| Autor: |
Ankang Cheng, Ying Min Low |
| Rok vydání: |
2022 |
| Zdroj: |
Volume 2: Structures, Safety, and Reliability. |
| Popis: |
Metamodeling (also referred to as surrogate modeling) has attracted increasing attention from researchers in the field of offshore engineering for its advantage in computation efficiency. Among the many metamodeling techniques, artificial neural network (ANN) has long been recognized as one of the most versatile and powerful. In this paper, metamodels of ANN type are constructed to predict time-domain dynamic responses of a steel lazy wave riser (SLWR) in irregular waves based on wave elevations. The work is first carried out with short-term variabilities in waves considered only. The capabilities of dynamic recurrent neural networks with exogenous inputs on the basis of nonlinear autoregression (NARX) is carefully studied. The prediction accuracy is evaluated by the coefficient of determination and the root mean squared error, and the possible overfitting is prevented using k-fold cross-validation. Then the work is expanded to include long-term variabilities in waves into consideration. The time series data of structural dynamic responses are generated from finite element method (FEM) based numerical simulations and are further used to train and validate the ANN. The results obtained suggest that time-domain dynamic responses of SLWRs subject to irregular waves can be effectively predicted by ANN type metamodels directly from wave elevations. |
| Databáze: |
OpenAIRE |
| Externí odkaz: |
|
