Subset Simulation Method

Overview

The Subset Simulation method (SSM) is a robust simulation technique that transforms a rare event into a sequence of multiple intermediate failure events with larger probabilities and efficiently approximates the probability of the mentioned rare event. That is, the failure event $\Omega_{f} = \{\vec{X}: g(\vec{X}) \leq 0\}$ is expressed as a union of $M$ nested intermediate events $\Omega_{f_{1}}$, $\dots$, $\Omega_{f_{M}}$, such that $\Omega_{f_{M}} \subset \dots \subset \Omega_{f_{1}}$ and $\Omega_{f} = \cap_{i = 1}^{M} \Omega_{f_{i}}$. The intermediate failure events are defined as $\Omega_{f_{i}} = \{\vec{X}: g(\vec{X}) \leq b_{i}\}$, where $b_{1} > \dots > b_{M} = 0$ are non-negative thresholds selected such that each conditional probability $P(\Omega_{f_{i + 1}} | \Omega_{f_{i}})$ equals a target conditional probability $P_{0}$.

\[P_{f} = P(\Omega_{f}) = P(\cap_{i = 1}^{M} \Omega_{f_{i}}) = P(\Omega_{f_{1}}) \prod_{i = 1}^{M - 1} P(\Omega_{f_{i + 1}} | \Omega_{f_{i}})\]

The threshold for the first failure event $b_{1}$ is computed using the Monte Carlo simulations. The thresholds for the following intermediate failure events $b_{i}$ are computed using the Monte Carlo Markov Chain samples generated from the conditional probability density functions $f_{\vec{X}}(\vec{x} | \Omega_{f_{i}})$.

Current implementation of the SSM follows Au and Beck (2001).

API

solve(Problem::ReliabilityProblem, AnalysisMethod::SSM)

SSM <: AbstractReliabililyAnalysisMethod

SSMCache