Entropy Based Method for Identification of Leading Risk Indicators

In this paper, a method is proposed for identifying leading risk indicators for prioritization of monitoring system parameters, and maintaining a desired level of reliability. The proposed method is applicable to risk and reliability models that are constructed using the Bayesian Belief Network (BBN) modelling framework. To identify leading risk indicators among the nodes of a BBN risk model, we propose a two-step method on the basis of concepts from Information Theory. The first-step metric applies the information entropy of system being in failed State, which measures the amount of information required to determine the system state as failed. In the second step the prior and posterior probability of the node-states are incorporated. By sorting the metrics associated with all the nodes and their states we can identify the most influential node-state pairs of the network. Such nodes are candidates as risk indicators. The paper provides an illustrative example to detail out and provide insights into the meaning and application of risk leading indicators. Moreover, the method is applied to a real industrial BBN model developed for the risk analysis of the corrosion of oil and gas pipelines.