ABSTRACT
Bayesian networks (BN) a powerful tool for analyzing system reliability. In BN, each node has a node probability table (NPT). For a child node, NPT is a conditional probability distribution which reflects the dependent relationship between the child node and its parent nodes. With the number of parent nodes of child node increasing gradually, the state combinations will increase exponentially, which will lead to the storage requirement of NPT exceeding the random access memory of the computer. To reduce the storage requirement of NPT for any complex multi-state systems, this chapter proposes a multi-state compression algorithm to reduce the storage requirement of NPT by compressing NPT to be a zip file. It discusses a multi-state inference algorithm to perform the inference of BN by the variable elimination algorithm. Multi-state compression algorithm integrates run-length coding technology, and Lempel-Ziv coding technology. The multi-state inference algorithm is based on the variable elimination algorithm.
