|
Title |
Optimization of Artificial Surfing Reef Geometry Using Artificial Neural Networks and Genetic Algorithms
|
|
Authors |
엄정우(Jeongwoo Um) ; 심현호(Hyeonho Sim) ; 이재모(Jaemo Lee) ; 장민우(Min Woo Jang) ; 백중철(Joongcheol Paik) ; 염상국(Sang-Guk Yum) ; 박민수(Minsoo Park) |
|
DOI |
https://doi.org/10.11112/jksmi.2026.30.1.113 |
|
Keywords |
인공 리프; 서핑; 데이터 증강; 인공신경망; 유전 알고리즘; 전산 유체해석 Artificial reef; Surfing; Data augmentation; Artificial neural network; Genetic algorithm; Computational fluid dynamics |
|
Abstract |
This study proposes a design optimization framework that integrates an Artificial Neural Network (ANN) surrogate model with a Genetic Algorithm (GA) to mitigate the significant computational costs associated with Computational Fluid Dynamics (CFD) analysis in the geometric design of artificial surfing reefs. The design variables are defined as the slope ratio (S), width (W), height (H). Surfing performance is quantified as a composite score through the normalization and weighted combination of wave height, wave duration, and wave angle. Initially, an ANN was trained using OpenFOAM simulation results from 225 design cases generated via Latin Hypercube Sampling (LHS). To enhance model performance, the dataset was augmented by assigning initial ANN predictions as synthetic data to grid candidate points across the entire design space. Analysis of ten repeated training sessions across seven augmentation cases demonstrated that the average validation R^2 and test R^2 improved to ranges of 0.64-0.67 and 0.82-0.84, respectively. The resulting surrogate models served as fitness evaluators within the GA to identify optimal reef geometries. Finally, the validity of the optimized designs and the overall effectiveness of the framework were verified by comparing OpenFOAM re-analysis results with the surrogate model predictions.
|