Correlation and Dependency in Multivariate Process Risk Assessment

Abstract

Process safety and risk assessment are often multidimensional and hence require the joint modeling of several potentially correlated random variables. Any effort to address the correlation among the input variables is important and could improve the accuracy in practical applications of risk assessment models. This paper discusses the problems with correlated variables used in risk assessment and presents a copula-based technique to model dependency among variables to improve uncertainty analysis. Using the copula approach, capturing the dependence structure among different risk factors and estimating the univariate risk marginals can be separated. This advantage simplifies the overall risk estimation for systems with multiple dependent risk sources. The advantage of the copula-based framework for generalization over the traditional correlation analysis technique is demonstrated using a case study. Methods are also presented for copula selection and estimation of the copula parameters.