We propose the use of techniques from Machine Learning for theprediction of tidal currents. The classical methodology of harmonicanalysis is widely used in the prediction of tidal currentsand computer algorithms based on the method have been used fordecades for the purpose. The approach determines parameters byminimizing the difference between the raw data and model outputusing the least squares optimization approach. However, althoughthe approach is considered to be state-of-the-art, it possesses severaldrawbacks that can lead to significant prediction errors, especiallyat locations of fast tidal currents and ’noisy’ tidal signal.In general, careful selection of tidal constituents is required in orderto achieve good predictions, and the underlying assumptionof stationarity in time can restrict the applicability of the methodto particular situations. There is a need for principled approacheswhich can handle uncertainty and accommodate noise in the data.In this work, we use Gaussian process, a Bayesian non-parametrictechnique, to predict tidal currents. The overall objective is to takeadvantage of the recent progress in machine learning to constructa robust yet efficient algorithm. The development can specificallybenefit the tidal energy community, aiming to harness energyfrom location of fast tidal currents.
展开▼
