面向作战决策智能体的融合赋权评估方法OA北大核心CSTPCD

In the adversarial operational experimental environment such as Wargame,the agent trained based on reinforcement learning method can be tested the effect of operational decision-making,so as to support decision-making process.Aiming at the limitation that victory ratio is often used as a single indicator in the current operational decision-making agent evaluation,a comprehensive evaluation method based on fusing weight is proposed.Firstly,we build the evaluation system that covers both technical and tactical indicators,including operational decision-making effectiveness,operational intention implementation,adaptability to operational environment,agent training efficiency,agent training system and agent training methods.Secondly,subjective and objective weighting methods are used to weight different types of indicators and weights will be fused into the evaluation system according to the properties of indicators.Then the evaluation of agent effect based on fusing weighting is realized.The simulation platform of Army synthesis unit's offensive combat environment is designed,and the comprehensive ranking of agents can be obtained using this method in the platform to evaluate six agents,then the limitation of single indicator evaluation can be avoided effectively.This method provides a new idea for the evaluation of operational decision-making agents,meets the needs of operational environment,and has certain application value.