Aiming at the problem that the existing methods of target damage assessment based on Bayesian network mainly determine the structural parameters of Bayesian network by giving conditional probability tables based on expert experience, which results in too subjective and having large errors in the evaluation results, an improved learning algorithm of conditional probability tables of Bayesian network is proposed in this paper. We divided the E step of the EM algorithm (Expectation Maximization Algorithm) into three steps. Firstly, the range of the missing variable is determined by expert experience, then the Gibbs sampling algorithm is used to complete the sample set, and finally the sample is weighted. The proposed algorithm is compared with EM algorithm, Gibbs algorithm, EM and Gibbs algorithm. The experimental results show that the proposed algorithm has good stability and high precision.
Paper
Full text
Bayesian Network Parameter Learning Algorithm for Target Damage Assessment
Semantic Scholar · Engineering · 2019
Abstract
Aiming at the problem that the existing methods of target damage assessment based on Bayesian network mainly determine the structural parameters of Bayesian network by giving conditional probability tables based on expert experience, which results in too subjective and having large errors in the evaluation results, an improved learning algorithm of conditional probability tables of Bayesian network is proposed in this paper. We divided the E step of the EM algorithm (Expectation Maximization Algorithm) into three steps. Firstly, the range of the missing variable is determined by expert experience, then the Gibbs sampling algorithm is used to complete the sample set, and finally the sample is weighted. The proposed algorithm is compared with EM algorithm, Gibbs algorithm, EM and Gibbs algorithm. The experimental results show that the proposed algorithm has good stability and high precision.