Bayesian Network And Its Application Research

Bayesian network is an inference tool combined with graph and probability.The inference result of Bayesian network which with strong visibility is trustworthy?Because of the firm foundation,Bayesian network has become an effective t means to solve the uncertainty problem,and got the wildly application.In this thesis,the theory of Bayesian network has been researched.We proposed three new methods based on three special situations by fusing neural network,Markova chain and Fuzzy set theory with Bayesian network.The forecasting and warning method has been proposed in this thesis.It inherited the predictive function of neural network and the diagnosis function of Bayesian network.In the case of oilfield development risk forecasting and warning.Firstly,we calculated the probabilistic from risk degree of each index use 3? method under its original dimension after the prognosis of neural network.Then,input the result which 1s named evidence into Bayesian network to infer the posterior probability distribution.This method improved the confidence level of evidence and the accuracy of diagnose.The result 1s fit for the fact.The method combined with Markova chain and neural network for forecast has been proposed.It forecast the risks from two aspects of vertical and horizontal,avoided the excessive fitting of the data when only use Bayesian network to make the forecasting,and enhanced the accuracy of the forecast.In the case of drilling risk forecasting,we compared the two forecasting result of the index at the bottom from Markova method and Bayesian network method.It shows that the forecasting degree of accuracy is 81,82%,when use Markova method to deal with the problem.And the accuracy of Bayesian network is 45.45%.It proves that the Markova method is better than Bayesian network when the forecasting object is the bottom index.When we use the combined new method to make the forecast of the upper index,we find that the accuracy for this case is 100%.In this paper,the fuzzy Bayesian network which is the combination of fuzzy theory and Bayesian network theory has been proposed.Theoretically demonstrated the relationship between the fuzzy degree and the probability distribution of index.Define the fuzzy degree of the fuzzy set which is consisted of the index states 1s equals to the subjective probability of the states.The fuzzy Bayesian network method 1s look at the combination of human's subjective experience and the data information.Because on the one hand,it employs the fuzzy degree function to confirm the initial parameters of Network,and on the other hand,it uses the data to make the parameters to be a computable number.In the case of oilfield development risk forecasting and warning,the inference result reflected the relationship between the observation index and the production at the high water cut stage of the oilfield.At last,we concluded the research content and the direction of further work are put forward.