Even small errors in the estimation of the objective function can cause a redistribution of individuals on the Pareto frontier, which can lead to divergence. High accuracy of the surrogate approximation is especially important for the multi-objective optimization task. The model has approximately two million parameters and consists of convolutional layers with 3x3 kernels, upsampling, max pooling, and skip connectionsįigure 4: Proposed surrogate-assisted evolutionary approach consisting of three main stages: random initialization, surrogate preparation, evolutionary stage. Outputs of the CNN is an output mask and wave heights at important points. The first output is a key factor in the model’s understanding of dependence between wave height in target points and breakwater configuration since input masks have no information about the field of sea waves. Only the values from the second output are needed to solve the initial problem and it is possible to think that the first output is useless. The second output is obtained from the first by taking a linear transformation from the intensity of the pixels corresponding to the important points. The first is designed to reconstruct the entire field of sea waves in the whole water area using three-dimensional array parameterization, the second obtain the heights of sea waves in important points. ![]() The main feature of the surrogate model is that it has two types of outputs. Authors implement kriging interpolation as a surrogate model to replace the computationally expensive CFD model. To analyze the ecological situation and to study hydrodynamic characteristics of eco-seawalls researchers used SDM (species distribution modeling) and CFD (computational fluid dynamic) models respectively. A multi-objective optimization problem was formulated, in which compromise between engineering cost and ecological protection was considered. In Īuthors developed computational framework SEO for eco-seawalls design. In wave protection structures optimization tasks surrogate modeling is also used. ![]() This approach has been tested on Monterey Bay and demonstrated good agreement with a numerical model with a 4,000 times improvement in computational speed. Also, a support vector machine was used to predict the wave period. , researchers utilized multi-layer perceptron based on numerical model outputs to solve the regression task and predict the wave height. ![]() This approach showed a good performance, the results were compared with numerical prediction. The surrogate model was trained on historical data from 1989 to 2009 and evaluated for the following year (2010). The output of the physics-based model (significant wave height, mean wave direction, mean zero-crossing period, and peak wave period) is used as input features to the random forest. In Īuthors apply a random forest model to predict wave conditions in significant points (buoy locations) across the considered domain (Cornwall, South West UK). Surrogate models are widely used in coastal engineering to avoid the unnecessary runs of metocean models.
0 Comments
Leave a Reply. |
AuthorWrite something about yourself. No need to be fancy, just an overview. ArchivesCategories |