贝叶斯网络故障诊断的先验概率评估研究

Study on Prior Probability Assessment of Bayesian Network Fault Diagnosis

针对贝叶斯网络中引入专家意见的先验概率准确度问题,对专家意见的误差判别、量化规则、意见综合等方面进行了研究。基于多重插补法对故障样本中的缺失统计数据进行概率插补,利用MATLAB进行概率拟合后得到了完备的先验概率集,提出了一种系统化的专家先验概率评估方法,并应用于大型桥式起重机故障诊断,验证了先验概率评估方法的可靠性。研究结果表明,该评估方法能够实现专家定性意见的精确量化,可以有效减小先验概率误差、提高故障诊断的效率,也可为先验概率赋值合理性的进一步研究提供依据。

Aiming at the problem of the accuracy of prior probability when introducing expert-opinions into Bayesian network, the article studies expert-opinion error determination, quantitative rule, opinion synthesizing, and so on. The probabilities of absent samples are calculated based on multiple interpolation method. A complete set of prior probabilities is obtained by probability fitting using MATLAB. A systematic expert prior probability assessment method is proposed and is applied to the failure diagnosis of large bridge crane, and the reliability of the method is thus verified. Research results indicate that this method is able to quantify the qualitative opinions of experts accurately, reduce the prior probability error effectively and improve the efficiency of failure diagnosis. It is well expected that this work will provide a strong basis for further research on the rationality of prior probability assignment.