Reinforcement learning training optimization method and device for multi-agent confrontation

Abstract

The embodiment of the invention provides a reinforcement learning training optimization method and device for multi-agent confrontation. The method comprises the following steps of a rule coupling algorithm training process including the steps of acquiring an initial first state result set of a red party multi-agent for each training step, if the initial first state result set of the red party multi-agent meets a preset action rule, obtaining a decision-making behavior result set according to the preset action rule, and otherwise, acquiring the decision-making behavior result set according toa preset reinforcement training learning algorithm; and performing reinforcement learning training on the red-party multi-agent by utilizing a training sample formed by the decision-making behavior result set and other preset parameters. The embodiment of the invention provides the reinforcement learning training optimization method and device for multi-agent confrontation. In the whole training process, the preset action rule can guide the multiple agents to act, invalid actions are avoided, the problems that in the training process in the prior art, invalid exploration is much, and the training speed is low are solved, and the training efficiency is remarkably improved.

Description

technical field [0001] The invention relates to the technical field of machine learning, in particular to a multi-agent confrontation-oriented reinforcement learning training optimization method and device. Background technique [0002] Artificial intelligence is a technical science that researches and develops theories, methods, technologies and applications for simulating and expanding human intelligence. One of the main goals of artificial intelligence research is to simulate human decision-making by intelligent agents (Agents), so as to be competent for some complex tasks that require human intelligence to complete. The limited functionality of a single agent to cope with complex tasks has driven the concept of multi-agent systems. A multi-agent system is composed of multiple agents that can make independent decisions and interact with each other. They share the same environment and have perception and execution mechanisms. At present, multi-agent systems have become a...

AI Technical Summary

Problems solved by technology

However, these methods have great limitations. The main problems include: when the number of agents increases, the training efficiency is low

As the number of agents increases, the size of the action-state space of the multi-agent system increases exponentially, and more and more time is required for trial-and-error exploration, resulting in low training efficiency.

Images