基于分布式贝叶斯网络的多故障诊断方法研究
2017-02-16郭文强彭程张宝嵘夏令君佘金龙
郭文强+彭程+张宝嵘+夏令君+佘金龙



摘 要: 针对复杂系统存在的不确定性、多故障以及传统贝叶斯网络诊断实时性差等问题,提出一种基于分布式贝叶斯网络的故障诊断方法。该故障诊断方法将大型、复杂系统故障诊断模型抽象为贝叶斯网络模型,并将其分解为若干贝叶斯网络子系统,基于消息传播机制完成多个子系统局部推理以及子系统间重叠子域紧凑的消息传播,实现分布式贝叶斯网络的故障推理与诊断。实验结果表明,该故障诊断方法可在复杂、不确定性系统中完成单故障和多故障推理、诊断任务,与传统贝叶斯网络故障诊断方法相比,该方法在推理速度上的优势尤为突出,具有广泛的应用前景。
关键词: 贝叶斯网络; 分布式推理; 多故障诊断; 复杂系统
中图分类号: TN711?34; TP391.4 文献标识码: A 文章编号: 1004?373X(2017)01?0119?06
Abstract: Since the complex system has the uncertain and multi?fault problems, and the traditional Bayesian network diagnosis has poor real?time performance, a fault diagnosis approach based on distributed Bayesian network is proposed. The large?scale and complex system fault diagnosis model is abstracted as the Bayesian network model with the fault diagnosis method. The Bayesian network model is decomposed into several Bayesian network subsystems. The information propagation mechanism is used to accomplish the partial inference of the multiply subsystems and compact information propagation of the overlap subdomain among subsystems, and realize the inference and diagnosis of the distributed Bayesian network fault. The experimental results demonstrate that the fault diagnosis method can accomplish the single fault and multi?fault diagnosis inference and diagnosis task in the complex and uncertain system. In comparison with the traditional Bayesian network fault diagnosis method, the method has the prominent advantage of fast inference speed and extensive application prospects.
Keywords: Bayesian network; distributed inference; multi?fault diagnosis; complex system
0 引 言
随着高新技术在大型复杂系统中的广泛应用,一方面极大地改善了系统的性能,使系统功能越来越强大;另一方面,由于技术和结构的复杂化,不确定因素及不确定信息充斥其间,系统的故障概率也大幅提高,并且由单故障转向多故障并发。为保证系统正常运行,必须采取有效的故障诊断方法,对系统进行实时观测,准确、及时地检测与排除故障。
近几年来,多故障诊断问题得到了国内外学者的关注[1?4]。文献[5]提出基于分布式神经网络的多故障诊断方法,需要系统中有足够的可观测节点才能进行相应的故障识别。文献[6?7]提出有向图模型的模糊多故障诊断算法,只考慮系统的结构模型,未充分考虑测试过程中的不确定性因素,因此在工程应用中受限。……
