退火递归神经网络极值搜索算法及其在无人机紧密编队飞行控制中的应用

An annealing recurrent neural network for extremum seeking algorithm and its application to unmanned aerial vehicle tight formation flight

针对无人机紧密编队飞行问题,以气动干扰引起的僚机俯仰角θw作为极值搜索变量,利用退火递归神经网络极值搜索算法,使僚机干扰俯仰角θw收敛至其极值,从而解决了无人机紧密编队飞行中僚机所需动力最小化的问题.将退火递归神经网络与极值搜索算法相结合,消除了传统极值搜索算法中控制量的来回切换问题和输出"颤动"现象,改善了系统的动态性能,同时简化了系统的稳定性分析.通过对无人机紧密飞行编队的仿真,验证了该算法的有效性.

In the unmanned aerial vehicle （UAV） tight formation flight, a novel annealing recurrent neural network combined with the extremum seeking algorithm （ESA） is used to search the extremum of the wingman pitch angle produced by the vortices, for minimizing the required power of the wingman. This combination eliminates the back-and-forth switching between control variables in the conventional ESA and suppresses the ＂chattering＂ in the output, greatly improving the dynamic performance of the system and simplifying the stability analysis of the controlled system. This algorithm is validated by a simulation of UAV tight formation.