEAM2020 CONFERENCE, İstanbul, Turkey, 6 - 08 December 2021, pp.1-8

  • Publication Type: Conference Paper / Full Text
  • City: İstanbul
  • Country: Turkey
  • Page Numbers: pp.1-8

Abstract

Ocean waves and currents are among the most beneficial forms of renewable energy. The literature on this subject is vast. Also, the prediction of the amount of energy available in the marine environment is one of the long-studied topics. Various numerical models, weather forecast methods, and even artificial intelligence techniques are studied in the literature. However, the prediction of the performance of the ocean wave and tidal energy harvesting devices in the marine environment is not as commonly studied. With this motivation, first of all, the temporal dynamics of a buoy type power take-off (PTO) device in the random sea environment are modeled. To model the random sea environment, a spectral approach is followed where the FFT routines are employed for the efficient modeling of the multi-spectral ocean wave field. The dynamics of a buoy-type PTO device in this multi-spectral environment are modeled using the transfer function approach. Secondly, the prediction of the heave velocities and converter power time-series are achieved via the Long Short-Term Memory (LSTM) deep learning network. It is shown that the LSTM can predict the device dynamics and the instantaneous converter power. Our findings indicate the LSTM based deep learning or similar approaches can be adopted for such predictions, which are expected to be useful in optimizing the power conversion, predicting resonant dynamics, and avoiding excessive loading and structural damage of the converter.