Multidimensional Risk Analysis

Abstract

Accidents involve critical consequences that require an appropriate and efficient form of risk management. A multidimensional risk analysis allows a broader view. MCDM/A approaches enable more consistent decision-making, taking into account the DM's rationality (compensatory or non-compensatory), DM's behavior regarding risk (prone, neutral or averse) and the uncertainties inherent in the risk context. This chapter presents numerical applications illustrating the use of multicriteria models in two different contexts: a natural gas pipeline and an underground electricity distribution system. Two different MCDM/A approaches are considered: MAUT (Multiattribute Utility Theory) and the ELECTRE TRI outranking method. In the numerical applications, MCDM/A approach steps for building decision models are presented: identifying hazard scenarios, estimating the set of payoffs, eliciting the MAU function (Multi-attribute Utility function), computing the probability function of consequences and estimating multidimensional risk. Loss functions are introduced in the models to calculate the probability distribution functions over the multiple criteria such as impact on humans, and environmental and financial losses. Therefore, Decision Theory concepts are applied to estimate risk in industrial plants and modes of transportation. Finally, other decision problems related to multidimensional risk analysis, using MCDM/A, are considered in different contexts, such as: power electricity systems, natural hazards, risk analysis on counter-terrorism, nuclear power plant. Justifying the Use of the Multidimensional RiskThe perceived level of risk is directly linked to the perceived intensity of consequences to people and society as well as to issues related to the level of probability. These consequences are multidimensional and are associated to the objectives, represented by criteria and can be approached with an MCDM/A or a multiobjective method (see Chap. 2). Many studies show that using a single dimension of risk may not be realistic (Morgan et al.