计算机工程与应用 ›› 2006, Vol. 42 ›› Issue (13): 26-.

• 博士论坛 • 上一篇    

GIBBS仿真方法运用在大型因果图的推理过程

汪成亮,陈娟娟   

  1. 重庆大学计算机学院
  • 收稿日期:2006-01-12 修回日期:1900-01-01 出版日期:2006-05-01 发布日期:2006-05-01
  • 通讯作者: 汪成亮 wclight wclight

The Application of GIBBS Sampling in the reasoning of huge Causality Diagram

,   

  1. 重庆大学计算机学院
  • Received:2006-01-12 Revised:1900-01-01 Online:2006-05-01 Published:2006-05-01

关键词: ~{Rr9{M<, ~}, ~{PE6HMx~}, Gibbs~{7BUf~}, ~{9JUOUo6O~}

Abstract: The Causality Diagram methodology, which was 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[1][2][3], and it was 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.

Key words: causality diagram, belief network, gibbs simulation, fault diagnosis