GIBBS仿真方法运用在大型因果图的推理过程

The Application of GIBBS Sampling in the Reasoning of Huge Causality Diagram

在信度网基础上发展起来的因果图模型,克服了信度网的一些不足,目前已发展成一个能够处理离散变量和连续变量的混合模型,特别适于运用在大型的工业故障诊断领域。但因果图在推理计算中,面临着与信度网的推理算法复杂度过高的同样问题。通过比较几种MarkovChainMonteCarlo(MCMC)方法,论文基于Gibbs仿真算法的思想,并对进入稳态条件、采样顺序判定准则、采样结束判据等进行深入分析,进而提出了一种改进的仿真推理新方法。利用该采样仿真算法能极大地提高故障诊断的速度和准确度,这对因果图模型在工业在线故障诊断领域中的应用具有重要意义。

The causality diagram methodology,which is based on belief network,overcomes some shortages in knowledge expressing and reasoning of belief network and has evolved into a mixed causality diagram methodology coping with discrete and continuous variables,and it is very useful for industrial fault diagnosis application.However,it is still confronted with a problem as belief network is,of high computation complexity.By comparing several Markov Chain Monte Carlo（MCMC） simulating algorithms,and analyzing the requirement for stable-condition,the principle of sampling sequence and the criterion of sampling ending,this paper puts forward an improved simulating reasoning algorithm based on Gibbs simulation.The simulating algorithm will improve the diagnosis speed and accuracy,which has an important significance for the application in industrial online fault diagnosis.