基于Gibbs抽样的诊断贝叶斯网络模型评估方法

Evaluation Method of Diagnostic Bayesian Networks Model Based on Gibbs Sampling

贝叶斯网络模型应用到故障诊断之前,必需对模型的正确性、故障诊断率、故障隔离率等进行全面的分析评估。通常的评价方法是用基于标准案例的测试程序对模型进行测试,但这些案例毕竟是有限的,不能穷尽所有情况。提出一种基于Gibbs抽样的诊断模型评估方法,不需要特殊的诊断案例,该方法采用等概率故障注入算法,自己产生测试案例,仿真故障传播和故障诊断过程,能够保证对系统的全面覆盖测试,对诊断模型进行全面评估。

The validity, faults diagnostic rate and faults isolated rate of the model must be evaluated all before Bayesian Networks model is used for faults diagnosis. General evaluation way is to use test program based on standard cases to test the model, but the cases are limited and do not include all instance. An evaluation method of diagnostic model based on Gibbs sampling was proposed, it did not need special diagnostic cases, adopted faults injecting algorithm with equal probability, and the diagnostic cases was produced by this evaluation method itself. Process of faults propagation and diagnosis was simulated, and the diagnostic system was guaranteed to be tested overally, and the diagnostic model could be evaluated roundly.