Path planning method based on multi-agent enhanced learning

Abstract

The invention discloses a path planning method based on multi-agent enhanced learning, and belongs to the technical field of aircrafts. Firstly, a global state division model of an air flight environment is established, and a global state transfer table Q-Table 1 is initialized, and the global state of a certain line is randomly selected to be used as the initial state s1; in all columns of the current sate s1, an epsilon-greedy algorithm is adopted to select a certain column to be recorded as a behavior a1; based on the selected behavior a1, the next state, a formula which is as shown in thespecification, of the current state s1 is obtained in the global state transfer table Q-Table 1; the specific element values corresponding to the current state s1 and the behavior a1 are updated in the global state transfer table Q-Table 1 according to a transfer rule of a Q-Learning algorithm; a formula which is as shown in the specification is updated to enter an inner layer cycle; and a local planning path corresponding to the updated state s1 is obtained by adopting the Q-Learning algorithm. The number of iterations of the outer-layer cycle is increased by 1 until N1, and global path planning of the aircraft in the air is completed. The aircraft can meet requirements of different environments, so that the survival rate and the task completion rate of the aircraft are improved, and theconvergence speed of the enhanced learning is improved.

Description

technical field [0001] The invention belongs to the technical field of aircraft, and relates to a path planning method based on multi-agent reinforcement learning. Background technique [0002] With the continuous development and improvement of air traffic management technology, accurate and fast path planning for aircraft is an important means to ensure the safe flight of aircraft in complex airspace environment and effectively improve the efficiency of air traffic. Path planning is often based on a certain evaluation standard system, in a given planning space, to find the optimal or most feasible trajectory of the moving object from the starting point to the target point and satisfying the constraints and certain performance indicators, so that the moving object can complete safely. Scheduled missions and the path selection of aircraft in the air have always been the focus of research. [0003] The more commonly used flight path selection method is to plan the path before...

AI Technical Summary

Problems solved by technology

For reinforcement learning, the number of states will directly affect the learning speed; with the expansion of the state and the subdivision of the state, it will lead to the "dimension disaster" of reinforcement learning

Images