Adaptive Flight Control Using Second Order Sliding Mode Online Learning
Adaptive Flight Control Using Second Order Sliding Mode Online Learning
复制标题
使用二阶滑模在线学习的自适应飞行控制
DOI：
10.2514/6.2013-5133
复制
发表时间：
2013
影响因子：
6.8
通讯作者：
P. Voersmann
中科院分区：
文献类型：
--
作者：
P. Schnetter;M. Marcinek;T. Krueger;P. Voersmann
针对小型无人机系统，提出了一种基于二阶滑模在线学习的非线性动态逆控制器的神经网络扩展方法。由于其体积和重量较小，这类飞机对大气湍流等非线性非常敏感，对系统退化特别敏感。像非线性逆动力学(NID)这样的非线性控制方法可以在已知模型参数的条件下抵消这些问题。由于这类飞机的系统辨识在整个飞行包线中往往不是有效的，因此使用神经网络来补偿这些模型误差。本文提出了一种基于高阶滑模控制(SMC)的自适应飞行控制策略，用于训练神经网络。这种学习策略源于变结构理论，并将神经网络的训练视为控制问题。它在保持SMC理论稳健特性的同时，实现了学习速率的动态和稳定计算，并提供了更快的收敛速度。在这项工作中，将一阶和二阶滑模学习与存在外部干扰和系统退化的标准梯度下降训练进行了比较。
A neural network expansion of a nonlinear dynamic inversion controller using second order sliding mode online learning is presented for a small unmanned aircraft system (UAS). Due to their small size and weight this class of aircraft is very susceptible towards nonlinearities like atmospheric turbulence and reacts especially sensitive to system degradation. Nonlinear control approaches like nonlinear inverse dynamics (NID) allow for counteracting these problems under the condition of well known model parameters. Because system identification for this class of aircraft often is not valid throughout the complete flight envelope, neural networks are used to compensate these model errors. In this work an adaptive flight control strategy is augmented with the concept of higher order sliding mode control (SMC) for the training of neural networks. This learning strategy is derived from variable structure theory and considers the training of a neural network a control problem. It enables the dynamic and stable calculation of the learning rate while maintaining the robust characteristics of SMC theory and offers a higher speed of convergence. In this work first and second order sliding mode learning are compared to standard gradient decent training in the presence of external disturbances and system degradation.