On-line Trajectory Planning for Launch Vehicles Based on Intelligent Decision-making and Aftereffect Compensation Under Fault Conditions

To address challenges associated with numerous decision-making branches，suboptimal decision-making， and significant aftereffect impacts on decision-making capabilities during launch vehicle power system failures，an online trajectory planning technique for launch vehicles based on intelligent decision-making and aftereffect compensation is developed. Initially，a random forest learning model is employed to train a residual task capability evaluation agent， enabling rapid decision-making by accurately predicting the launcher state at the shutdown point and assessing the reachability of the original target orbit. When the original target is unreachable，a feasible degraded mission target orbit is provided. Subsequently，a deep neural network is utilized to establish a mapping between the initial state parameters and the velocity increment during the aftereffect stage，allowing for the determination of the flight state increment induced by the thrust aftereffect and significantly reducing the aftereffect estimation time while maintaining accuracy. Based on the decision outcomes，the position and velocity variations caused by the thrust aftereffect are incorporated into the trajectory optimization model. A terminal virtual target constraint is constructed to compensate for the thrust aftereffect，and the model is further discretized，convexified，and solved using a convex optimization algorithm to achieve high-precision orbit insertion. Finally， simulation results demonstrate that the proposed method effectively enables payload rescue under fault conditions.