Development of a novel wave-force prediction model based on deep machine learning algorithms

The future knowledge of the waves and force is indispensable for the model identification and the real-time control of ocean engineering devices. In order to effectively control the motion of the offshore structures in a real-time manner, it is required to have an accurate and efficient prediction of the waves. Machine learning has been widely applied in ocean engineering field as it offers compromise between prediction accuracy and computational cost. The present study focuses on wave-force prediction of offshore structures based on deep machine learning algorithms. A novel wave-force prediction model is proposed, which makes full use of the efficient processing characteristics of Long Short-Term Memory Recurrent Neural Network (LSTM RNN) and Nonlinear Autoregressive Exogenous Feedback Neural Network (NARX FNN) for time series data processing. The relationship between the wave height and the wave height is non-causal and nonlinear which need future wave height knowledge for current wave excitation force. Therefore, The LSTM RNN is firstly utilized for multi-step prediction of the time series of wave elevation. The NARX FNN is used to address the model system identification between the wave heights and the wave force. Then, the LSTM RNN is further applied to predict the future force of offshore structures for the real-time control of the structure motions. After that, the proposed deep machine learning algorithm is utilized for wave-force prediction based on the experimental data obtained in Kelvin Hydrodynamic Laboratory and the optimal horizon can be specified for the test model by comparing the performance of different prediction horizons. The results indicate that LSTM-NARX model can successfully predict the time series of the waves and force.