Multi-agent confrontation action prediction method and device

Abstract

The embodiment of the invention provides a multi-agent confrontation action prediction method and device, and the method comprises the steps: determining a calculation graph of a multi-agent confrontation scene, wherein the calculation graph comprises global attribute features, and the node information and relation information of each agent; inputting the calculation graph into an adversarial action prediction model to obtain execution action information of each agent output by the adversarial action prediction model, wherein the adversarial action prediction model is obtained based on a sample calculation graph and sample execution action information training of each sample intelligent agent in the sample calculation graph, the adversarial action prediction model is used for fusing the global attribute characteristics and the node information and the relationship information of each intelligent agent to update the node information of each intelligent agent, and determining the execution action information of each intelligent agent based on the updated node information. According to the method and the device provided by the embodiment of the invention, the information of the multi-agent confrontation scene is effectively represented, and the effective processing of the variable-dimension multi-agent characteristics is realized.

Description

technical field [0001] The invention relates to the technical field of artificial intelligence, in particular to a multi-agent confrontation action prediction method and device. Background technique [0002] Multi-agent confrontation mainly studies how to drive the internal members of the agent formation to effectively complete the task of confrontation with the enemy. Basically, the formation is driven to engage enemy targets entering the area. It can be seen that the multi-agent confrontation problem is actually a typical multi-agent cooperation problem, and its core is the cooperative decision-making of the agents in the formation. Deep reinforcement learning is one of the means to achieve multi-agent collaborative decision-making. In specific applications, for different task instructions, the task area, task time, number of formation members and member attributes are dynamically changing. Therefore, the neural network model is required. The data input dimension can cha...

AI Technical Summary

Problems solved by technology

[0004] The embodiment of the present invention provides a multi-agent confrontation action prediction method and device to solve the technical problem that the existing multi-agent action prediction method cannot effectively process variable-dimensional agent features

Images