运载火箭推力故障下在线计算轻量化任务重构方法

为了避免运载火箭推力下降故障引起发射任务失败，基于径向基神经网络，提出了一种在线计算轻量化的任务重构方法，可快速在线计算最优救援轨道对应飞行轨迹(最优轨迹)的近似解。在离线部分，结合凸优化与自适应配点法产生“故障状态-最优轨迹”数据集。数据集被用来训练径向基神经网络，建立轨迹决策模型来构建故障状态到最优轨迹的动力学关系。在线应用时，不需要迭代求解轨迹优化问题，只需将离线训练好的神经网络正向传播应用即可。仿真校验了本文方法在圆轨道、椭圆轨道两种救援类型情况下的有效性，与直接法相比，本文方法可获得最优轨迹的近似解，并且在线部分的计算时间降低三个数量级。

Mission Reconstruction Method with Lightweight Online Computation for Launch Vehicles under Thrust Drop Fault

To avoid the launch failure of launch vehicles caused by thrust drop faults, an online mission reconstruction method based on radial basis function neural network (RBFNN) is proposed, which can quickly obtain the approximate solution of flight trajectory from the fault position to the optimal rescue orbit (optimal trajectory, OT) online. In the offline part, mission reconstruction problems under numerous fault states of the thrust drop are solved by the convex optimization and the adaptive collocation method to generate the dataset about the fault states versus the OT. The dataset is used to train the RBFNN to establish a trajectory decision making model for mapping the relationship from the fault states to the OT. During the online application, instead of iteratively solving the trajectory optimization problem, the RBFNN trained offline is used for forward propagation, the approximate solution of the OT can quickly be obtained by the trajectory decision making model. The effectiveness of the proposed method in the case of circular orbit and elliptical orbit is validated by the numerical simulation. The results show that the online solving times of the proposed method are decreased by more than three orders of magnitude, compared with the direct method.