The Application of Different Surrogate Models in Engineering Predictions
Wenqi Chen · Applied and Computational Engineering · 2024
Surrogate models are widely used in engineering predictions to reduce computational costs and improve efficiency, especially in complex systems where direct simulations are time-consuming and expensive. This paper explores the application of four commonly used surrogate models: Response Surface Methodology (RSM), Radial Basis Function (RBF), Kriging, and Support Vector Machine (SVM). Each model's strengths, weaknesses, and suitability for various engineering scenarios are discussed. RSM is shown to be effective for process optimization in systems with moderately nonlinear responses. RBF excels in real-time predictions and nonlinear systems, while Kriging offers high accuracy in spatial data prediction along with uncertainty quantification. SVM demonstrates strong performance in high-dimensional classification tasks. Additionally, this paper addresses strategies for reducing computational costs when applying these models, including the use of efficient optimization techniques. The findings suggest that the selection of an appropriate surrogate model depends heavily on the specific application and the complexity of the system being modeled. Future research could focus on improving the computational efficiency of these models, especially for large datasets.