Research On Bayesian Network Structure Learning Based On Hybrid Rice Algorithm

Learning Bayesian networks is an NP-hard problem.When the data set is relatively large,it will be very difficult or even impossible to rely on experts to build a Bayesian network.Therefore,the use of effective algorithms for structural learning on data sets is the main direction of current research.Bayesian network learning is mainly divided into structural learning and parameter learning.Parameter learning often requires the network structure to be determined in advance.Therefore,structural learning is the core of Bayesian network research.The classical algorithms in Bayesian network structure learning include K2 algorithm,hill-climbing algorithm,particle swarm algorithm and MWST algorithm,but these algorithms have obvious deficiencies in some applications.The hybrid rice algorithm is a newly proposed bionic optimization algorithm in recent years.The hybrid rice algorithm has good convergence,robustness,and is not easy to fall into the local optimal solution.Therefore,the hybrid rice algorithm is used in Bayesian network structure learning.(1)Introduces the origin and development of Bayesian networks,research status,and current applications of Bayesian networks.The basic theoretical knowledge of Bayesian network is introduced in detail.Based on this,several scoring methods and search algorithms for Bayesian network structure learning are introduced.(2)Introduced the basic principle of the hybrid rice algorithm,and then based on the Bayesian network storage method proposed a hybrid binary rice algorithm based on discrete binary.The improved algorithm firstly combines the multi-point crossover operator of the genetic algorithm and the crossover operator of the hybrid rice algorithm on the continuous dataset to propose a discrete binary crossover operator;then the subtraction operator of the particle swarm algorithm is applied to restore rice and The self-crossing process of optimal rice;Finally,the structural characteristics of more Bayesian networks improve the resetting strategy of resetting operators.(3)In this thesis,the maximal weighted spanning tree algorithm MWST algorithm is used to generate the maximum weighted tree,and the initial population of the hybrid rice algorithm is generated by adding edges,subtracting edges and deleting edges on the basis of this tree.Then the hybrid rice algorithm was used to learn the initial population,and the results were compared with the K2+MWST algorithm and the greedy algorithm to prove the effectiveness of the algorithm.Compared with the genetic algorithm and the particle swarm algorithm,the high sensitivity of the algorithm was proved.Awesome.(4)This paper summarizes the research of this paper and points out the problems encountered in this study.It points out the learning algorithm of missing data sets,and how to obtain higher-ranking Bayesian networks in illegal structure maps.