圆轨道任务火箭推力故障下的在线任务重构研究

To maximize the utilization of remaining fuel for a rocket on a circular orbit mission to enter a degraded orbit in the event of a thrust drop failure preventing mission completion，a mission replanning algorithm that integrates deep neural networks（DNN）with lossless convex optimization is proposed for two novel types of degraded orbits. Initially，a two-layer optimization algorithm is formulated to diminish the dimensionality of the original non-convex problem through equivalent transformation，lossless relaxation of constraints，and the incorporation of extra coasting segments. This approach converts the problem into an inner layer that is amenable to lossless convexification and an outer layer that can be optimized using a single variable to guarantee convergence. Subsequently，DNN is employed to substitute for the outer layer algorithm in generating a single variable，while preserving the inner layer's lossless convex optimization algorithm to eliminate reliance on DNN output accuracy，thereby achieving an optimal enhancement in computational efficiency with minimal risk. Ultimately，numerical simulations demonstrate that the hybrid algorithm not only ensures accuracy and optimality but also exhibits rapid convergence and stable performance，making it highly valuable for engineering applications.