基于深度网络的蓝军装备模型参数生成方法研究

蓝军装备的建模仿真是构建对抗仿真环境不可缺少的部分。针对蓝方系统可获取的参数有限、具有“贫信息”“小样本”特性的问题，提出一种基于深度网络的蓝军装备模型参数生成方法。通过设定信息注入蓝军装备的仿真模型，生成仿真数据，利用数据训练深度神经网络。得到的网络对该型装备的未知参数预测具有一定的泛化能力，可直接用于预测或作为迁移学习的源模型。以蓝军某型拦截弹的建模仿真为例对该方法进行应用验证，使用了多层感知机和循环神经网络2种网络对比例导引系数进行学习与预测，均获得了较好的效果。

Research on Parameter Construction Method of Blue Army Equipment Model Based on a Deep Network

The modeling of blue army equipment is an indispensable part of adversarial simulation environment construction. Aiming at the limited available parameters of "information-poor" and "small sample" characteristics to the blue system, a deep network-based method is proposed to generate the parameters of blue army equipment model. By injecting the information into the simulation model of the blue army equipment, the simulation data is generated and trained in the deep neural network. The obtained network has a certain generalization ability to the unknown parameters prediction of the same type of equipment and can be used directly in prediction or be the source model for migration learning. The application is verified by the modeling simulation of a certain type of interceptor of blue army in which two kinds of networks, multilayer perceptron and recurrent neural network, are used to learn and predict the proportional guidance coefficients, and the results are good.