Analysis of Support Vector Regression for Approximation of Complex Engineering Analyses

Abstract

A variety of metamodeling techniques have been developed in the past decade to reduce the computational expense of computer-based analysis and simulation codes. Metamodeling is the process of building a “model of a model” to provide a fast surrogate for a computationally expensive computer code. Common metamodeling techniques include response surface methodology, kriging, radial basis functions, and multivariate adaptive regression splines. In this paper, we investigate support vector regression (SVR) as an alternative technique for approximating complex engineering analyses. The computationally efficient theory behind SVR is reviewed, and SVR approximations are compared against the aforementioned four metamodeling techniques using a test bed of 26 engineering analysis functions. SVR achieves more accurate and more robust function approximations than the four metamodeling techniques, and shows great potential for metamodeling applications, adding to the growing body of promising empirical performance of SVR.

Recent advances in surrogate-based optimization

“…While there is no doubt that SVR is powerful method for prediction, particularly with large, high dimensional data sets, there is only a slim body of literature detailing its use in engineering design. An example among few is Clarke et al [9] who compare SVR with other surrogate modelling methods when predicting a variety of engineering problems. No doubt the lack of published material is partly because the method is still young, but also perhaps because the expense of engineering analyses means that we are rarely faced with the problem of very large data sets.…”

Digital elevation model uncertainty and hazard analysis using a geophysical flow model

“…Instead, output results can be determined by evaluating the emulator. There are indeed many methods-kriging, metamodels, support vector machines-by which such surrogates may be constructed and there exists a body of literature on the topic (Simpson et al 2001;Clarke et al 2005). One often used emulator is the GAuSsian Process (GASP) emulator, which assumes the regression has the form of a trend plus a Gaussian (Kennedy & O'Hagan 2001;O'Hagan 2006;Bayarri et al 2009;Conti & O'Hagan 2010).…”

State-of-the-Art Review: A User’s Guide to the Brave New World of Designing Simulation Experiments

“…Regression software is abundant, so the most common analysis tool is readily available and need not be discussed further. Newer surface-fitting methods are also available, including Kriging, neural nets, radial basis functions, splines, support-vector regression, and wavelets; see Clarke et al (2005) and Antioniadis and Pham (1998). These are metamodel construction methods that can be applied to data collected using a variety of experimental designs and may do a better job fitting certain complex response surfaces.…”