MS_114 — Bayesian networks for engineering risk analysis

The other method uses a Limited Memory Influence Diagram (LIMID) where cost and decision nodes are included directly in the Bayesian network, see (Lauritzen and Nilsson 2001). The method is an approximation because the decision nodes only include the direct parents in the policy tables; thus the decisions have to be updated each time new information is entered in the network, in order to make the optimal decisions. The Bayesian network for this case is shown in Figure 2. For this method, the expected costs can be calculated directly from the network, but only approximately, and Monte Carlo sampling is used to find the correct expected costs.