Structure Learning Of BN Using Improved Cloud Genetic Algorithm

Bayesian Network provides a graphical model to represent the probability distribution of variables, with explicit structure, flexible reasoning ability, easy decision-making mechanism and efficient learning mechanism. Now it has been an effective tool in uncertainty knowledge describing, data analysis and reasoning. Bayesian Network learning includes structure learning and parameters learning. The parameters can be easily got under given structure and data set. Structure learning is NP-hard problem, so it's of great significance for Bayesian Network to find effective way and algorithm on structure learning.In this paper, we have got deep research on BN structure learning. Meanwhile, we propose cloud model adaptive mechanism, immune theory and allied strategy based on traditional Genetic Algorithm, and use this improved algorithm in BN structure learning. The main works of this paper is as follows:Introduce basic theory of BN, and generalize and summarize main frame of Bayesian Network.Introduce basic theory of cloud model in details, and mainly get research on the Cloud-based Genetic Algorithm combined with the characteristics of randomness and stable tendency of cloud model. Besides, we studied the strategy of adaptive crossover rate and immune rate.Based on the characteristics of BN structure, we improved the basic operations of Cloud Adaptive Genetic Algorithm. Then, we introduced the strategy of multi-group allied parallel evolution to enhance the overall algorithm performance and speed up global optimization. Besides, the immune operator was used in this improving algorithm. We injected immune vaccines to individuals in order to prevent population from degradation.Finally, we finished two experiments using this algorithm. The experiments recover that the new improved algorithm can be effectively used on BN structure learning, and it has higher learning efficiency.