Research On Bayesian Networks’ Theory And Application Based On Fuzziness And Randomness

Bayesian Network is an effective method for uncertain knowledge representation andreasoning. However, the value of the attribute variables usually are ambiguous in practicalapplications, which leads to the appearance of the uncertainty reasoning mixed of randomnessand ambiguity. For example, predicting the possibility of “light rain” for tomorrow. Whether itwill rain tomorrow is a random event, but “the rain is a light rain” is a fuzzy event. How toapply Bayesian networks to solve the uncertainty knowledge inference problem mixed ofrandomness and fuzziness has become a hot research focus. A new approach of fuzzyBayesian network learning algorithm based on randomness and fuzziness is proposed in thispaper. Firstly, define the hybrid event and its fuzzy probability, and on this basis, define thejoint fuzzy probability, conditional fuzzy probability and conditional fuzzy probability table,which solve the problem of representation for variables which are possessed of randomnessand fuzziness simultaneously. Secondly, genetic algorithm is used to optimize the networkstructure and parameters, i.e., conditional fuzzy probability table. Finally, Knowledgeinference based on network structure and parameters is introduced.Nowadays it is difficult to diagnose the risk in advance and give a safety warning in thefield of special pressure equipment’s inspection. According to the characteristics of the data inthe field, taking the industrial boilers’ external inspection from the field for example, involveddata is obtained from Guangzhou special pressure equipment inspection and research institute.The knowledge inference model of industrial boilers’ security alerts and control is built usingthe proposed algorithm in this paper. Results shows that, compared to other classicalalgorithms, the proposed algorithm has a higher computational complexity and takes morerunning time because of importing membership degree into the process of building a network,but due to the use of fuzzy logic, it not only plays a better performance in reasoning correctrate, but also be able to make uncertain inference where the variables are mix of randomnessand fuzziness in practical applications.