Fuzzy Reasoning Of High Dimensional Causal Network For The Management Of The Colleges' Assets

As an effective tool to deal with the causal relationship problem between the variables, Bayesian network has been widely used in the field of artificial intelligence and data mining.But the traditional Bayesian network also has three problems as follow, which restrict the further development.At frist,traditional Bayesian network can only deal with discrete variables and processing other types of variables need to be discrete. So that the "edge sharpening" problem is easy to arise. Especially in the face of the fuzzy variables, discretization will cause a lot of information loss, which affect the accuracy of reasoning.Then the structure learning of Bayesian network is a NP-HARD problem, when deal with the high dimensional data. Under these condition, the search space of its network structure will show exponential growth, traditional structure learning algorithm will become slow.In the end,the traditional structural learning algorithm can not identify Markoff equivalence class. When there is a large number of Markoff equivalence classes in the search space, neither of the efficiency and the accuracy is high to this algorithm.In view of the frist problem, this paper extended the traditional Bias network system based on the fuzzy theory and give the hybrid Bias network complete system which is compatible with the fuzzy variables. For the second question, this paper proposed a new scheme of combination and reduction. This scheme can reduce the influence of high dimensional data by decomposed the network construction problem into multiple subnets. At the same time, the method also determines the connection point between the sub networks at the same time, so that avoids the second search and reduces the computational complexity compared with the same kind of algorithm. The clustering process uses the similarity based on causality, which can reduce the negative impact on the quality of the subnet structure. Finally, in view of the problem 3, it combined the information-geometry causal inference with the mountain climbing method, and put forward the improved mountain climbing algorithm IGCI-HC. This new algorithm can solve the problem of identify the Markov equivalence class,which the traditional structure learning algorithm can not solve.The construction and development of universities are directly influenced by the asset management of colleges and universities.There are so many factors,affecting the assets management of universities.This problem has feature of high dimension and relates to many types of factors.How to excavate the factors and asset management efficiency between causal analysis and knowledge reasoning decision-making is key points of the application.This paper establishes a hybrid causal network combined with the reduction algorithm based on the asset management related data of 72 universities. The experimental results show that the hybrid Bayesian network has some computational complexity due to the processing of fuzzy variables. However, the computational complexity is reduced by combining the reduction algorithm and the IGCI-HC algorithm. In addition,the problem of edge sharpening is solved by using the fuzzy probability to represent uncertainty of variable.Compared with the traditional Bayesian network, hybrid Bayesian networks is more outstanding in the quality of network and the accuracy of reasoning.