Research on simulation methods of evaluation for diagnostic Bayesian networks

Abstract

Bayesian Networks that based on probability inference is proposed to solve problems of uncertainty and imperfection. It has more advantages to solve faults caused by uncertainty and relevancy of complex devices. Diagnostic models of Bayesian Networks must be evaluated roundly before used for diagnosis. The usual way to evaluate diagnostic models is using standard cases to test the model, but the cases are limited and the quality of these cases depend on their source, and these cases could not include all instances. Based on sample means of Monte Carlo, an algorithm of evaluation for diagnostic models is proposed this article; this algorithm does not need special diagnostic cases. Faults injecting algorithm with equal probability of every components are adopted, test cases are produced by this algorithm of system itself, and course of faults propagation is simulated by this algorithm of system too. This algorithm could test diagnostic models roundly, and make overall evaluation of diagnostic models.