基于神经网络响应面的鱼雷总体性能参数仿真优化方法  

Optimization Method of Overall Performance Parameters of Torpedo via Simulation Based on Neural Network Response Surface

作　　者：范若楠[1] 曾进[2] 任庆生[3] 李斌[1] 

机构地区：[1]中国船舶重工集团公司第705研究所,陕西西安710075 [2]上海交通大学数学系,上海200240 [3]上海交通大学计算机科学与工程系,上海200240

出　　处：《鱼雷技术》2015年第5期326-331,共6页Torpedo Technology

摘　　要：利用水声寻的仿真进行鱼雷总体性能参数优化匹配时,需要获取大量的仿真样本数据,为节省优化时间,节约优化成本,提出了一种基于神经网络响应面(NN-RSM)的鱼雷总体性能参数仿真优化方法。利用神经网络构建能够快速计算的响应面,并通过空间映射方法获得水声仿真系统的最优解,最后利用该方法对自导作用距离、毁伤半径等鱼雷总体性能指标进行了仿真优化。优化结果显示,该方法不仅能够获得原问题有效的优化结果,且优化计算时间小于普通优化方法的1/20。该方法可以进一步拓展应用到其他复杂系统多参数、多目标的仿真优化中。Underwater acoustic homing simulation can exhibit operational effectiveness of a torpedo in a real environment, however a large number of sample data is needed in the simulation for the overall performance parameters optimization of the torpedo, thus optimization time becomes the bottleneck. Based on the quick neural network response surface method（NN-RSM）, this paper uses the space mapping method to reduce the cost of optimization calculation and directly get the optimal solution to the underwater acoustic homing system. Subsequently, the overall performance indices of a torpedo, such as homing range and damage radius, are optimized. Optimization results show that effective optimization of the original problem is achieved with computation time less than 1/20 that of the general optimization methods. The present method can be applied to optimization of other complicated system with multiple parameters and multiple objectives.

关 键 词：鱼雷 总体性能优化 神经网络 响应面 空间映射 

分 类 号：TJ630.1[兵器科学与技术—武器系统与运用工程]