Research On Target Threat Estimation Based On DBN With Limited Samples

Threat assessment is the basis of command decision-making and action control,which plays an important role in the research on combat auxiliary decision-making.Target threat assessment is an important part of situation threat assessment.The diversity of modern warfare means and modes of operations leads to the incompleteness and uncertainty of information on the battlefield,which affects the accuracy of target threat estimation,and then affects the command decision on the battlefield.Dynamic Bayesian Network(DBN),as the time extension of static Bayesian network,is very suitable for target threat estimation in the battlefield environment because of its ability to model the interdependencies between variables,its ability to evolve over time and its ability to deal with good uncertain information.Therefore,this paper studies the target threat estimation based on dynamic Bayesian network under limited samples.Due to limited sample data,it is not possible to accurately indicate the causal relationship between target threat estimation feature quantities,which affects the effectiveness of structural learning and reduces the accuracy of target threat estimation.To solve this problem,it is proposed a DBN hybrid structure learning method based on improved slime mold algorithm(ISMA).Firstly,it is deleted the connecting edges between feature quantities that have no direct causal relationship through conditional independence testing,which reduces the search space and improves the learning efficiency.The slime mold algorithm is improved by combining the crossover operator and mutation operator to make it more suitable for the search of DBN structure of target threat estimation,and ISMA is used to search the edges between variables to obtain the causal relationship between each feature quantity,so as to obtain the optimal network structure of DBN of target threat estimation.The experimental results show that the accuracy and timeliness of the dynamic Bayesian network hybrid structure learning method based on ISMA are better than other similar comparison algorithms,and the algorithm can learn the correct target threat estimation DBN structure.Because the limited sample data can not correctly describe the causal relationship between the characteristic quantities of target threat estimation,it affects the accuracy of parameter learning,and then reduces the accuracy of target threat estimation.To solve this problem,a fuzzy K-nearest neighbor algorithm based on the correlation of feature(CF-FKNN)is proposed for DBN parameter learning.First,the correlation between the feature quantities is calculated and introduced it as a reference factor into the fuzzy K Nearest Neighbor(KNN)algorithm to fill the data,which changes the incomplete data into complete data.On this basis,the construction and reasoning of DBN are completed to achieve target threat estimation.The experimental results show that the accuracy of CF-FKNN filling is better than other algorithms of the same type,and this algorithm can obtain DBN parameters that are more suitable for the actual situation.Finally,CF-FKNN is applied to the target threat estimation in the battlefield environment,and the target threat level estimation results are consistent with the assumptions.In order to solve the problem that the DBN needs to be expanded along the time slice when using DBN for target threat level reasoning,resulting in a large amount of computation and high computational complexity,the junction tree inference algorithm based on the Triangulation Optimization Algorithm on Gray Wolf Optimization(TOA-GWO)is proposed to complete the target threat level estimation.First,use TOA-GWO to complete the triangulation of the network,get the optimal node elimination order,thus get the junction tree of the minimum state space,and on this basis,carry on the message transmission,complete the reasoning.The experimental results show that TOA-GWO has better convergence speed and can obtain the junction tree with lower state space value than the similar algorithms.In addition,the junction tree reasoning algorithm based on TOA-GWO is applied to target threat estimation in battlefield environment,and the result of target threat level estimation is completely consistent with the hypothesis.