基于神经网络的二阶波-流和结构物相互作用OA北大核心CSTPCD

Wave diffraction by a horizontal semi-circle in a uniform current was evaluated based on the potential flow theory through the time-domain second-order approach.The time histories of first-and second-order wave and hydrodynamic force and their amplitudes or peaks of their superpositions were obtained and taken as the training set.The back propagation neutral network with Levenberg-Marquardt algorithm(LM-BP)was employed to make fast and accurate prediction of the wave and force including their histories and amplitudes or peaks at any wave frequency and current speed.Research results show that higher accurate results on wave and force amplitudes or peaks with small sample can be obtained using a single hidden layer with fewer neurons,while two hidden layers at least with more neurons are required for predicting the histories of first-and second-order waves and forces with very large sample.