SobolGSA is the general purpose GUI driven global sensitivity analysis and machine learning software. It can be used to compute various sensitivity measures and/or to develop metamodels either from explicitly known models or directly from data. There is a choice of different metamodeling techniques, including Bayesian Sparse Polynomial Chaos Expansion, (Quasi) Random Sampling-High dimensional model representation and Radial Basis Function methods. SobolGSA can be applied to both static and time-dependent models.
Developed metamodels are produced in a form of self-contained MATLAB, Python or C# files which can be used as accurate, reliable and very fast surrogates of the original CPU-expensive full scale models. SobolGSA can be linked directly to MATLAB, Python or other systems and packages. SobolGSA makes use of the most advanced sampling methods: Mersenne twister (Monte Carlo), standard and scrambled Sobol’ sequences (Quasi Monte Carlo). SobolGSA is also a tool for global sensitivity analysis (SA). Global SA methods evaluate the effect of a factor while all other factors are varied as well and thus they account for interactions between variables and do not depend on the choice of a nominal point like local sensitivity analysis methods.
Global SA is used to identify key inputs whose uncertainty most affects the output and the results are used to rank inputs, fix unessential inputs and decrease problem dimensionality (see diagrams below). Over the last decade global SA has gained acceptance among practitioners in the process of model development, calibration and validation, reliability and robustness analysis, decision-making under uncertainty, quality-assurance, and complexity reduction. Global SA has been shown to dramatically improve results in model -based Optimal Experimental Design, process monitoring and control. SobolGSA tutorials present examples of applications of global SA in various setting including the area of dynamical systems. The set of available in SobolGSA global SA techniques include screening methods such as Morris, variance (Sobol’ indices) and derivative based global sensitivity measures. SobolGSA can be downloaded using this link: www.imperial.ac.uk/process-systems-engineering/research/free-software/sobolgsa-software.
Comments and questions can be sent to s.kucherenko@imperial.ac.uk.
