Research On Bayesian Network Inference Algorithm Based On Union Tree

Bayesian networks,also calls causal networks or probability networks.It is not only an effective means to study uncertain events in the field of artificial intelligence,but also an important method of graph model reasoning.The joint tree algorithm represents the joint probability distribution by transforming the bayesian network into the graph of the joint tree,thus completing the reasoning operation of the Bayesian network.At present,the joint tree inference algorithm has become a fast and widely used inference algorithm,which provides an important method for solving practical problems.Although the development of Bayesian network reasoning based on joint tree has achieved milestone progress,there are still many problems in the process of development.In this paper,the problems in the joint tree algorithm are studied deeply.Firstly,the basic knowledge of bayesian networks and their reasoning is introduced.On this basis,the joint tree triangulation problem is studied,and the minimum weight method is proposed.In the process of triangulation,the proper order of elimination is found and the chord graph with the minimum weight is obtained.The rationality and validity of the algorithm are verified by an example.In addition,the algorithm can also find out all the cliques involved in the chord graph,and lay a foundation for the establishment of the joint tree.Finally,the paper makes an in-depth study on the information transmission process of the joint tree.Under the condition of Lazy propagation,the LAZY-ARVE algorithm is improved,and a minimum filling edge algorithm is proposed to determine the order of variable elimination according to the score function,which is compared with LAZY-ARVE.The results show that the algorithm reduces unnecessary filling edges in information transmission.At the end of the paper,the problem of example is solved by using the algorithm of arc reversal.