The Improvement And Implementation Of Bucket Elimination Inference On Multiply-connected Bayesian Networks

The inference tasks are NP-hard in general Bayesian Network, and more hard in multiply-connected Bayesian Network. The present inference algorithms could not reach the practice need in computational precise and rate. This paper analyses the deficiencies in the research of Bayesian Networks and the status quo and tendency of present Bayesian-networks technical development, then researches the theories of inference of multiply-connected Bayesian Networks. Many things have been done on the research of inference methods based on multiply-connected Bayesian networks, and the idea of parallel compute based on the bucket-elimination algorithm also been put forward on this paper. This paper also builds an integrated experimental platform for the test and further research of Bayesian Networks, and the improved algorithm is applied on the reliability assessment of power system. Experimental results indicate that the improved BE is more effective than the traditional BE and that Bayesian network have rich applicaton foreground and values. Hu Weihua(Computer Applied Technology)Directed by Prof.Zhu Yongli...