Object-oriented Bayesian Networks And Itsapplication In Risk Assessment

With the background of the complex system of multi-target risk assessment, the construction and reasoning of Bayesian network and the risk evaluation of complex system with multi-target were studied in this paper. The main contents are as follows:Firstly, the construction method of object oriented Bayesian networks parameters based on cloud model is proposed. Risk assessment as a means of expression of subjective knowledge is applied to the risk assessment by experts. But the construction workload of Bayesian networks explosive growth with the increase of the number of parent nodes. The traditional Noisy-OR construction method is only limited to the node state for two values. So this paper proposes a method of constructing Bias network parameter based on the cloud model, which extends the state of the Noisy-OR model from two states limit to multiple states. The reasoning ability of Bayesian network combined with the expression ability, based on the State combination weight, which make use of a small amount of expert knowledge construct cloud model, and then use the uncertainty of the State combination weight generate the parameters of Bayesian networks, so as to achieve the purpose of saving the experts. Experimental results show that the statistical results of the method to generate the parameters of Bayesian networks matches the expert given initial state weights match, and the technique can effectively express expert knowledge and improve the efficiency of modeling.Secondly, an index value elimination algorithm is proposed. According to dense tree type structure Bayesian network inference calculation of high complexity, difficult to simplify the problem, put forward index value elimination inference algorithm, which analyzes the discretization feature of hard evidence and the continuity characteristics of soft evidence probability and divided the ordinary CPT table into the index table and calculating table. The index table is used in in calculating the hard evidence and the calculating table is used in the calculation of the soft evidence, so that the computational complexity is only in depends on the number of soft evidence. Experiments are made to prove that if the algorithm in the ratio is higher than 40% of the hard evidence of conditions, it has the common advantages of value elimination method and the junction tree algorithm, and efficiency is greatly improved.Then, the risk assessment method of multi-target risk assessment based on Bayesian networks object fusion is put forward. According to the changes in the number of complex system threat sources and difficult to modify or maintain models, combines object-oriented Bayesian networks and information entropy theory, and the method of multi-target comprehensive risk evaluation based on the object cumulative fusion of put forward. Bayesian networks classes for each type of threat source were designed, and according to the threat source population the threat source objects are dynamically instantiated and are inferred. And then the results by removing the uncertainties of the result of single threat assessment based on the entropy theory are accumulatively fused, so as to get the final results of the assessment of the multi-target complex risk assessment system.Finally, the comprehensive risk assessment system software is designed and implementation. According to the object-oriented Bayesian network construction, reasoning and multi-target risk assessment methods, the use case diagrams are designed, the system is decomposed into different model, and the interface is given through the analysis of the needs of the risk assessment. Finally, the reliability and validity of the system are verified by experiments.