Validation of Human Centred Bayesian Networks - Case Study on a Cable Cut of an Export Cable of an Offshore Wind farm

Abstract

Bayesian networks are a commonly used method in the risk and reliability domain to assess the likelihood of certain scenarios or the resilience of an infrastructure. This study is conducted based on a literature review of Bayesian networks in the maritime domain. Even though there are about 78 journal papers published on Bayesian networks in the maritime domain between 2018-2022, it is challenging to gather suitable data in the development process. Further, either a-priori or conditional probabilities are necessary depending on the node. Especially acquiring conditional probabilities is challenging due to that fact that for every state of the parent node(s) a conditional probability needs to be defined. In this work, we investigate validation methods for Bayesian networks. The selected methods have been chosen based on a careful review of the state of the art and encompass benchmark exercises, formal walkthroughs, qualitative feature tests and sensitivity analysis. The validation methods formal walk through and qualitative feature are then discussed at hand of a case study. For this purpose, a Bayesian network has been developed for a maritime security scenario. The results indicate that the validation process improves the design as well as the parametrization of Bayesian networks.