Study of the Performance of Deep Learning Methods Used to Predict Tidal Current Movement

Abstract

To predict tidal current movement accurately is essential in the process of tidal energy development. However, the existing methods have limits to meet the need for accuracy. Recently, artificial intelligence technology has been widely applied to solve this problem. In this paper, a tidal current prediction model combining numerical simulation with deep learning methods is proposed. It adopts three deep learning algorithms for comparative investigations: multilayer perceptron (MLP), long-short term memory (LSTM) and attention-ResNet neural network (AR-ANN). The numerical simulation was carried out using ROMS, and the observation collected in the Zhoushan region were used to validate the results. Compared with the numerical simulations, deep learning methods can increase the original correlation coefficient from 0.4 to over 0.8. In comparison, the AR-ANN model shows excellent performance in both the meridional and zonal components. This advantage of deep learning algorithms is extended in the tidal energy resource assessment process, with MLP, LSTM and AR-ANN models reducing the root mean square error by 32.9%, 34.4% and 42%, respectively. The new method can be used to accurately predict the hydrodynamic of tidal flow in the process of tidal energy extraction, which contributes to determine the suitable location for energy generation and tidal turbine design.