A self-evolution generation method for multi-agent action strategy

Abstract

The invention discloses a multi-agent action strategy self-evolution generation method, which has strong robustness and self-adaptive ability, and is suitable for the rapid generation of the action strategy of the agent under the highly dynamic changing situation of the battlefield. Initialize the public neural network, which contains the actor network and the critic network. After initializing the settings, calculate the reward value obtained after the current time node performs the action; update the time node t, that is, t increments by 1. Calculate the state reward value of the last time node t in the current time series; update the state reward value of time node t+1, update the gradient value of the actor network parameters in the current thread, and update the critic network parameters in the current thread. The gradient value of network parameters; Update two global parameters of the public neural network. After the training of the public neural network is completed, a strategy generation model is formed, and in the face of a new air combat environment change, the state features and actions of the new air combat environment are input into the strategy generation model, and an action strategy is output.

Description

technical field [0001] The invention relates to the technical field of agents, in particular to a method for self-evolution and generation of multi-agent action strategies. Background technique [0002] In the real natural environment, unmanned equipment can be regarded as an intelligent body with perception and action ability. Since the situation information is usually in the process of constant change, the surrounding environment, allocatable resources, and macro tasks of the intelligent body may appear in a short period of time. And the rapid change of information such as the agent's own capabilities. In the limited time for action planning, the ability to quickly generate action strategies and realize the ability of multi-agent unified deployment and coordinated operations is the current focus of research on action strategies for multi-agent complex systems. It has a very wide range of applications in the fields of intelligent scheduling, industrial intelligence and com...

AI Technical Summary

Problems solved by technology

[0003] The commonly used theories such as analytic hierarchy process, evidence fusion method and multi-attribute decision-making mainly rely on expert knowledge and experience database for decision-making, which lacks enough cases for judgment, and the agent does not have self-exploration ability. In the battlefield environment, the increase, decrease, and clustering of agents in the entire intelligent system are not considered; the strategy generation method based on neural networks relies on large-scale supervised learning, and cannot effectively give action strategies in the face of small sample combat cases

Images