摘要
考虑实际无人系统指挥控制环境中各类随机扰动对武器目标分配问题建模与求解的影响,研究了3类不确定性扰动约束,建立了一个多目标动态传感器武器目标分配模型;针对扰动导致模型性质变化、传统单算子求解算法鲁棒性不足的问题,提出了一种基于DQN的多算子约束多目标进化框架。该算法在目标和决策空间中描述种群的收敛性、多样性和可行性,并建立从状态到再生算子的映射模型,实现动态调整的再生策略。仿真实验验证了算法的优异性能和求解模型时的鲁棒性。
The impact of various random disturbances in the actual command and control environment of unmanned systems on problem modeling and solving of weapon target assignment was considered,and three types of uncertainty disturbance constraints were investigated.A multi-objective dynamic sensor weapontarget assignment model was established.By considering the issues of model property changes caused by disturbances and insufficient robustness of the traditional single-operator solving algorithm,a multi-operator constrained multi-objective evolutionary framework based on the deep Q-network was proposed.The algorithm described the convergence,diversity,and feasibility of the population in both the objective and decision spaces.It established a mapping model from states to reproduction operators,achieving a dynamically adjusted reproduction strategy.Simulation experiments validated the algorithm's excellent performance and robustness when solving models.
作者
白臻祖
侯一帜
何章鸣
魏居辉
周海银
王炯琦
Bai Zhenzu;Hou Yizhi;He Zhangming;Wei Juhui;Zhou Haiyin;Wang Jiongqi(College of Science,National University of Defense Technology,Changsha 410073,China)
出处
《系统仿真学报》
北大核心
2025年第12期2967-2980,共14页
Journal of System Simulation
关键词
动态武器目标分配
鲁棒性
深度强化学习
多算子再生
元启发式算法
dynamic weapon target assignment
robustness
deep reinforcement learning
multi-operator reproduction
meta-heuristic algorithm
