多兵种交战中作战指数迭代计算方法及收敛性研究  

Iterative Calculation Method and Convergence of Operational Index in Multi-arms Engagement

作　　者：张昊 巫银花[1] 吴涛[1] 文韬 朱智[2] ZHANG Hao;WU Yin-hua;WU Tao;WEN Tao;ZHU Zhi(Warfare Research Laboratory,Naval Command College,Nanjing 210016,China;Train Management Department,Naval Command College,Nanjing 210016,China)

机构地区：[1]海军指挥学院作战实验室,南京210016 [2]海军指挥学院训练管理系,南京210016

出　　处：《火力与指挥控制》2021年第9期143-148,共6页Fire Control & Command Control

摘　　要：对于大规模多兵种交战,作战指数评估是进行军事决策的重要依据。针对传统作战指数评估方法依赖专家经验知识的问题,借鉴强化学习理论,提出了多兵种交战兰彻斯特方程中作战指数迭代计算方法,利用作战指数和火力分配策略的递推关系,循环更新作战指数直至结果收敛,重点考察了固定更新率、梯度指数递减方法和动量梯度方法对迭代收敛性的影响,通过调节更新率、递减强度和动量强度等超参数取值,显著提高了迭代收敛速度和稳定性。For large-scale multi-arms engagement,the evaluation of operational index is an important basis for military decision-making.In view of the shortcoming that traditional operational index methods rely on expert experience and knowledge,this paper proposes an iterative calculation method of operational index in Lanchester equation based on reinforcement learning theory.By the recursive relationship between operational index and fire assignment strategy,the operation index is updated circularly until the result converges.To improve the convergence of iteration,the fixed update rate method,gradient exponentially decay method,and gradient with momentum method are investigated.By adjusting the super parameter such as learning rate,exponentially decay strength and momentum strength,the convergence speed and stability of iteration are significantly improved.

关 键 词：作战指数 多兵种交战 火力分配 兰彻斯特方程 

分 类 号：TP301.6[自动化与计算机技术—计算机系统结构]