Surrogate modeling of time-dependent metocean conditions during hurricanes

Metocean conditions during hurricanes are defined by multiple parameters (e.g., significant wave height and surge height) that vary in time with auto- cross-correlation. In many cases, the nature of variation these characteristics is important to design assess risk offshore structures, but a persistent problem measurements sparse history simulations using metocean models computationally onerous. Surrogate modeling an appealing approach ease computational burden modeling; however, time-dependency surrogate challenging because at one instant dependent on not only also previous instances. this paper, time-dependent height, peak period, direction, storm explored database site. Three types models, including Kriging, multilayer perceptron (MLP), recurrent neural network gated unit (RNN-GRU), evaluated, two different structures considered for Kriging model training set sizes MLP model, resulting total five evaluated paper. The performance compared terms accuracy sensitivity toward hyperparameters, RNN-GRU demonstrated have extraordinary prediction context.