Research On Structure Learning Algorithm Of Bayesian Networks Based On Node Order

Bayesian network(BN)is a modeling tool combining probability and statistics and graph theory,which has been widely used in machine learning,data mining,and object recognition.In Bayesian network learning,structure learning is the basis of parameter learning and reasoning,so it is also the focus and difficulty of Bayesian network research.In structural learning,the commonly used search space includes network space and node order space.Compared with the search algorithm in network space,the search space based on node order is smaller,so its search efficiency is higher.However,the existing Bayesian network structure learning algorithms under the node order space still have problems such as insufficient accuracy.Therefore,this paper studies the Bayesian network structure learning algorithm based on node order space,aiming to improve the learning accuracy of existing algorithms,and applies the proposed algorithm to battlefield ground target type identification to verify its effectiveness and feasibility.The capital research contents and innovations of this paper are as follows:Aiming at the problem of low search efficiency of Bayesian network structure learning algorithms in network space,this paper studies the mapping relationship between Bayesian network structure space and node order space,and explores the advantages of learning Bayesian network structure in node order space.From a theoretical point of view,the division of Bayesian network structure space by node order is analyzed,and then the task of structure learning is transformed from Bayesian network structure space to node order space,which reduces the search space and improves the probability of obtaining a better scored structure.On this basis,a search operator in node order space is designed,which provides a basis for subsequent research on Bayesian network structure learning algorithm based on node order.In view of the existing Bayesian network structure learning algorithms in node order space usually have problems such as insufficient node order optimization,low learning accuracy,and easy to stop at a local optimum state.Therefore,this paper studies the local search algorithm in node order space,and proposes an IWINOBS algorithm to improve the learning accuracy of Bayesian network structure by optimizing the node order search operator.In this paper,the iterative local search algorithm is combined with the window operator to search the neighborhood of a given node order space,so as to reduce the probability of the algorithm dropping into a local optimal value and seek a higher quality network structure.In order to test the performance of the algorithm,a Bayesian network structure learning algorithm for optimizing node order search operators proposed in this paper is applied to the field of target type recognition in the battlefield environment.Taking the attribute information of the required target identification,such as the flight distance of the enemy artillery projectile,the flight speed,the scattering cross-sectional area of the radar,the ballistic curvature of the projectile,and the external rectangular area,aspect ratio,invariant moment 1 and invariant moment 4 of the platform as the evaluation indexes.On the basis of discretization of indexes,the proposed algorithm is used to construct a complete target type recognition model based on Bayesian network.This paper studies the Bayesian network structure learning algorithm based on node order,and the experimental simulation shows that compared with the K2 and HC algorithms in network space and the OBS and IINOBS algorithms in node order space,the IWINOBS algorithm in this paper can improve the accuracy on the basis of ensuring learning efficiency.It has certain theoretical research significance.