Prediction of irregular wave (current)-induced pore water pressure around monopile using machine learning methods
Abstract
The fluid-solid-soil coupling mechanism and seabed stability problems surrounding marine structures are highly complicated. Artificial intelligence models trained by specific samples could make highly fast and reliable pre-dictions on physical phenomena over a broader parameter range. Thus providing some reference for optimization simulation and calculation of fluid-solid-soil interaction phenomena and field disaster forecasting. In order to investigate the applicability of machine learning (deep learning) methods for seabed response prediction around pile foundation, several machine learning models were trained and tested based on a series of collected irregular wave (current) induced oscillation pore water pressure time series in a non-cohesive seabed from the wave -current flume. Moreover, the impact of time-lag correlation and physical laws of pore water pressure response on model errors was further investigated. The results show that the machine learning methods proposed in this study achieve satisfactory performances in predicting the pore water pressure oscillating response at multiple locations within the seabed by using a single point near the surface sand layer. The deep learning models rep-resented by Gated recurrent unit (GRU) accomplish reasonable estimation on the time-frequency characteristics of the pore water pressure oscillation response inside the seabed when accompanied by long-term forecasting capabilities. In addition, this study evaluates the model fitness, prediction bias, and oscillation properties under various random wave spectra, wave periods, wave heights, and wave-current conditions. Parametric analysis results show that the frequency characteristics of the pore pressure dataset and the penetrability of the wave group spectrum become crucial factors affecting the prediction bias of the deep learning model in amplitude attenuation. However, the performance and interpretability of the above model in the poor permeability seabed and long-term forecasting in the field still deserve more investigation.