Trajectory Correction of the Rocket for Aerodynamic Load Shedding Based on Deep Neural Network and the Chaotic Evolution Strategy With Covariance Matrix Adaptation

This article studies the problem of trajectory correction optimization for aerodynamic load shedding in the rocket's ascending phase considering the wind. Based on the statistical horizontal wind field, the weakest wind field is proposed to plan the nominal trajectory and the strongest wind field is proposed to test the performance of the designed trajectory. Considering the weakest wind field, two time-varying correction coefficients are calculated by one deep neural network unlike the traditional methods, and used to plan the rocket's flight attitudes when optimizing the rocket's trajectory. There are a large number of parameters to be optimized in this problem, so the traditional trajectory optimization techniques may suffer from poor convergence issues. By introducing the chaotic function into the traditional evolution strategy with covariance matrix adaptation, a novel variant C-CMA-ES is proposed to solve the trajectory optimization problem, and its efficiency is demonstrated by some popular test functions. Besides, the cost function to be minimized is shaped based on the rocket's maximal normal aerodynamic load and final state deviations. Finally, compared with the other three strategies, the efficiency of the proposed trajectory correction strategy is demonstrated by multiple simulation scenarios considering the strongest wind field and the Monte Carlo method considering the random wind fields.