中文摘要：

采用深度学习方法和自编码机的神经网络建模和优化方法,求解空天飞行器的控制分配问题。将空天飞行器控制分配问题的期望控制量看作自编码网络的输入,将舵面实际产生的控制力矩看作自编码网络的输出,通过构建一种特殊形式的深度神经网络,建立自编码机和控制分配问题的等价模型,在不需要用优化算法计算训练样本的前提下,实现了非线性控制分配。提出了一种全新的智能控制分配方法,与早期的基于神经网络的控制分配方法有本质不同。新方法能够很好地处理气动数据的非线性特性,具有较强的工程实用性。

英文摘要：

The deep learning method to solve the nonlinear control allocation problem of space reentry vehicle is introduced. The similarity between the stacked auto-encoder network in deep learning and the control allocation problem is built by linking the expected control moments to the network＇s input,and the actual moments produced by the airplane to the network ＇s output. The stacked auto-encoder is trained by using the output to reconstruct the input. This neural network is trained in an unsupervised manner,which is the crucial advantage over the traditional neural networks where a sufficient amount of training data has to be generated beforehand. The nature of neural networks allows them to deal with nonlinearity at a high accuracy. Therefore,this intelligent allocation method proposed in this paper could be a brand new direction in control allocation with a solid engineering validity.