Aircraft maneuvering control method based on hierarchical reinforcement learning

Abstract

The invention relates to the technical field of flight control, in particular to an aircraft maneuvering control method based on hierarchical reinforcement learning. The method comprises the following steps: S1, acquiring an action embedding vector of an intelligent agent calculated by a neural network; S2, respectively outputting a probability list of a horizontal angle, a probability list of a vertical angle and a shooting probability list according to the action embedding vector; and S3, carrying out sampling according to the probability lists, determining the horizontal control mode and the vertical control mode of the intelligent agent, determining whether shooting is carried out or not, and controlling the intelligent agent. Through the mutual combination of the horizontal tactical maneuvering intention and the vertical three-dimensional maneuvering intention, a large number of maneuvering styles with more diversity and flexibility can be generated.

Description

Technical field [0001] The present application relates to the field of flight control, and more particularly to aircraft mobile control method based on hierarchical reinforcement. Background technique [0002] The super-visual air war is the main form of modern air combat. Its main interpretation is to detect tracking and launch missiles, and to avoid the enemy. This process involves a large number of mobile decision-making processes, how to make a mobile placement to occupy a good offensive situation and how to make motor circumvention to maximize the protection of the enemy's missile, these are superior air warfare Key issues considered. In recent years, how to make uncontrolled intelligent body make a problem of tactical decision-making behavior of human pilot, have become a hot spot in unmanned air combat research. The existing AI air combat method mainly includes rule-based expert system methods, probability model / fuzzy logic, and computational intelligent mixing methods, ...

AI Technical Summary

Problems solved by technology

The probabilistic model / fuzzy logic and computational intelligence hybrid method requires experts to build a probabilistic reasoning network or design a heuristic objective function, which cannot cover all air combat states and is very complex to design Difficult

Machine learning methods rely heavily on a large amount of real air combat data, which are usually rare or even unavailable, and tend to limit the behavior of the agent to the capabilities that the data can provide

The deep reinforcement learning method automatically generates tactical strategies for air combat through self-game reinforcement learning training without human knowledge supervision, but its maneuvering style is fixed, which greatly lacks diversity and flexibility compared with humans

Images