Trajectory online optimization for unmanned combat aerial vehicle using combined strategy

This paper presents a combined strategy to solve the trajectory online optimization problem for unmanned combat aerial vehicle (UCAV). Firstly, as trajectory directly optimizing is quite time costing, an online trajectory functional representation method is proposed. Considering the practical requirement of online trajectory, the 4-order polynomial function is used to represent the trajectory, and which can be determined by two independent parameters with the trajectory terminal conditions; thus, the trajectory online optimization problem is converted into the optimization of the two parameters, which largely lowers the complexity of the optimization problem. Furthermore, the scopes of the two parameters have been assessed into small ranges using the golden section ratio method. Secondly, a multi-population rotation strategy differential evolution approach (MPRDE) is designed to optimize the two parameters; in which, 'current-to-best/1/bin', 'current-to-rand/1/bin' and 'rand/2/bin' strategies with fixed parameter settings are designed, these strategies are rotationally used by three subpopulations. Thirdly, the rolling optimization method is applied to model the online trajectory optimization process. Finally, simulation results demonstrate the efficiency and real-time calculation capability of the designed combined strategy for UCAV trajectory online optimizing under dynamic and complicated environments.

基于卷积神经网络的结冰翼型气动特性建模

提出了基于卷积神经网络(CNN)的结冰翼型气动特性预测方法,设计了输入层结冰翼型图像规范,克服了复杂冰形在翼面同一位置法线方向存在多值,单值函数难以描述的问题。预测模型可同时预测多个迎角对应的升阻力系数,实现了直接从冰形图像到气动特性的快速预测,对升力系数和阻力系数预测结果的平均相对误差均可控制在8%以内。重点研究了不同卷积层数量、卷积核数量、卷积核尺寸对模型性能的影响规律:CNN的不同层次特征对应不同滤波频率,卷积层数增加会捕获更多高频特征量;增加卷积核数量可提取更多冰形特征,提升模型性能,但数量过多会增加冗余特征,降低模型泛化性能;阻力系数预测模型对卷积核数量的最低要求大于升力系数,其原因在于,相较升力系数,阻力系数不仅受翼面压差影响,还受摩阻特性影响,其建模所需的关键特征数量多于升力系数;增大卷积核尺寸,可扩大卷积操作“视野”,增强对冰形整体特征信息的提取,有利于提升模型泛化性能。相关结论为飞机结冰气动特性实时动态预测与监测提供了新的思路和方法支撑。