Sensitivity Analysis of a Large-scale Computer Simulation Model - A Practical Guide - Revision 2

Oct 26, 2026

A practical guide presents a 10-step sensitivity analysis procedure for large-scale computer simulation models, emphasizing identification of the most influential input parameters. It uses tornado diagrams as an initial screening tool, while sensitivity plots and spider plots capture nonlinear input-output behavior. Regression metamodels with experimental design support understanding of sensitivities and interactions, and the guidance addresses sampling error, input uncertainty, and rare-event failure probability estimation.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages71

Reading Time2 h 59 min

AccessFree to download

Background, Purpose, and Lexicon

Sensitivity analysis definition and terminology

Key distinctions in model inputs and outputs

South Texas Project Risk-‐Informed GSI-‐191 Evaluation A Practical Guide to Sensitivity Analysis of a Large-scale Computer Simulation Model Document: STP-‐RIGSI191-‐ARAI.01 Revision: 2.1 Date: December 9, 2013 Prepared by: David Morton, The University of Texas at Austin Jeremy Tejada, The University of Texas at Austin Reviewed by: Ernie J. Kee, South Texas Project Zahra Mohaghegh, University of Illinois at Urbana-‐Champaign Seyed A. Reihani, University of Illinois at Urbana-‐Champaign Ron Holloway, Wolf Creek Nuclear Operating Corporation Craig Sellers, Enercon A Practical Guide to Sensitivity Analysis of a Large-scale Computer Simulation Model David Morton and Jeremy Tejada The University of Texas at Austin Abstract We describe a 10-step sensitivity analysis procedure that applies to a large-scale computer simulation model. We propose using tornado diagrams as the initial tool for identifying the input parameters to which the simulation's outputs are most sensitive. Sensitivity plots and spider plots complement tornado diagrams by capturing nonlinear responses in outputs to changes in inputs. Regression metamodels, and associated experimental design, help understand sensitivities to, and interactions between, input parameters. Our motivating model from GSI-191 has a number of distinguishing features: (i) The model is large in scale in that it has a high-dimensional vector of inputs; (ii) Some of the model's inputs are governed by probability distributions; (iii) A key output of the model is the probability of system failure|a rare event; (iv) The model's outputs require estimation by Monte Carlo sampling, including the use of variance reduction techniques associated with rare-event simulation; (v) It is computationally expensive to obtain precise estimates of the failure probability; (vi) We seek to propagate key uncertainties on model inputs to obtain distributional characteristics of the model's outputs; and,(vii) The overall model involves a loose coupling between a physics-based stochastic simulation sub-model anda logic-based PRA sub-model via multiple initiating events in the latter sub-model. We review a subset of a much larger literature, guided by the need to have a practical approach to sensitivity analysis for a computer simulation model with these characteristics. We illustrate our proposed 10-step procedure on a simple example of a simulation model for system reliability. Important themes repeat throughout our recommendations, including the use of common random numbers to reduce variability and smooth output analysis, a focus on assessing di􀀋erences between two model con􀀌gurations, and proper characterization of both sampling error and uncertainties on input parameters. 1 Background, Purpose, and Lexicon Paraphrasing Kleijnen [9], Sensitivity analysis, of a computer simulation model, estimates changes in the model's outputs with respect to changes in the model's inputs. In this report, we review approaches to sensitivity analysis of a computer simulation model, focusing on speci􀀌c approaches that we see as practically viable in the context of resolving GSI-191 through a risk-informed approach. And, we propose a 10-step sensitivity analysis procedure. Before proceeding, some remarks on our characterization of sensitivity analysis, adopted from Kleijnen, are in order. 1. We deliberately use the verb estimates because the \true" value of our model's outputs (e.g. , probability of system failure) cannot be computed. Rather this output must be estimated using Monte Carlo sampling in our implementation via a computer simulation model. 2. For the moment, we are purposefully vague about how to make changes in the model's inputs because this is a key distinguisher of di􀀋erent approaches to sensitivity analysis. We explore this issue in some detail in this report. 3. Another key distinguisher of approaches to sensitivity analysis is the manner in which we quantitatively and

FAQs

What is the core idea of the sensitivity analysis procedure in the guide?

The procedure estimates how simulation outputs change with respect to changes in simulation inputs, using a structured sequence of steps and visualization/analysis techniques.

How does the guide identify which input parameters the model outputs are most sensitive to?

It proposes tornado diagrams first to screen the inputs most affecting outputs, then supplements them with sensitivity and spider plots to capture nonlinear responses.

Why are regression metamodels and experimental design included?

They help quantify sensitivities and interactions among input parameters, improving understanding beyond visualization and supporting uncertainty-focused analysis.

Oct 26, 2026

Upload by Genevieve