Intelligent control method for vertical recovery of carrier rockets based on deep reinforcement learning

Abstract

An intelligent control method for vertical recovery of carrier rockets based on deep reinforcement learning is disclosed. A method of autonomous intelligent control for carrier rockets is studied. Theinvention mainly studies how to realize attitude control and path planning for vertical recovery of carrier rockets by using intelligent control. For the aerospace industry, the autonomous intellectualization of spacecrafts is undoubtedly of great significance whether in the saving of labor cost or in the reduction of human errors. A carrier rocket vertical recovery simulation model is established, and a corresponding Markov decision-making process, including a state space, an action space, a state transition equation and a return function, is established. The mapping relationship between environment and agent behavior is fitted by using a neural network, and the neural network is trained so that a carrier rocket can be recovered autonomously and controllably by using the trained neural network. The project not only can provide technical support for the spacecraft orbit intelligent planning technology, but also can provide a simulation and verification platform for attack-defense confrontation between spacecrafts based on deep reinforcement learning.

Description

technical field [0001] The invention relates to a vertical recovery control method of a carrier rocket based on deep reinforcement learning, which is applicable to the field of carrier rocket guidance and control. Background technique [0002] The standard reinforcement learning framework is that an agent continuously interacts with its environment in discrete time. It mainly consists of four elements: reward and punishment feedback function, value function, strategy selection, and interactive environment. Such as figure 2 shown. [0003] The process of the agent interacting with the environment is as follows: (1) The agent (Agent) perceives the current state of the environment (state); (2) According to the current state and the reward value (reward), the agent selects an action (action) and executes it. The action; (3) When the action selected by the agent acts on the environment, the environment is transferred to a new state and a new reward is given; (4) The agent calcu...

AI Technical Summary

Problems solved by technology

Under such a major premise, the trajectory planning and control of the launch vehicle is very dependent on manpower, and any human error may cause huge losses

Images