Uncertainty Analysis in Software Reliability Modeling by Bayesian Analysis with Maximum-Entropy Principle

摘要: In software reliability modeling, the parameters of model are typically estimated from test data corresponding component. However, widely used point estimators subject to random variations in data, resulting uncertainties these parameters. Ignoring parameter uncertainty can result grossly underestimating total system reliability. This paper attempts study and quantify modeling a single component with correlated large numerous components. Another characteristic challenge testing is lack available failure test, which often makes difficult. poses bigger analysis modeling. To overcome this challenge, proposes utilizing experts' opinions historical previous projects complement small number observations uncertainties. done by combining maximum-entropy principle (MEP) into Bayesian approach. further considers at level, contains multiple components, each its respective model/parameter/ uncertainty, using Monte Carlo Some examples different approaches (NHPP, Markov, Graph theory) illustrated show generality effectiveness proposed Furthermore, we illustrate how approach for considering various components improves large-scale model.