0
Research Papers

Sensitivity Analysis of a Bayesian Network

[+] Author and Article Information
Chenzhao Li

Department of Civil
and Environmental Engineering,
Vanderbilt University,
2301 Vanderbilt Place PMB 351826,
Nashville, TN 37235
e-mail: chenzhao.li@vandebilt.edu

Sankaran Mahadevan

Department of Civil
and Environmental Engineering,
Vanderbilt University,
2301 Vanderbilt Place PMB 351826,
Nashville, TN 37235
e-mail: sankaran.mahadevan@vandebilt.edu

1Corresponding author.

Manuscript received August 31, 2016; final manuscript received January 27, 2017; published online September 7, 2017. Assoc. Editor: Yan Wang.

ASME J. Risk Uncertainty Part B 4(1), 011003 (Sep 07, 2017) (10 pages) Paper No: RISK-16-1112; doi: 10.1115/1.4037454 History: Received August 31, 2016; Revised January 27, 2017

In a Bayesian network (BN), how a node of interest is affected by the observation at another node is a main concern, especially in backward inference. This challenge necessitates the proposed global sensitivity analysis (GSA) for BN, which calculates the Sobol’ sensitivity index to quantify the contribution of an observation node toward the uncertainty of the node of interest. In backward inference, a low sensitivity index indicates that the observation cannot reduce the uncertainty of the node of interest, so that a more appropriate observation node providing higher sensitivity index should be measured. This GSA for BN confronts two challenges. First, the computation of the Sobol’ index requires a deterministic function while the BN is a stochastic model. This paper uses an auxiliary variable method to convert the path between two nodes in the BN to a deterministic function, thus making the Sobol’ index computation feasible. Second, the computation of the Sobol’ index can be expensive, especially if the model inputs are correlated, which is common in a BN. This paper uses an efficient algorithm proposed by the authors to directly estimate the Sobol’ index from input–output samples of the prior distribution of the BN, thus making the proposed GSA for BN computationally affordable. This paper also extends this algorithm so that the uncertainty reduction of the node of interest at given observation value can be estimated. This estimate purely uses the prior distribution samples, thus providing quantitative guidance for effective observation and updating.

Copyright © 2018 by ASME
Your Session has timed out. Please sign back in to continue.

References

Figures

Grahic Jump Location
Fig. 1

Possible Bayesian inference results: (a) desired inference and (b) undesired inference

Grahic Jump Location
Fig. 2

Auxiliary variable for a CPD

Grahic Jump Location
Fig. 3

Auxiliary variable for a BN

Grahic Jump Location
Fig. 4

Deterministic function for the path X1→XN

Grahic Jump Location
Fig. 6

Equally probable interval

Grahic Jump Location
Fig. 7

Steps to realize the proposed algorithm

Grahic Jump Location
Fig. 8

Beam with mass–spring–damper

Grahic Jump Location
Fig. 10

Posterior distributions at observation value of A3=4500

Grahic Jump Location
Fig. 11

Dynamic BN of example 2

Grahic Jump Location
Fig. 12

Observations, example 2

Grahic Jump Location
Fig. 13

Posterior distribution of state variables

Grahic Jump Location
Fig. 14

VRR of the state variables

Tables

Errata

Discussions

Some tools below are only available to our subscribers or users with an online account.

Related Content

Customize your page view by dragging and repositioning the boxes below.

Related Journal Articles
Articles from Part A: Civil Engineering
Related eBook Content
Topic Collections

Sorry! You do not have access to this content. For assistance or to subscribe, please contact us:

  • TELEPHONE: 1-800-843-2763 (Toll-free in the USA)
  • EMAIL: asmedigitalcollection@asme.org
Sign In