DATA-DRIVEN MODELING OF LONG-TERM RISER RESPONSE TO IRREGULAR WAVES CONSIDERING WAVE DIRECTIONALITY

Prediction of riser response to irregular waves has long been of industrial interest. But numerical simulation is often time consuming due to system nonlinearity and excitation stochasticity, especially in the context of reliability assessment where tons of repetitions are targeted. In this regard, data-driven modelling techniques have been proposed and applied by many researchers in the past decades. However, most established surrogate models were constructed in essence for single sea states and require vessel motions or their derivatives for input. These facts preclude them from fully playing the role of numerical models in prediction. In this paper, a data-driven model of long-term riser response considering wave directionality is developed based on nonlinear autoregressive networks with exogenous inputs (NARX) using fluid velocities. The adoption of fluid velocities stems from wave physics. It not only guarantees the good model performance in conjunction with the ensemble technique, but also allows prior knowledge about ocean environment to apply. Through design of experiments, only a limited number of numerical simulations for riser response to random short-term sea states need to be conducted to prepare data for training networks. The resultant model works well in the most critical case where all long-term wave variables including wave direction come into effect. A floating production system is used as an example for demonstration.