基于DNN与IMPDE的武器系统效能优化  

Optimization of Weapon System Effectiveness Based on DNN and IMPDE

作　　者：王国梁 赵滟[1] 杨卓鹏[1] 杨超[1] WANG Guo-liang;ZHAO Yan;YANG Zhuo-peng;YANG Chao(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：《计算机仿真》2022年第6期71-75,85,共6页Computer Simulation

摘　　要：针对武器系统结构复杂,多变量多峰值的特点,难以稳定、准确地获得最佳参数组合,达到系统效能最大化的问题,建立以效能影响因素为输入、效能值为输出的深度神经网络模型,并提出一种改进多种群差分进化算法(IMPDE)。算法引入多种群协同进化策略和参数自适应调整策略,采用各具特点的差分进化算法对不同的子种群执行进化操作,增强了种群个体的多样性,保证了算法快速收敛的同时能有效地跳出局部极值点,提高了算法全局寻优的能力。仿真对比结果表明,上述算法具有更好的寻优精度和稳定性,可以高效准确地寻找到参数最佳组合。Due to the complex structure of the weapon system, and its multi-variable and multi-peak characteristics, it is difficult to stably and accurately obtain the optimal parameter combination and maximize the system effectiveness. So in view of these problems, a deep neural network model was built with the influence factors of effectiveness as input and its value as output, and an improved multi-group differential evolution algorithm was proposed(IMPDE). The algorithm introduces multi-group co-evolution strategy and parameter self-adjusting strategy, and adopts differential evolution algorithms with different characteristics to perform evolution operations on different subpopulations, which enhances the diversity of individual populations, ensures that the algorithm can quickly converge while effectively jumping out of local extreme points, and improves the algorithm’s ability to globally optimize. The results of simulation comparison show that the algorithm has better optimization accuracy and stability, and can efficiently and accurately find the optimal parameter combination.

关 键 词：效能优化 深度神经网络 差分进化算法 多种群协同进化策略 参数自适应调整策略 

分 类 号：TP301.6[自动化与计算机技术—计算机系统结构] TJ765.43[自动化与计算机技术—计算机科学与技术]