Distribution method oriented to simulation Q learning attack targets

Abstract

The invention discloses a distribution method oriented to simulation Q learning attack targets. The method comprises the following steps: (1) determining an initial state, acquiring two-side air combat situation information of a red side and a blue side, wherein the two-side air combat situation information comprises the number of formation aircrafts and relevant parameters of formation aircrafts, and providing input for target distribution and air combat model calculation of the red side; (2) determining an action set performed by the formation of the red side, strictly stipulating complete state-action pairs, determining a proper probability epsilon value and selecting actions of the red side by using an epsilon-greedy strategy; and (3) stipulating a Q learning algorithm reward function, an ending state and a convergence condition and distributing iteration to attack targets of the red side by using a Q learning algorithm until the convergence condition is met. The method gets rid of the dependence on the priori knowledge; due to the introduction of the epsilon-greedy strategy, the local optimum trap is avoided; by setting the parameter epsilon, the balance between the algorithm efficiency and the local optimum problem can be sought.

Description

technical field [0001] The invention relates to the technical field of combat simulation, in particular to a simulation-oriented Q-learning attack target allocation method. Background technique [0002] In air combat simulation, the traditional method of target assignment is to find a pairing scheme in the pre-given two columns (task and execution unit) elements to maximize the value (or consume the least) after pairing, but because it depends on the gradient function and thus It is easy to fall into the local optimal trap; the ant colony algorithm has the advantage of good optimization ability when used for target allocation, but the calculation time is long, especially when facing complex systems, and the efficiency is low; compared with the ant colony algorithm, the particle swarm algorithm When used for target allocation, it has the advantages of fast optimization speed and simple algorithm, but it is still easy to fall into local optimum when dealing with discrete probl...

AI Technical Summary

Problems solved by technology

[0002] In air combat simulation, the traditional method of target assignment is to find a pairing scheme in the pre-given two columns (task and execution unit) elements to maximize the value (or consume the least) after pairing, but because it depends on the gradient function and thus It is easy to fall into the local optimal trap; the ant colony algorithm has the advantage of good optimization ability when used for target allocation, but the calculation time is long, especially when facing complex systems, and the efficiency is low; compared with the ant colony algorithm, the particle swarm algorithm When used for target allocation, it has the advantages of fast optimization speed and simple algorithm, but it is still easy to fall into local optimum when dealing with discrete problems; when genetic algorithm solves such problems, although it makes good use of the algorithm's self-organization, Adaptability, parallelism, uncertainty and other characteristics, but failed to overcome the defect of poor local search ability, resulting in slow algorithm convergence, greatly affecting the search efficiency, and prone to premature phenomena

Images