Abstract

In interdependent critical infrastructures (ICIs), a disruptive event can affect multiple system elements and system resilience is greatly dependent on uncertain factors, related to system protection and restoration strategies. In this paper, we perform sensitivity analysis (SA) supported by importance measures to identify the most relevant system parameters. Since a large number of simulations is required for accurate SA under different failure scenarios, the computational burden associated with the analysis may be impractical. To tackle this computational issue, we resort to two different approaches. In the first one, we replace the long-running dynamic equations with a fast-running Artificial Neural Network (ANN) regression model, optimally trained to approximate the response of the original system dynamic equations. In the second approach, we apply an ensemble-based method that aggregates three alternative SA indicators, which allows reducing the number of simulations required by a SA based on only one indicator. The methods are implemented into a case study consisting of interconnected gas and electric power networks. The effectiveness of these two approaches is compared with those obtained by a given data estimation SA approach. The outcomes of the analysis can provide useful insights to the shareholders and decision-makers on how to improve system resilience.

Keywords

Critical infrastructure; System resilience; Importance measure; Sensitivity analysis; Artificial neural networks; Ensemble of methods

Link

Ver publicación