Wave excitation force forecasting using neural networks

Many wave energy conversion applications require future knowledge or forecasting of the wave excitation force values. Most wave energy converter (WEC) control strategies need to forecast the time-series excitation force for wave energy harvesting maximization. The main aim of this study is to forecast the wave excitation force experiences by a two-body heaving point absorber WEC (as a case study) using three forecasting neural network methods. The wave excitation force is calculated based on the hydrodynamic characteristics of the considered device in the frequency and time-domain simulations. The nonlinear autoregressive neural (NAR) network, group method of data handling (GMDH) network, and Long Short-Term Memory (LSTM) network are fitted to the wave elevation time-series data to forecast the future values of the excitation force. The performance of the examined methods is evaluated for various irregular incident waves that are created using different wave spectrums. Moreover, sensitivity analyses to sampling period and algorithms input parameters are performed to investigate the accuracy and generalizability of the discussed methods at different conditions. Each data set is divided into training and test sets. The results show that the performance of all discussed methods is satisfactory in training data sets and short-term ahead forecasting, but the NAR network method provides a relatively better agreement with test target data compared to other methods.