ABSTRACT
The sensitivity analysis approach is based on the novel ideas discussed initially in (Taflanidis 2009). Foundation of this methodology is the definition of an auxiliary probability density function that is proportional to the integrand of the risk integral
π( ,Z) ( ) (Z) ( ) (Z)= ∝h p( ,Z) p H
h p( ,Z) p (2)
where ∝ denotes proportionality. Comparison between π(θ,Ζ) and the prior probability model p(θ) p(Ζ) expresses the sensitivity of the performance measure to the various model parameters; bigger discrepancies between these distribution indicate greater importance in affecting the system performance, since they ultimately correspond to higher values for h(θ,Ζ). This idea can implemented to a specific group of parameters (or even to a single parameter), denoted y herein, by looking at the marginal distribution π(y). Comparison between this distribution π(y) and the prior distribution p(y) expresses the probabilistic sensitivity with respect to y. Uncertainty in all other model parameters and stochastic excitation is explicitly considered by appropriate integration of the joint probability distribution π(θ,Ζ) to calculate the marginal probability distribution π(y). Different grouping of the parameters, for defining y, will provide information for each group separately. In this process the stochastic sequence Ζ should be always considered as a separate group since it represents a fundamentally different type of uncertainty in the hazard description.
