基于深度置信网络效能拟合的火控系统精度全局敏感性分析

In view of the fact that most of the current researches on the accuracy of fire control system require high data integrity and only analyze the influence factors of single error source, this paper proposes a novel global sensitivity analysis (GSA) method based on deep belief network (DBN) effectiveness fitting for the accuracy of fire control system. Starting from the advantages and disadvantages of the traditional GSA methods, the limitations of the traditional methods in the case of incomplete data are analyzed. Then, a fire control system performance fitting model is constructed and trained by utilizing the excellent feature extraction ability of DBN and a combination of unsupervised training and supervised fine-tuning methods. The proposed GSADBN method is compared with traditional Sobol method, classical Fourier amplitude sensitivity test (FAST) method and improved BNSobol method. The experimental results show that the proposed GSADBN method can not only meet the accuracy requirements, but also obtain the excellent accuracy analysis results when the data from aviation fire control system is incomplete. Furthermore, GSADBN method can give a reasonable scheme of accuracy value when designing each module of fire control system, so as to provide reference and theoretical support for the design of fire control system and the improvement of combat effectiveness.