基于生成对抗网络和深度神经网络的武器系统效能评估  

WEAPON SYSTEM EFFECTIVENESS EVALUATION BASED ON GENERATIVE ADVERSARIAL NETWORK AND BP NEURAL NETWORK

作　　者：李健[1] 刘海滨[1] 胡笛 Li Jian;Liu Haibing;Hu Di(China Aerospace Academy of Systems Science and Engineering,Beijing 100048,China)

机构地区：[1]中国航天系统科学与工程研究院,北京100048

出　　处：《计算机应用与软件》2020年第2期253-258,共6页Computer Applications and Software

摘　　要：武器系统的效能评估受很多因素的影响,神经网络是现代武器系统效能评估的重要方法,但受样本量的限制,很难达到预期的训练效果。针对这一问题,选取少批量真实数据训练生成对抗网络,待网络达到纳什均衡后,利用生成网络产生同分布的伪数据。将伪数据与真实数据结合形成扩增样本,使用扩增样本训练深度神经网络用以评估。同时,生成对抗网络中的判别网络也能为专家评估提供一定的参考。Effectiveness evaluation of weapon system is affected by many factors.Neural network is an important method for evaluating the effectiveness evaluation of modern weapon system.However,due to the limited by the sample size,it is difficult to achieve the desired training effect.In order to solve this problem,a small batch of real data was selected to train the generative adversarial network.When the network reached Nash equilibrium,the generative network was used to generate pseudo data with the same distribution.Then,the pseudo data was combined with real data to form an amplified sample.The amplified sample was used to train the depth neural network for evaluation.In addition,the discriminative network in the generative adversarial network can also provide a certain reference for expert evaluation.

关 键 词：武器系统 效能评估 生成对抗网络 扩增样本 深度神经网络 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程]